Changelog

Updates and improvements to Pathmode. We ship regularly.

October 18, 2026

Changed

•An implementation review passes only when the reviewer marks every outcome the intent states as met. The bar used to be a majority of outcomes. On the labelled examples we check the grader against, the two bars gave the same verdict every time: incomplete changes met half or fewer of their outcomes, and correct changes met all of them. The stricter rule closes one way a partial change could pass. It does not make a green check proof that the behaviour works: on those same examples the reviewer still credited outcomes whose deciding code was outside the diff. Read a pass as “nothing shown contradicts the intent”, and review what the diff does not show yourself.

Fixed

•Your agent now sees all of Pathmode’s instructions. Claude Code cuts MCP server instructions at 2,048 characters, and the connected instructions were about four times that, so the rules it read ended before preflight-before-code for an existing intent.md, the adopt flow, and what to do when the server rejects a key. Both the connected and the keyless instructions now fit under that limit with those rules intact (@pathmode/mcp-server 1.43.2, plugin 0.1.53).
•When a product choice is still open, your agent’s readiness check leads with the not-ready verdict and the choice to resolve, even when all six field checks pass. The field checks used to appear first, and a waived check could read as ready while a decision was outstanding (@pathmode/mcp-server 1.43.2).
•After your agent confirms an intent field, the reply tells it to run the readiness check again before further edits or a summary. The confirmation used to read as the end of the step, so an agent could report ready from the confirmation alone (@pathmode/mcp-server 1.43.2).
•The daily agent-activity email is held when the workspace’s queue address cannot be looked up, and is retried once it can. It used to be sent with a link to the wrong page.

October 17, 2026

Features

Review pull requests from your own GitHub workflow

The Pathmode GitHub Action now reads the code changes in your pull request and sends them to Pathmode for review against the linked intent. It uses your workflow’s GitHub permissions, so you can get a review even when the Pathmode GitHub App is not installed or cannot read the pull request. The result appears as a Pathmode intent review check, with code references for its findings and the exact proposal version and commit it reviewed. Human resolutions carry forward when they still apply to the next revision. The review is advisory: it gives your team evidence to consider before merging.

Removed

•New workspaces no longer start with a "Connect your coding agent" intent, and existing ones have been removed. It sat in your intent list beside your own work and described a setup flow that has since changed. Connect a repository from Home or from the start screen after sign-up instead.

Improvements

•We now check implementation reviews against a fixed set of examples labelled before the grader sees them. The examples cover incomplete changes, misleading citations, and rules that do not apply to a change. We compare proposed grader changes with the current behavior, track false passes and false blocks separately, and reserve unseen examples for a final check. Recorded responses can be replayed to test the review rules without another model call.

October 14, 2026

Features

Try a review from both sides

The reviewer guide walks through answering questions, checking what the coding assistant understood, and approving a specific version or asking for changes. Try the interactive examples yourself: nothing is saved or built, and approval does not release anything to customers.Running the agent? The coding agent guide shows the reviewer’s page beside the terminal at each step, with setup instructions and explanations of readiness blockers, listening limits, and when the agent must stop.

October 13, 2026

Features

Your agent works from the version you approved

In a connected repository, approving now reaches your agent’s copy of the intent. If your agent is listening for your review, it brings intent.md in the directory it works in to the version you approved, including a fresh worktree, so nobody copies the file over by hand. If the file cannot be brought level, the listener says why and does not tell your agent to build. Your agent’s readiness check says whether its intent.md matches the approved version, and Pathmode hands it the implementation prompt only when it does. An out-of-date copy is brought current with npx @pathmode/mcp-server@latest pull. Local edits are never overwritten silently: your agent saves them for review, or discards them on purpose and names the fields it replaced. Reconnect your MCP client after updating (@pathmode/mcp-server 1.43.0, plugin 0.1.52).

See what Pathmode handed your agent

After you approve, the review page and the proposal show what Pathmode recorded: when it issued the implementation prompt for this version, the name of the key that asked for it, and the latest note or finding your agent recorded since. Until something is issued, the page says no delivery is recorded instead of promising a build. When an agent asks to build a version that is not approved, or one that changed after it checked, Pathmode refuses and the page shows the refusal with its reason, for example that the approval is for another version. The record shows what Pathmode issued, not whether your agent built it.

Fixed

•Your agent’s readiness check now applies the app’s approval rule exactly. A repository intent needs its exact version approved, whoever wrote it, and an unfinished intake review blocks building.
•When your agent saves a new version while the review page is open, the page moves to that version so you can review and approve it. It used to stay on the earlier version behind an error that Try again could not clear.
•A workspace invitation now names the address it was sent to before you choose how to sign in. If you sign in with a different address, Sign Out and Use Correct Email takes you to account entry with the invited address filled in, instead of a password sign-in that could not create an account. To join with the address you used instead, the screen asks you to have the invitation sent to that address.

Improvements

•While a version waits for your approval, your agent starts listening for it and ends its turn instead of checking back on a timer. The commands your agent is told to run match the server version it uses, so a pinned plugin setup is not switched to the latest release.
•A connected save leaves intent.md matching what the workspace stored, and says when the workspace kept something differently. For example, a verification sent without a checks list keeps its stored checks.

October 12, 2026

Features

Review a proposal without knowing how the software works

Contribute product knowledge through a review page in your browser. For example: a cancellation proposal ends access immediately. Maria points out that the customer has already paid and asks to keep access through the paid period. The agent revises the proposal, and Maria reviews and approves the changed behavior separately. The review page opens with what happens today, what will change, and what stays the same, with numbered references to the agent’s original wording. Questions appear one at a time, with the full list visible so you know the size of the review. The page marks where you need to act next and tells you when nothing more is needed. After a revision, the sentence your answer changed is marked so you can confirm it or ask for another change.

Illustrative example (33 seconds): four product decisions hiding in a cancellation request, reviewed before the agent builds. Watch on YouTube

Ask for changes in your own words

On a proposal your agent wrote, Ask for changes takes a sentence instead of a form. Your words are kept exactly as written, stay open through every update to the proposal, and keep it from being approved until you close the request. Your agent reads them, revises the proposal, and explains what it changed. Close the request when the reply addresses it, or say what is still not right and the same request stays open with the whole exchange. Closing a request never approves anything (@pathmode/mcp-server 1.42.0, plugin 0.1.51).

October 11, 2026

Fixed

•Connecting a repository intent to your workspace dropped any code block from its Implementation Context. Code examples in that section now reach the hosted intent intact; they still do not count as outcomes, checks or other requirements. Agents proposing a correction to an outcome are now told to send the complete outcome, so their first proposal is accepted instead of failing on a bare sentence (@pathmode/mcp-server 1.41.3, plugin 0.1.50).

October 10, 2026

Features

Say whether evidence supports or challenges a claim

When you link evidence to a claim, choose Supports, Challenges, or Context. A challenge you confirm marks the claim as contested, so evidence against it stays visible alongside evidence for it. Your agent can propose a relation, but a person must confirm it. The review page, implementation prompt, and exports distinguish confirmed support from challenges, context, and unconfirmed suggestions.

Read the original passage beside the finding

Newly extracted evidence keeps its original source passage when one is available. Open the evidence to compare that passage with the editable finding; editing the finding leaves the passage intact. The original wording also appears while you review extracted evidence. Source passages stay in Pathmode and are not included in agent prompts or exports.

Fixed

•Security updates to Next.js and to the libraries bundled into the MCP server, including the MCP SDK (1.31.0). None of the advisories applied to how Pathmode uses them, so no action is needed on your side; updating removes them from your security scanners (@pathmode/mcp-server 1.41.2, plugin 0.1.49).

October 9, 2026

Features

Your agent builds when you authorize, if you asked it to

When your request asked your agent to implement the change, authorizing the proposal is your last step: your agent builds it without asking again. The review page tells you before you decide. The button reads Authorize and build, and the confirmation names the revision your agent will build. If you asked only for a proposal, your agent still asks after you authorize. Tell your agent to hold off and it keeps listening for your answers, but asks before building; if that changes while you are deciding, the page shows it and asks you to decide again. Your agent builds only a revision that passes the execution checks. When one does not, the page does not offer the build, and your agent tells you what blocks it (@pathmode/mcp-server 1.41.0, plugin 0.1.47).

Your agent listens for your review answers

After it hands a proposal over for review, your agent can keep listening instead of stopping: it runs npx @pathmode/mcp-server@latest watch-review in the background. When you answer a product choice in Pathmode, it picks the answer up and applies it, and the review page shows the revised proposal without a reload. The page says whether your agent is listening or has picked your answer up. Listening lasts up to four hours per review, and your agent starts again only when you ask. When you authorize, it tells you and asks before implementing. In a client that cannot run background commands nothing changes: go back to your agent and ask it to apply your answers (@pathmode/mcp-server 1.40.0, plugin 0.1.46).

Withdraw an answer, answer again, or ask your agent to retry

An answer to a product choice used to be stuck once you sent it, even when your agent could not apply it. Withdraw answer brings it back into the form so you can change it. When intent.md moved after you answered, Answer again does the same. For any other refusal, Ask the agent to retry keeps your answer exactly as it is and asks your agent to try again.

Fixed

•In a connected workspace, your agent builds only from a spec that passes the execution checks. The implementation prompt still offered a way to proceed after you accepted a failing check, and that path led to a prompt telling your agent not to implement. Accepting a named risk still works in keyless mode (@pathmode/mcp-server 1.41.1, plugin 0.1.48).
•When a repository has no intent.md, the grill, verify, split and handoff skills ask which saved intent you mean instead of picking the workspace’s most recent one, so their edits and notes land on the intent you chose. The workflow interview no longer starts by connecting the repository (plugin 0.1.48).
•The compile-intent conversation checks its draft against the six preflight gates, records behavior you left undecided as open product choices, and saves a new spec as a local draft until you adopt it. A failing check is shown but never blocks the save. Its checklist used to accept a deferred verification and out-of-scope lines in place of a constraint, both of which preflight rejects (@pathmode/mcp-server 1.41.1).
•Adopting a second repository into the same workspace against a local or staging server no longer re-points the other repositories on that machine. The one credential file per workspace used to be overwritten, so the first repository’s agent traffic went to the new server with the new key, and nothing warned. Each repository now holds the credential for the API origin it was adopted at, installs from earlier versions keep working, and a repository binding that names an origin with no stored credential connects nowhere rather than borrowing another one (@pathmode/mcp-server 1.41.0).
•Connecting a second repository on a free plan reuses the key your first repository holds. Adopt presented only a configured API key, never the key an earlier adoption stored, so the approval offered to replace the workspace’s only key, and collecting it deleted the key the first repository used. Adopt now presents keys stored on this machine for the same server, never for another one, and a browser-started adoption code tries each instead of stopping at the key limit (@pathmode/mcp-server 1.41.0).
•A message too long for an intent’s discussion is no longer saved as the assistant’s reply. Pasting a long document used to add "Invalid intent interview request" to the conversation, and a short follow-up could be refused the same way. The message box now keeps your text and says how much to cut: one message holds up to 20,000 characters. Nothing is added to the conversation, and a message refused for any other reason comes back to the box too.
•When import redaction is enabled, a research file whose redaction fails is skipped with a retry message. It used to be saved with its original content. Privacy settings now explain which imports redaction covers and which content it leaves unchanged.
•Workspace privacy and integration controls are read-only for people who are not owners or admins. Connection details remain visible, and the settings explain who can change them.
•You can save the signing secret from your Linear webhook in workspace settings, including on an existing connection without entering its API key again. The connection used to generate an unrelated secret, leaving Linear’s signed updates unable to verify.
•MCP setup instructions only fill in a newly created key when it has full access. A key limited to evidence ingestion no longer appears in a setup command that needs broader access.

Improvements

•While an answer is with your agent, the review page checks for the revised proposal and loads it without a reload. If you have unsaved changes, it waits until they are saved, and it tries again if loading fails.
•Your agent gets a clearer account of the Pathmode tools. The execution prompt says when it refuses and what it carries, every product-choice field is described, and recording a finding points to proposing the exact correction. Pressure wording in the prompts and skills is replaced with plain statements (@pathmode/mcp-server 1.41.1, plugin 0.1.48).

October 8, 2026

Fixed

•An answer you sent to your agent no longer reads "Answer requested", which looked like a request for the same answer again. The row says "Answer sent" and whether your agent has read it, a proposal that depends on an answered choice says it waits for your agent, and the header tells you to go back to your agent and ask it to apply your answers.
•Your agent can apply your answer to a product choice when the recommendation adds a requirement the intent already has. The apply used to be refused, as if intent.md had moved after you answered. The existing requirement is kept as it is, and a refusal for any other reason no longer says intent.md moved.

Improvements

•Your own decisions stand out in the product choices: the "Your decision" label is marked as complete, and the decision itself reads darker than the page’s own labels.

October 7, 2026

Fixed

•Workspace home no longer asks your role every time you return to it. An unanswered setup question used to come back on each visit; one person saw the role question 13 times. It now waits a day before asking again, and answering or skipping it still settles it for good.
•If you said after signup whether you will run the coding agent yourself, workspace home no longer asks again after your role. The second answer used to overwrite the first.

October 2, 2026

Features

Delete an intent from its page

The intent page has an Intent actions menu with Delete intent, for the workspace’s owners, admins and editors. The confirmation says what goes: the intent, its review comments and its guest link, for everyone in the workspace. Linked evidence stays, and a repository’s intent.md is not changed. Until now an intent could not be deleted in the app.

Changed

•Your agent can record an answer to an open question that has no recommendation, such as one from the workspace assisted intake. Cost and reopen trigger are optional for those answers. adopt refuses an intent that still holds assisted questions or answers, and says to keep it local or use the workspace intake (@pathmode/mcp-server 1.39.0).

Fixed

•After you approve a connection, the waiting page says when your agent has stopped checking for it, and tells you to go back and ask it to continue before the approval expires. It used to promise that the page would move on by itself.
•Rerunning adopt after its approval request expired starts a new request straight away. It no longer prints the expired code first, or reports an expired request as denied in the browser (@pathmode/mcp-server 1.39.0).
•intent.md saves preserve confirmations in files with Windows line endings, and repository roots supplied as Windows file URLs resolve to the correct directory (@pathmode/mcp-server 1.39.0).
•The email telling you your agent added an intent links to that intent. It used to open the workspace and leave you to find it; the review queue now has a link of its own.

October 1, 2026

Features

Start an intent with a brief, then edit the draft

Create intent saves your brief and organizes its exact passages into the problem, intended behavior, outcomes, and boundaries. Edit the wording directly on the page, or open the original brief to mark passages yourself. Automatic placement remains a draft: implementation requires the readiness checks and your review of the current content. If the behavior is undecided, leave the question open and gather input.

Choose between the ways your brief says it is done today

When a brief says the same situation is handled in more than one way, the draft quotes each way under the behavior question, with nothing preselected. Pick one to start your answer from its exact wording, then say when it applies. Something else, Leave undecided and These aren’t alternatives are still there. The way you don’t pick is not recorded as ruled out. Guests on the review link see the quotes and can propose an answer from one; you still decide.

Ask a colleague to propose an answer

Leave a decision undecided and Share for input becomes the next action. The guest review link leads with the question and the problem it refers to. Guests can propose an answer with an optional reason, or comment without an account. Workspace editors can use the proposal, edit it before recording, decline it, or reply. The original proposal stays visible alongside the recorded answer. A proposal cannot be accepted after its question or problem changes, and guest feedback alone never resolves the decision or authorizes implementation.

Prepare for implementation on the draft page

The header points to the next missing step, with readiness prompts beside the wording they concern. Add failure cases and verification checks inline, see progress before opening preparation, and review the complete draft once the checks and decisions are resolved. Content changes invalidate that review. The final action opens the implementation prompt and intent.md export, with the agent connection status shown; exporting does not send or start work.

Fixed

•Evidence counts and searches continue beyond 1,000 records instead of silently stopping at the first page.
•Queued intent autosaves run in order, preventing two saves from competing against the same revision. Verification checks retain unique IDs when older and newer check formats coexist.
•Undoing a tracker import preserves evidence linked to an intent, including a link added during the undo. Repeated URLs in one import no longer create duplicate evidence items.
•Failed Linear, Jira, and Slack webhook processing remains retryable instead of being treated as already completed. Retried shipment updates do not duplicate the shipped note.
•Account deletion reports cleanup failures instead of claiming success. Content belonging to a surviving workspace is preserved.
•Unreadable product-choice records keep readiness blocked instead of disappearing from the decision count.

Improvements

•The original brief shows which sentences have no current source selection. Clear automatic roles removes untouched automatic additions while preserving manual edits and separately marked requirements.
•Brief titles use a complete sentence or a short generated title instead of cutting the opening sentence in half. A generated title never replaces one you have edited.

September 28, 2026

Changed

•Preflight grades intents written in Finnish. The objective, outcomes, title and constraint checks read Finnish wording as strictly as its English translation, so a Finnish spec no longer fails on its language alone. Other languages are still read but left unconfirmed (@pathmode/mcp-server 1.38.0, @pathmode/cli 2.7.0, plugin 0.1.45).
•Preflight counts a refusal as a measurable outcome when it says which case is refused or what the refusal names: "An import with a duplicate ID is refused and names the existing row" now passes. A bare refusal such as "Bad requests are rejected" does not. Blocked and prevented follow the same rule, so "Bad requests are blocked" no longer passes either (@pathmode/mcp-server 1.38.0, @pathmode/cli 2.7.0, plugin 0.1.45).

September 27, 2026

Features

Your agent starts the connection, you approve it

In a repository with an intent.md, your agent can now run npx @pathmode/mcp-server@latest adopt without a code. It prints a link and a short code. Open the link, check that the browser shows the same code, choose the workspace and Product, and approve. The agent collects the approval and uploads intent.md, so nobody copies an API key. Nothing is uploaded before you approve. If the workspace is at its key limit, the approval names the key it would replace, and Cancel changes nothing. The command generated in the browser still works (@pathmode/mcp-server 1.36.0).

Changed

•After you save an intent.md without a workspace, your agent points team review to adopt instead of the website and a manual API key. setup, the compile-intent prompt, and the Claude Code plugin say the same, and the plugin’s optional key can stay empty. An existing pm_live_ key still works for workspace access without a local intent (@pathmode/mcp-server 1.36.1, plugin 0.1.43).
•Preflight reads a `## Why` section as the objective when there is no Objective or Problem section. A `## What` section is not read as outcomes, because it usually describes the solution. When outcomes are missing, the verdict says the What section was not evaluated and that success criteria belong under Outcomes (@pathmode/mcp-server 1.37.0, @pathmode/cli 2.6.0, plugin 0.1.44).

Fixed

•Workspace editors can finish connecting a repository. The last step used to refuse them after the browser had generated their command.
•If a connection response is lost, or a temporary server error interrupts adoption, rerunning adopt resumes the same connection. It keeps the saved credential and avoids creating another key or uploading the same spec twice (@pathmode/mcp-server 1.36.0).
•setup stores your API key in a file only you can read, and checking the key no longer counts as using it. It refuses anything that is not a pm_live_ key before writing any config, including the settings placeholder, which used to report success (@pathmode/mcp-server 1.36.0).
•Saving intent.md no longer deletes a Problem section written beside Objective, or a Proposed outcome beside Outcomes. Preflight reads only Objective or Outcomes when both are there, and the save used to treat the other section as spec and drop it without saying so. It is now kept, and the save message lists it. If the save leaves that field empty, the kept section would be read in its place, so the save removes it and names it (@pathmode/mcp-server 1.37.0, plugin 0.1.44).
•Saving intent.md also keeps a prose Proposed outcome section when the file lists its outcomes in frontmatter. A list under that heading replaces the frontmatter outcomes; prose does not, so the prose is now kept as your text instead of being deleted (@pathmode/mcp-server 1.37.0, plugin 0.1.44).
•The web preflight now treats an empty Objective, Problem or Outcomes heading the way the CLI does: the field stays empty instead of being filled from a Background or Summary section, so the same file gets the same verdict in both.
•Preflight no longer calls an objective too vague when it found none. An intent.md with no objective section now gets "No objective found", and the vague message is kept for text it actually read (@pathmode/mcp-server 1.37.0, @pathmode/cli 2.6.0, plugin 0.1.44).

September 26, 2026

Features

Every check passed. Wrong duck.

A new animated explainer follows the same request through two paths: build a duck, then discover it needs to be a pirate to complete a set; or settle that choice in the proposal, review the agent's revision, and authorize it before building. Both paths end with the right duck. One needs rework after the build.

An illustrated example: the same question, answered before the build or after.

Answer product choices without a workspace

With no workspace connected, your agent can record your answer to each open product choice in intent.md, in your words: use the recommendation, choose the alternative, exclude it, or defer it. The answer adds what the choice promises to the spec, and reopening it takes back exactly what it added. Preflight then reads Ready, with 4 product choices answered locally (unverified) and quotes each answer. It refuses one answer that covers every recommendation, and a choice answered before the choice it depends on. A file in the repository can't show who decided, so verified sign-off stays with the workspace review (@pathmode/mcp-server 1.35.0).

A warning before an answer can fail the outcomes check

Preflight now grades the outcomes an open choice would add, before anyone answers it. If accepting a choice would add an outcome preflight can't read as measurable, the verdict says so, because accepted as written it can fail a spec that passes today. Answer it with a measurable outcome instead. The warning never blocks a save (@pathmode/mcp-server 1.35.1, @pathmode/cli 2.5.1).

Changed

•The Ask an AI section no longer calls its answer independent. The prompt is ours, and the answer draws largely on what we wrote. It stays on the comparison pages and is gone from pricing.

Fixed

•Changing a choice from open to accepted in intent.md, with nothing else, no longer passes preflight. Without a workspace, an accepted choice counts only when the spec includes what it promised. This keeps the file consistent. It does not verify who decided (@pathmode/mcp-server 1.35.0, @pathmode/cli 2.5.0).
•The preflight.pathmode.io paste box reads product choices, scope, and decisions. It used to call a spec with four open choices ready.
•Decisions whose ruled-out text contains a closing parenthesis are no longer dropped when a local save rewrites intent.md (@pathmode/mcp-server 1.35.0).
•Two accepted choices that promise the same outcome, edge case, or constraint add it to the spec once. It used to appear twice (@pathmode/mcp-server 1.35.2).
•Without a workspace, the verdict no longer sends you to a proposal review you don’t have. It says a person answers or excludes each open choice before implementing. After a save, the note on connecting a workspace names what it adds: private evidence, a reviewer who answers and authorizes, and staleness across repositories (@pathmode/mcp-server 1.35.2, @pathmode/cli 2.5.2).
•Answering a choice with its own recommendation, to reword an outcome the warning flagged, keeps the choice’s alternative. It used to replace the alternative with the answer, so the decision read as ruling out the option it chose. Same fix in the workspace review (@pathmode/mcp-server 1.35.3).
•Intent.md writes from your agent run one at a time, so two answers sent together are both recorded. The second used to overwrite the first while both reported success (@pathmode/mcp-server 1.35.4).
•Without a workspace, your agent reads intent.md from the repository your editor has open. It used to read the folder the MCP server was started in, so after switching checkouts it kept serving the old spec until the server was restarted (@pathmode/mcp-server 1.35.5).

Improvements

•The pricing page starts with a choice between the keyless repository workflow and the Free workspace, and the first price shows without scrolling.
•intentspec.org/intent-md checks an intent.md written in the format of Anthropic’s AI-native SDLC playbook: the verdict it gets as written, and the same file after three repairs. pathmode.io/intent-md redirects there.
•intentspec.org documents the intent/<feature>/intent.md layout beside a root intent.md, and which one preflight reads.
•The CLI and MCP READMEs open with a recorded first run you can reproduce with npx -y @pathmode/cli preflight.

September 25, 2026

Features

Green CI does not validate a test the same change edited

When a change edits the test named by a verification check, a green CI run no longer marks that check passing. A previous pass returns to unknown; a failure stays failing. Changes to shared test setup also hold CI-backed checks for review. The pull request review explains which tests changed and asks a person to confirm that they still check the intended behavior. Naming a test file in a check narrows which test edits hold it.

Check the build against the decisions you made

When a pull request is reviewed against its intent, each recorded decision is now graded alongside the outcomes, hard boundaries, and failure cases, and held as strictly as a hard boundary. Building the option a decision ruled out fails the review, even if every outcome is met. A decision's reopen trigger is context for a person to reconsider it, not permission for the agent to reverse it. Decisions appear as their own group in the verification panel and in the GitHub check.

Preflight finds intent.md in an intent/ folder

Anthropic's AI-native SDLC playbook keeps intent.md in an intent/ folder, with spec.md beside it. When a repository has no intent.md at its root, preflight now reads intent/intent.md or intent/<feature>/intent.md. Only a file named intent.md counts, so spec.md is never graded as the intent. With several features, it asks which one to check instead of picking. A save writes back to the same file (@pathmode/mcp-server 1.34.0, @pathmode/cli 2.4.0).

Mark a choice that doesn’t need a product decision

An open choice can be closed without answering it again. Already answered in the brief quotes the brief that settles it, and the quote becomes the requirement. Implementation agent should decide hands it to the coding agent, within the existing outcomes, scope, and hard boundaries. Neither accepts the agent's recommendation or authorizes implementation.

Changed

•Green CI requires a complete changed-file list compared against the PR base commit IDs captured in its event. Runs without those IDs, including runs before a PR exists, leave checks held. A failed run still records failing checks.
•The site navigation shows Log in and Sign up. Installation stays on the homepage and the Developers page.
•Headings inside a field, such as a ## heading in implementation context, are written one level below the field’s own heading, so they stay part of it (@pathmode/mcp-server 1.34.0).

Removed

•The full intent editor, including Open advanced editor. Intents open in the proposal view, and conversation continues in its discussion drawer.

Fixed

•Changed claims no longer display evidence backing from an earlier revision while new records load or after a failed fetch.
•An unrelated workspace folder no longer blocks a save to the repository being reviewed (@pathmode/mcp-server 1.34.0).
•Undo in a brief no longer drops its decisions.
•Onboarding recognizes when you are already signed in.
•Invitation and membership changes complete as one step, so simultaneous requests can’t leave a workspace half-updated.
•Saving a hand-written intent.md keeps the text under its title and every section Pathmode doesn’t read, such as Open questions. A save used to delete them (@pathmode/mcp-server 1.34.0).
•A save writes only to the repository preflight checked, and never outside it, including through a symbolic link (@pathmode/mcp-server 1.34.0).

Improvements

•Returning to a revised proposal starts with "What changed since your last decision", above the remaining choices.
•Repository proposals wait briefly for review records before appearing, and same-content refreshes keep existing records on screen.
•Cursor setup starts with a keyless configuration and Preflight. Connecting a workspace is a separate step.
•Edit a claim in place on the proposal.
•When verification blocks handoff, "Add a verification check" saves a repeatable check to run. Saving does not record a pass.
•Blockers other than open choices are listed under the choices as "Also blocking handoff".
•"Answer the open choice" brings that choice to the top of the page and marks it for a moment.
•Invitation sign-up shows the invited email address and workspace.
•check_intent_readiness takes repoRoot, so agents in clients without workspace roots, such as Cursor and Codex, can name their repository (@pathmode/mcp-server 1.32.0).
•Preflight reads specs with "## Problem" and "## Proposed outcome" headings, with or without frontmatter (@pathmode/mcp-server 1.32.1, @pathmode/cli 2.3.1).

September 23, 2026

Features

Confirm or reject one claim in an agent’s proposal

Workspace owners, admins, and editors can confirm or reject a single part of an agent's proposal: the problem, an outcome, a hard boundary, or a failure case. The judgment is recorded under your name, tied to the wording you read, and included in what the coding agent sees next. It does not authorize implementation or change the readiness result; authorization stays a separate decision. On a repository proposal, a rejection offers Ask the agent to correct it, which carries your reason into the correction. The header counts how many claims are confirmed, rejected, or changed since you judged them.

Discuss a product choice before you answer it

Each open choice on a repository proposal has Discuss this choice, with starter prompts to explain the trade-off, compare the alternatives, or ask what evidence supports it. When the discussion suggests an answer, Review proposed answer opens the choice with that answer ready to use. Nothing is recorded until you submit it yourself.

Invite a new reviewer to one feature

The Share dialog now has Invite a new reviewer. Enter an email to create an editor invitation for that feature. Once they accept, the invitee lands on the feature they were asked to review.

Changed

•The Intents board is now a searchable list.
•New workspaces start from a brief.
•Outcomes can be recorded only after an intent ships.

Removed

•The Relationships view. Agents can still read relationships with analyze_intent_graph.

Fixed

•Shipped or authorized intents no longer ask for authorization again.
•Proposals with product choices now list their outcomes.
•Opening repository setup no longer discards unfinished attempts elsewhere.
•A brief that fails to draft is kept for manual editing.
•Generated briefs with empty optional fields now save.
•Your own brief is no longer marked as an agent suggestion.
•Keeping a single suggestion no longer fails to save.
•People who stay in onboarding now get the welcome email.
•check_intent_readiness uses the intent you name when several folders contain an intent.md (@pathmode/mcp-server 1.31.2).

Improvements

•The intent page shows one primary action, such as "Answer the open choice" or "Authorize as is".
•Each action has one name. "Request changes" always goes to the repository agent.
•The intent page states each fact once.
•The selected product choice opens inside its own row.
•Workspace home shows intents with unanswered product choices.
•Unfinished briefs are kept in your browser and can be resumed from product home.
•Repository setup is reachable from Settings, product home, and the product list.
•Evidence items show their source, and "Add evidence" opens in place.
•Intent rows show titles on up to two lines.
•The developer docs start with the review workflow.
•The About page lists key company facts.

September 20, 2026

Features

Answer product choices before authorizing the build

Repository proposals can carry open product choices with recommendations, rationale, alternatives, and trade-offs. Your answer requests a change; the coding agent applies it to the repository intent. Review the resulting revision before separately authorizing implementation. Leaving a choice undecided keeps it open and blocks authorization.

Recorded in the Pathmode app, in an example workspace. Fieldnote is a fictional product.

Follow a decision from the request to the next coding session

The homepage walkthrough now separates an answer request, an agent-applied revision, and authorization. Follow the same cancellation example through access timing, resuming, and overdue payments in the complete review and the engineer’s handoff walkthrough.

Fixed

•Settled product choices no longer trigger a competing generated review question.
•Prerequisite conflicts explain which decision needs attention instead of reporting a generic save failure.
•Applied-choice summaries show the recorded answer instead of incorrectly reporting that product choices were removed.

Improvements

•Repository reviews lead with the recorded problem and a numbered choice list showing dependencies, pending answers, and resolved decisions.
•Advanced editing is available from More options, keeping proposal review as the default page.

September 13, 2026

Features

Review what the agent understood before it builds

Compare your original ticket or brief with the agent's proposal, with a key question to settle and a suggested answer. Request a correction, ask a question, or authorize implementation of the whole proposal at the revision you reviewed. Corrections go back to your coding agent; authorization is a separate decision. When the agent returns, the page leads with what changed after your correction.

Keep the original request beside the proposal

Your coding agent can now attach the ticket or brief verbatim when it creates a proposal, so you can check its interpretation against the original. @pathmode/mcp-server 1.28.0 adds attach_original_request. Workspace editors can also attach the request directly from the review page. Attached requests arrive in the evidence inbox as unreviewed and do not count as evidence backing the proposal's claims.

Share one feature with the teammate who decides on it

Share this feature gives you a teammate review link tied to the revision you are reviewing. Workspace members land on that feature after sign-in and see whether the product manager or coding agent needs to act next. If the proposal has changed since the link was sent, the page shows both revision numbers and asks them to review the current proposal before acting.

Invite the product manager as an editor

Team invitations now default to editor, so invited product managers can request corrections and authorize implementation. Each role explains which actions it allows. The free plan includes three editor seats for an owner, a product manager, and an engineer.

Fixed

•If an agent cannot apply your correction, the review now shows why and who acts next. It tells you whether the agent can retry or you need to request the correction again against the current revision
•Failed review analyses now say they could not be completed, instead of suggesting the proposal matches your request
•A failed relationship rerun no longer shows its error beside "No new relationships found"
•Invited teammates can create an account and return to their invitation without being asked to create a separate workspace
•Linking evidence to a claim during review no longer blocks your coding agent from applying a correction
•Requests attached by an agent now appear reliably in the proposal review. Retrying the same attachment does not create a duplicate
•Failed evidence links now report an error instead of reporting success
•Incomplete AI findings are rejected before they can be saved

Improvements

•Reviewers share one question and recommendation per revision, so repeat visits use no additional AI allowance. Re-importing the ticket clears the outdated question
•Correction fields start with the current wording. Choose "Use this wording" to adopt Pathmode's suggestion before sending
•Review questions and recommendations use the language of your request. Suggested corrections use the language of the text they replace
•Workspace home lists each intent by title, with its revision status underneath, making pending proposals easier to tell apart
•When a discussion suggests a change that must go through your coding agent, you can open a correction with that wording directly from the notice

September 11, 2026

Features

Sign up with an email link

Create an account by clicking a link sent to your email, with no password to choose. Google signup is also available. Existing accounts can still sign in with their password.

Name your workspace during signup

Choose your workspace name and URL as you sign up. The URL follows the name until you edit it yourself, and you can see when a URL is already taken.

See when your coding agent is connected

New workspaces start with a Connect your coding agent intent. It completes when an agent reads it using your workspace key. If setup needs to wait, the step stays open so you can return to it later.

Follow your teammate’s setup progress

A dedicated handoff page tracks Invitation sent, Teammate joined, and Repository connected. Start your brief while your teammate connects, then return to the handoff from workspace home. Connection status reflects the invited teammate’s setup and shows when their credential needs renewal.

Removed

•New workspaces no longer include the fictional Acme HR product. Existing workspaces keep their sample content, and demos still include it

Fixed

•Email signup links now open an existing Google account for that address instead of creating a duplicate
•You can go back to correct your email address and request a link immediately for the corrected address
•Invalid workspace names return you to the form with an explanation and your input preserved
•Confirming signup in two tabs now opens the same workspace in both
•The email confirmation screen now describes the signup path you chose

Improvements

•The welcome email arrives after you confirm your address and explains how to get started in your workspace
•The signup page keeps pricing visible and simplifies the steps before account creation

September 9, 2026

Features

Connect a repository as you get started

Setup now asks who runs your coding agent. Connect the repository yourself, invite the teammate who runs it, or start from a brief if you have no repository ready. Your repository holds the spec, while Pathmode holds its evidence and review decisions.

Invite a teammate to connect the agent

Send an invitation that takes your teammate directly to repository setup. You can start writing while they connect, and follow the connection status from your own screen.

Try the readiness checks before signing up

Run six deterministic readiness checks in your browser on an example spec or one you paste in. Nothing is uploaded and no account is needed.

Fixed

•Signup works when the preceding decision did not name a product
•The Pathmode logo on the preflight page now returns to the homepage

Improvements

•Questions deliberately left open are distinguished from questions you have not reviewed, both in the review and the exported spec
•Pre-signup decision reviews let you accept, edit, or leave each suggestion open individually
•Continue from a product decision to signup and proposal review without a separate workspace step
•Briefs record where in the product they were started

September 8, 2026

Features

Review a proposal from your agent

Start with a short description and choose Draft with agent, or open a proposal created through the API. The review leads with the problem, intended outcomes, trade-offs, success measures, and open questions. Implementation details remain available alongside the review.

Review suggested additions

Agent additions are marked Suggested until you keep, edit, or remove them. Review individual claims or keep the remaining suggestions together. Authorizing the whole proposal is a separate, recorded decision.

Authorize the revision your agent will use

Agent-originated intents require a human decision before implementation. For repository-owned intents, authorization applies to the exact revision you review, and a new revision requires a fresh decision. Request changes from the proposal and follow your coding agent’s response.

Improvements

•Workspace and Product Home prioritize proposals needing a decision, suggestions to review, and work waiting on an agent
•Intent rows distinguish proposal review, suggestion review, revised proposals, and waiting for an agent from delivery status
•The conversation editor remains available for discussion and detailed editing
•Agent-written outcomes and edge cases retain their suggestion and evidence references through ordinary saves

September 3, 2026

Features

See what backs each claim

Objectives, outcomes, constraints, and edge cases now show Backed, Unreviewed, Assumption, or No evidence. Expand a claim to inspect its sources and reviewers, or choose Use as backing to review a signal and link it to that claim. Agents can capture signals, but only people can review them. A claim without evidence remains visible as a decision and does not itself fail readiness.

Keep evidence and assumptions distinct in handoffs

Agent prompts, exports, and review links distinguish reviewed backing, unreviewed signals, and AI-drafted assumptions. Archived and dismissed items are excluded from these handoffs, and assumptions do not count as evidence coverage. MCP server 1.24.0 includes these classifications and adds get_intent_evidence to retrieve the items cited by an intent.

Review pull requests against the intent before merge

Pull requests referencing an intent receive an advisory Pathmode intent review check in GitHub when opened or updated. Each review records the code commit and intent revision it assessed. Workspace writers can accept, reject, or waive findings with a reason; the check does not block merging. Incomplete reviews are marked unavailable.

Keep private evidence out of committed files

intent.md, CLAUDE.md, AGENTS.md, and .cursorrules now reference evidence by type and ID. Quotes and source names stay in Pathmode, where connected agents can retrieve them through MCP.

Removed

•The Evidence board’s Kanban, Flow, and Clusters layouts have been removed. The Inbox and evidence detail panel remain
•Evidence Browse is now a flat, searchable list. Grouping, bulk selection, and Suggest intents have been removed; review claim backing on the intent itself
•New evidence links in the signed-in UI must identify the claim they support. Existing general links remain visible, and API and MCP linking behavior is unchanged

Improvements

•Evidence records distinguish human-reviewed items, imports awaiting review, and agent-created signals
•Sample tickets and example cards are labeled as unverified assumptions
•Paste text into the composer to extract and review signals inline; use the import dialog for files

August 31, 2026

Features

Connect an existing intent.md to Pathmode

Choose a workspace and product during setup, then run the adoption command from your repository. It preserves your intent’s text, adds its Pathmode identity, and opens a review link for your teammates. Available in @pathmode/mcp-server 1.20.0.

Connect repositories to separate workspaces

Repository setup now keeps each repository’s workspace binding separate. API keys stay outside git, and generated .mcp.json files contain no credentials. Connecting another repository no longer changes the workspace used by your other repositories.

Improvements

•If repository adoption fails during a network step, you can resume without creating duplicate credentials
•Adoption codes are single-use, expire after 15 minutes, and have rate-limited exchange attempts

August 30, 2026

Features

Use Pathmode with OpenSpec

Check an OpenSpec change with the free local preflight, or add the experimental pathmode-intent schema to write intent.md before drafting the proposal. Repository files carry summaries and references, while private evidence stays in Pathmode. This is an independent integration. Explore the OpenSpec integration.

Contribute to the IntentSpec standard

The specification, JSON Schema, conformance examples, and governance now live in pathmodeio/intentspec. The spec also defines how to keep committed intent files free of private evidence, without changing the schema.

Read IntentSpec consistently across tools

The reference implementation, GitHub Action, browser, and MCP/CLI parser now share a normalization test corpus. All accept the documented edge-case formats, so the same spec produces the same structured content across tools.

Use the repository’s intent for local checks

Readiness checks, review-against-intent, and handoff-intent now prefer the repository’s intent.md. Workspace data adds findings, notes, and authorization to that intent. Workspace selection is used only when no repository-bound file exists. Available in MCP server 1.17.0 and plugin 0.1.12.

Pass handoff notes to the next agent

Agent prompts now include the ten most recent implementation notes alongside open findings. If older notes are omitted, the prompt shows how many remain. Notes provide context for the next session while the spec continues to define the work.

See intent status when Claude Code starts

Plugin 0.1.12 shows the intent title, status, days since its last edit, and subsequent commit count when you open a repository containing intent.md. This experimental notice runs locally without a key or network request.

Fixed

•The Engineering Lead and First Intent use case now opens the flow that works without an API key
•Plugin installation instructions are consistent across the homepage, preflight, intent compiler, and developer docs
•Page titles and blog metadata now describe their current content

August 29, 2026

Features

Import and export OpenSpec changes

intent_import reads an OpenSpec change into a new intent while preserving its text and readiness result. Export in OpenSpec format creates proposal.md, a spec delta, and verification tasks in a new change folder. The receipt lists content that stays in OpenSpec, including design prose and existing task lists. Available in MCP server 1.16.0; imports create a new intent and exports do not overwrite existing change folders.

Confirm readiness dimensions in your workspace

confirm_intent_dimension now works with workspace intents as well as local files. Confirmations bind to the content reviewed, and concurrent edits produce a conflict so you can review the latest version. Cloud waivers remain unavailable because they require a human decision.

August 28, 2026

Fixed

•Workspace readiness checks now read edge cases correctly, fixing a false failure for specs that already defined them. Current-intent reads and outcome rubric exports also include edge cases. Fixed in MCP server 1.14.1

August 26, 2026

Features

Bring code context from your agent into the spec

Your coding agent can record relevant files, current behavior, risks, and verification suggestions from the repository, including uncommitted work. This context reaches the next agent through Pathmode. Suggested risks and checks enter the Outcomes rubric only after you accept them, and you can withdraw that acceptance.

Link a pull request to verify the merged work

Reference an intent in your branch name or pull request body. On merge, Pathmode grades the diff against the spec, records the result, and marks the intent Shipped. Agent instructions now include adding this reference and leaving the shipped transition to the merge.

Save agent-created specs to your workspace

With a workspace connected, intent_save now creates or updates the workspace intent and writes its local intent.md. Local specs also receive IDs that can be referenced in pull requests. If multiple products could own a new intent, the save asks you to choose.

Fixed

•Merged pull requests now update intents when a product has its own GitHub connection
•A merge only updates intents associated with that repository

Improvements

•Concurrent edits to a spec report a version conflict instead of silently overwriting each other
•The new record_implementation_context MCP tool lets agents submit code context

August 23, 2026

Features

Try Preflight in your browser

The free Claude Code plugin has a new home at preflight.pathmode.io. Try six readiness checks on two example specs directly in your browser. The plugin uses the models you already have access to in Claude Code.

Authorize agent proposals before implementation

Agent-authored specs start pending human authorization, and their prompts tell the agent to wait. Authorize or reject a proposal from your signed-in session; rejection notes reach the agent verbatim. You can revise your decision later.

Import a Linear or Jira backlog as evidence

Import open issues from a Linear team or Jira project into the Evidence Inbox. Issue URLs prevent duplicates, and labels help classify bugs as friction and other issues as requests. Undo removes unlinked imported items and detaches the others from the import.

Verify work against the pull request diff

Provide a pull request to have Pathmode fetch its GitHub diff and assess outcomes, constraints, constitution rules, edge cases, and verification checks. You can still submit a written summary, which is treated as a claim about the work.

Continue from a finished spec to agent handoff

After you accept Finalize, the assistant offers the next step for your workspace: connect an agent, complete an unused setup, or hand over the spec. Shipped work receives no setup prompt.

Describe how the product works today

The Current State field records existing behavior for agents to consider when making a change. It appears in intent.md, agent exports, and the Outcomes rubric, and can be read and written through MCP server 1.10.0.

Removed

•Creating GitHub Issues from intents has been removed. GitHub repository analysis and merge verification remain available, as do issue creation in Linear and Jira

Improvements

•The setup checklist now includes connecting a repository for codebase analysis
•Pasted customer quotes and measurements retain their types and source references

August 22, 2026

Features

Find agent-created specs awaiting your review

Specs created through the API or MCP show agent authorship and appear at the top of the workspace review queue. Workspace owners receive at most one notification email per day. Opening the spec clears that review notification.

Measure outcomes after shipping

Record a baseline, measured value, and target result for each shipped outcome. Product Home summarizes progress and highlights outcomes awaiting measurement. Reordering outcomes preserves their measurement history.

Record decisions directly in the spec

Add, edit, or remove decisions, including rejected alternatives and reasons to revisit a choice. A decisions API lets integrations and agents add to the same record. When cited evidence changes or disappears, the review queue highlights affected decisions.

Tailor setup and interviews to your role

Choose your role in the first interview to focus the conversation on product outcomes or engineering constraints and verification. Product roles get setup guidance for evidence and teammate review. You can change your role in your profile.

Install the Claude Code plugin

Run /plugin marketplace add pathmodeio/claude-plugin, then /plugin install pathmode@pathmode to install the MCP server and seven skills. Without an API key, specs stay in your local intent.md. Add a workspace key to use connected features; the key is stored in your OS keychain.

Fixed

•The shipped-status gate now recognizes verification checks created in the current editor
•Unsubscribe links now require a valid signed token
•Outcome measurements require a workspace writer role; viewers retain read access

Improvements

•Connector imports now identify their integration source
•The product setup checklist detects team and agent connections and closes when complete
•Developer docs now include decisions and implementation-findings endpoints

August 21, 2026

Fixed

•Installing without an API key now automatically starts local mode and exposes the tools that work with intent.md
•Saving the same intent updates it; replacing a different intent requires an explicit overwrite
•Reading intent.md now preserves multiline sections, scope, and decisions
•The setup command now completes without an API key

June 21, 2026

Features

Draft your first intent before signup

Describe what you are building on the landing page and review a proposed objective and outcomes. Assumptions remain labeled separately from evidence. Signing up saves the draft you reviewed.

Build product context from a document

Paste a PRD, positioning page, or notes to extract your vision, north star, audience, constraints, and assumptions. A context strength meter highlights gaps. Additional documents fill empty fields and offer conflicting content for review without overwriting your choices.

Bring implementation findings back into the spec

Verification checks now travel with agent handoffs. Agents can record contradictions discovered during implementation, and open findings reach the next agent’s prompt. Review or dismiss each finding in the editor; findings that invalidate a check mark it failing. Read Verification Is a Feedback Loop.

Update repository instructions with sync_context

The MCP server can update the Pathmode section of CLAUDE.md or AGENTS.md from your intent. Repeated runs update the existing section without adding duplicates or changing unrelated content.

Fixed

•Archived evidence is excluded from agent prompts and exports

Improvements

•The landing-page draft asks a focused follow-up question to clarify a vague description
•Product Home displays the strength of your product context
•Pasting a substantial document during signup handoff starts extraction without asking you to paste it again
•Pre-signup AI requests now use bot checks and rate limits

June 18, 2026

Features

Define scope and verification in IntentSpec v1.2

The IntentSpec format adds optional scope and verification fields. Specify what a change may touch and how outcomes will be checked through automated tests, manual checks, and production signals. The update is backward-compatible; the specification and JSON Schema include both fields.

June 15, 2026

Features

Organize evidence by its properties

Group evidence by type, severity, source, or status, and switch between list and board layouts. Edit a signal’s text and properties in its detail panel, and link or unlink intents there.

Import evidence from a remote MCP server

Add a remote MCP server through Add Evidence → MCP source to extract typed signals from its resources. This first version supports manual sync of resources, and imported items show a Synced label.

See which decisions need your attention

Workspace and Product Home prioritize specs needing review, unresolved questions, untriaged evidence, and critical friction without an intent. Each item links to its next action, with recent work available below.

Edit a scanned workspace before signup

Scan a URL, then edit the draft product, evidence, and header directly in the preview. Sign in from the same page to save the draft you reviewed.

Let teammates join through their company email

Team-plan workspaces can enable automatic viewer access for teammates with verified company email addresses. An owner or admin must verify and claim the same domain first. Personal email providers are excluded.

Ask an AI about Pathmode

Links on the homepage, pricing, comparison, and product-team pages open Claude, ChatGPT, or Perplexity with a question ready to send. Competitor pages include a comparison question.

New essay: The Cost of Being Worth Using

Explore why faster software delivery still needs product judgment to create something people use. Read the essay.

Changed

•The role-selection dialog no longer blocks the first workspace screen. Your role remains editable in Profile
•Legacy /journey-map URLs redirect to /product-teams

Fixed

•Atlassian and Jira integrations now verify installation bindings before resolving intents or assigning workspaces, preventing access across workspace boundaries
•The Anonymization Firewall holds data back if redaction cannot complete
•Google signup analytics now record correctly, and signup forms prevent duplicate submissions
•Evidence suggestions no longer trigger an infinite render loop
•A failed workspace load no longer sends you to workspace creation

Improvements

•Settings and the Intents board use simpler controls and quieter styling
•Creating a first product without a scan now opens its intent editor
•New accounts derive a display name from the signup email
•Account deletion now records an analytics event and notifies administrators

June 12, 2026

Features

Capture and review evidence inline

Type a signal in the Evidence composer for automatic classification, or use / to choose its type. Paste a support ticket to extract multiple signals, then keep or discard each one inline.

Edit evidence in place

Hover an evidence row to edit its text, type, and severity without opening another page.

Improvements

•Empty products show sample signals you can adopt to get started
•The evidence composer is available in List, Clusters, and Flow views
•Single-line observations are automatically classified like longer pastes

June 8, 2026

Features

Explore advanced Playbook guides

Three new guides cover product constitution rules, verification feedback loops, and coding-agent setup. The Playbook now separates introductory and advanced learning paths.

Read developer docs generated from the API

The developer reference now draws its tool lists, arguments, and OpenAPI schema from REST API and MCP server definitions. Automated checks help keep the documentation aligned with shipped behavior.

Fixed

•Health Metrics now appear in Markdown agent prompts as well as Claude Code and Cursor exports
•Broken Get started links in Playbook guides now open onboarding

Improvements

•Playbook guides include previous, next, and related links with clearer difficulty levels
•Playbook, glossary, and comparison pages now explain Scope and Health Metrics alongside the other IntentSpec fields

June 6, 2026

Features

Explain and enforce workspace rules

Constitution rules now include a rationale, scope, enforcement level, and examples. Edit each rule and preview how agents will see it. AI Spec Review, exports, and execution prompts carry these details, including advisory, required, or blocking enforcement.

See exactly what blocks readiness

The Spec panel and AI now use the same deterministic readiness checks. Each requirement shows its state and any blocking reason, so the coverage display agrees with the readiness verdict. Evidence remains advisory.

Install agent skills and export an outcome rubric

MCP server 1.6.0 bundles seven Claude Code skills for drafting, challenging, reviewing, and verifying intents. Install them with npx @pathmode/mcp-server install-skills. The new outcome-rubric export turns your outcomes, constraints, edge cases, and checks into writer instructions and a separate grader rubric for Claude Managed Agents.

Fixed

•Intent conversations offer a retry when the AI fails
•Saves no longer race between competing copies of the spec
•Fast conversations no longer create duplicate specs, and autosave keeps each intent’s baseline separate
•AI streaming preserves updates split across response chunks
•Scope entered in conversation now appears correctly in the spec
•Returning to a conversation phase no longer repeats the greeting
•Demo requests submitted from the site are now saved correctly

Improvements

•The Spec panel places the title and key properties above the document body
•Constitution templates include enforcement levels and reasons for their rules
•Site scans have stronger URL and network checks, more write operations are rate-limited, and integration webhooks reject replay attempts
•Security updates cover dependencies and database permissions

May 27, 2026

Features

Include evidence in IntentSpec v1.1

The IntentSpec format adds an optional evidence field for the signals behind outcomes, constraints, and edge cases. The update is backward-compatible. The standard and schema now have a dedicated home at intentspec.org.

Compare Pathmode with other tool categories

New comparison pages explain how Pathmode relates to issue trackers and spec-as-contract tools. Individual vendor comparisons link to their category, and Linear has a new comparison page.

Fixed

•Returning users reopen their existing workspace instead of creating a duplicate
•Site scans no longer reject visitors solely because a bot-check script failed to load; rate limits remain in place
•Reopening a spec no longer records past conversation turns as new activity, and autosave failures are tracked

Improvements

•You can skip the site scan or continue without it after a scan fails. New workspaces still include sample content
•Product pages now explain how evidence, decisions, and code context support work before and after shipping

May 25, 2026

Features

Ground a spec in your existing codebase

Choose Analyze codebase to identify relevant files, current behavior, risks, and suggested verification in a connected GitHub repository. Available on paid workspaces with a GitHub App connection and a default repository. Read about Implementation Context.

Include code context in agent handoffs

Claude Code, Cursor, and MCP agent prompts now include the intent’s Implementation Context: relevant files, existing behavior, risks, and verification suggestions.

Choose a repository for each product

A product can use a different GitHub repository from the workspace default. Set it in product settings under Code binding.

May 21, 2026

Features

Start a workspace from your website

Paste a public URL to draft a product, extract signals from your site, and get questions for your first intent. Review and edit the preview before signing up to save it.

Choose which product to scan

For sites with multiple products, choose the one you work on before Pathmode drafts its context. The scan then focuses on that product.

Describe your product when a scan cannot read it

If a site requires login or cannot be scanned, paste a description to draft the context. The original URL stays attached as a source reference.

Start your first intent from the product page

Scanned workspaces show a Draft your first Intent prompt until you create one.

Changed

•Public signup buttons and bookmarked signup links now open the site-scan flow
•The sign-in dialog links new users to the same signup flow
•Signing in without a workspace now opens site-scan onboarding

Removed

•The previous first-run welcome panel has been replaced by site-scan onboarding

Improvements

•The setup checklist reflects completed site scans and the next step of drafting an intent
•A scanned description satisfies basic product setup; vision and north star can be added later

May 19, 2026

Features

Gemini 3.5 Flash for global workspaces

Global workspaces now use Gemini 3.5 Flash for AI features, including intent conversations. EU-residency workspaces continue to use Gemini 2.5 Flash.

May 11, 2026

Features

Pathmode for Jira — Atlassian Marketplace

Submitted "Pathmode for Jira" to the Atlassian Marketplace review queue. A Forge app that surfaces the linked Pathmode intent — objective, evidence, edge cases, status — inside every Jira issue. Authenticated with Atlassian-signed JWTs verified against the public JWKS; read:jira-work scope only.

Push to Jira → linked from inside Jira

The existing Push to Jira flow now creates an issue with the intent ID embedded in the description. The Forge panel on the resulting Jira issue resolves and renders the intent automatically — title, APPROVED/SHIPPED badge, evidence count, and a link back to Pathmode.

Bidirectional status sync

When a linked Jira issue moves to Done, the corresponding Pathmode intent flips to Shipped automatically, with an implementation note recording who shipped it. Native Jira webhook + connection-id capability token, no shared secret required.

Atlassian Forge App workspace binding

New section in Workspace Settings → Integrations for linking an Atlassian site to a Pathmode workspace. Detected installs surface as one-click claim cards; the bridge endpoint auto-registers installs on the first JWT-verified call so the binding flow works even when the Forge lifecycle webhook is missed.

Improvements

•Privacy policy gained a dedicated section covering Atlassian integration data flows — what we receive (cloud id, site URL, issue keys, account IDs), what we send (intent summary + description on push), how we authenticate, where it is stored, retention, and uninstall behavior
•New /support page with Pathmode for Jira–specific troubleshooting and security reporting contact
•Jira Test Connection now works after Save — server-side test against stored credentials instead of requiring the user to re-type the API token
•JiraIssueModal portals to document.body so it renders above the conversation panel in the Intent Editor

May 10, 2026

Features

AI Spec Review

A new review pass critiques your spec across six dimensions before you ship it to an agent: outcome quality, edge-case gaps, verification coverage, constitution conflicts, evidence grounding, and implementation drift. Two modes — Approval (gate before execution) and PR (post-implementation review).

Codex Integration Page

New Codex guide mirrors the Claude Code agent page. Both now share a single component and registry, so adding another coding agent is a data change, not a copy.

Use Cases Hub

New Use Cases index plus ten concrete walkthroughs — anchoring outcomes to evidence, auditing a spec for edge cases, founder ask to testable goal, support tickets to a prioritized spec, hand-off to Claude Code, and more.

Auto-Linked Glossary

Blog, playbook, and glossary content now auto-link the first occurrence of each glossary term. New entries propagate to old content automatically — no manual back-linking, no rot.

AI Turn Rating

Rate AI turns inline during an intent conversation. Per-turn signal flows into analytics so we can tune prompts on real data instead of vibes.

Workspace Ownership Transfer

Workspace owners can now transfer ownership to another member from the Danger Zone settings. Backed by a role-checked Postgres function so the transfer is atomic and safe.

Evidence Board 101 Playbook

New beginner playbook on running the Evidence Board as a diagnostic play — collect symptoms, spot patterns, decide where to operate — instead of a feedback dumping ground.

Blog Subscribe

Readers can now subscribe to the Pathmode blog. Email capture, confirmation, and rate-limited handling all in-app.

Changed

•Robots.txt moved from static `public/robots.txt` to dynamic `app/robots.ts` — adds disallows for /api/, /intent/, demo, and review routes, references the sitemap
•Dropped `offers` from SoftwareApplication structured data — Pathmode is SaaS, shopping-result fields do not apply

Removed

•Legacy Figma plugin package (was drifting from the current intent model)
•Legacy "Journey Mapping 101" playbook — replaced by Evidence Board 101
•Standalone `ClaudeCodePage` component — replaced by the shared `AgentIntegrationPage` used by both /claude-code and /codex

Improvements

•Spec Panel — Refactored, splitting execution hand-offs (Jira / Linear / Build Queue) and lifecycle controls (status, verification, review) into focused components. Smaller surface, fewer re-renders
•Intent Quality scoring centralized in one library so the Spec Panel, AI Spec Review, and readiness checks all use the same rubric
•Intent Interview JSON schema (with tests) replaces implicit field-name agreements between AI and client
•Glossary entries now cross-link to each other and to core terms — a small built-in knowledge graph
•New unified workspace integrations API across Jira, GitHub, Linear, and Slack with proper credential redaction
•Workspace settings — new Rename and Export Context endpoints, plus tracking for conversation-fired AI guidance rules
•Landing page hero, copy, and section structure tightened around "know what to build before you build it"

April 30, 2026

Features

First-Run Workspace Onboarding

New empty-workspace onboarding hero asks for one concrete product problem, creates the first Product and Intent, and sends the seed straight into the Intent Editor. Fresh users now see Pathmode form a spec immediately instead of landing in an empty workspace.

Onboarding-Aware Intent Conversation

The Intent Editor now supports onboarding mode with an open spec panel, seed handoff through session storage, first-response analytics, and a softer first AI turn that still drafts an objective right away.

Improvements

•Free plan now includes 3 editor seats (up from 1) — small teams can get started without hitting the paywall
•Product creation now returns full metadata, allowing onboarding and analytics flows to tie created products to follow-up intents without extra lookups
•Paid-plan CTAs open the in-app contact modal when billing is not enabled, keeping interested teams inside the Pathmode experience
•Updated primary landing CTAs to "Start for free" and tightened homepage messaging
•Simplified product demo animation state to reduce timer/state drift
•Onboarding analytics across the full activation funnel — hero views, submissions, skips, first AI response, and first spec update

April 6, 2026

Features

Intent Readiness Scoring

Pathmode now evaluates whether your spec is ready to hand off to an AI agent. Six blocking checks cover title, objective, outcomes, constraints, edge cases, and verification; supporting evidence is reported separately as an advisory signal.

Smarter Context Interviews

The product context builder runs a deeper interview flow, producing richer product context that feeds into every spec your team writes. Better context in, better specs out.

Agent Prompt Generation

The MCP server now generates full execution prompts for intents with readiness checks. AI agents get structured instructions with objectives, outcomes, constraints, and verification steps.

Plan Changes & Team Management

Workspaces can now upgrade, downgrade, and manage team members directly from billing settings. Full Lemon Squeezy integration for checkout, portal, and subscription sync.

Fixed

•Billing status hook, checkout flow, and portal redirect issues across multiple plan types

Improvements

•API Reliability — Rewrote intent API routes to share a single, tested handler layer with consistent validation
•MCP Server — Expanded tool surface: enhanced local file reading, better intent compilation, improved export formatting
•Context Health — Updated indicators so teams see at a glance which products have strong context and which need attention

March 29, 2026

Features

Functional Roles

Introduced role-aware Pathmode onboarding and profile settings for Product, Engineering, Design, Founder, and Mixed. Pathmode now tailors AI interview prompts and question flow based on the lens you bring to the work.

Improvements

•Role-aware onboarding now prompts first-time users to choose their lens from workspace home
•Functional role selection is saved to the user profile and can be updated later in Profile Settings
•Added database support for functional_role with a safe default and validation constraint

March 28, 2026

Features

Evidence Inbox

The biggest update since Pathmode went live. New dedicated layout mode for triaging evidence at speed. Single-item focused view with keyboard-driven actions: j/k to navigate, a to accept, d to dismiss, l to link to an existing intent. Pagination-aware — advances to the next batch automatically. Includes a Link to Intent modal that ranks existing intents by keyword match.

AI Intent Synthesis

New "Synthesize from Evidence" workflow: select evidence items and AI proposes intent specs grounded in real user problems. A new intent chooser modal lets you pick between blank/template creation and evidence-based synthesis. Also added Suggest Relations — AI proposes dependency relationships between intents.

Structured Outcomes

Outcomes upgraded from plain strings to structured objects with stable UUIDs and optional MoSCoW priority (must/should/could). Evidence anchor keys now reference outcome UUIDs instead of array indices, making anchors stable across reordering and insertion.

Intent Scope

New scope field on IntentSpec with explicit inScope and outOfScope lists that tell AI agents what they can and cannot touch. Supported in the editor, API, and MCP server.

Intent Engineering vs Context Engineering

New blog post exploring why intent is upstream of context when building with AI. Teams optimize AI briefing before figuring out what to ask — better context just helps AI fail more efficiently.

Changed

•Removed beta banner — header now shows "Pathmode has launched!" announcement
•Restructured workspace settings with new API Keys tab

Improvements

•MCP Server — Structured outcome support, live quality checking with explicit rubric, scope field support, product strategy in exports
•Document View — Keyboard navigation in outcome lists, structured outcome rendering with priority labels, provenance view integration
•Evidence Board — New Inbox and Sources layout modes, layout selection in sidebar sub-menu, triage status counts in badges
•PostHog — Server-side event tracking with session and distinct ID headers

March 21, 2026

Features

Conversation-First Intent Editor

Complete redesign of the intent editing experience. Spec crystallizes from dialogue rather than form fields. Natural language conversation is now the primary surface, with a collapsible "Spec panel" sidebar showing structured output. The AI displays spec updates with immediate auto-apply, visual indication of what changed, and a 10-second undo window.

ThinkingQuality Scoring

4-dimensional assessment (specificity, testability, coherence, tension) scored 0-3 each, replacing the previous SpecCompleteness metric for more nuanced feedback.

Evidence-Anchored Specs

AI auto-links outcomes and edge cases to product evidence items. Specs display inline badges showing evidence links that persist across updates.

Changed

•Legacy form-based editor available via ?mode=form query parameter; conversation-first is now default

Fixed

•Intent editor save flow to properly persist spec updates and undo history
•Navigation issues between intent editor and product home

Improvements

•Intent Editor Performance — Consolidated state machine and reduced re-renders
•UI Consistency — Spec panel uses workspace design system with gray tones and no decorative hover effects

March 14, 2026

Features

Constraints Document Discoverability

The constraints document feature is now surfaced across four key touchpoints. Upload directly from the Context Builder guardrails phase. See a tip when creating a new intent. The AI mentions the feature when no constraints exist. And the Product Home always shows a CTA to add one.

Improvements

•Context Builder guardrails phase now includes a file upload button for constraints documents (PDF, DOCX, TXT, MD, CSV)
•New Intent modal shows a tip about constraints documents when none is uploaded for the product

March 11, 2026

Features

Smarter, Faster AI Conversations

Major overhaul to the Intent Editor and Context Builder conversations. The AI now responds instantly with real-time streaming, stays on topic instead of repeating itself, and progresses naturally through the spec-building process. Selecting quick-reply chips sends your answer immediately — no extra click needed.

Fixed

•AI would loop on the same topic or re-ask questions you already answered
•Conversations could get stuck on "Thinking..." with no way to recover — retry options now appear on failure

March 9, 2026

Features

PostHog Analytics Integration

Full-stack analytics with client-side event tracking and server-side error capture via PostHog EU, routed through a reverse proxy.

In-App Feedback System

"Send feedback" in the sidebar now opens a modal that emails the team directly via Resend (replaces mailto: link).

Product Creation Onboarding Flow

After creating a product, a guided welcome step offers "Get Started" (Context Builder) or "I'll explore on my own".

Changed

•Evidence Board redesign — Tabbed layout (Evidence / Sources), sort dropdown, more compact cards, dedicated "Suggest intent" button
•Intent Editor spec panel widened from 480px to 560px for better readability
•Product Home — Strategy section split into compact two-column card (Vision | North Star); tighter spacing throughout
•Insights Digest — Renamed from "Signals Digest"; critical items now use amber highlights; "View all" links to Evidence
•Hero demo — Streamlined from 7 steps to 6 (Signals step removed)

Removed

•Entire Signals system — Views, hooks, services, API routes, cron job, MCP tools, Slack notifications, and all UI references
•Glyphex analytics tracker — Replaced by PostHog
•Outcome quality scoring — Removed per-outcome quality badges (Testable/Decent/Vague) from the Intent Editor

March 6, 2026

Features

Hand-Drawn Illustrations

Added custom hand-drawn editorial illustrations to the Philosophy and Product Teams pages. The ink-and-teal style matches the existing blog illustration series — depicting the traceability pipeline (Friction → Evidence → Intent Spec → AI Agent → Shipped) and the three-role collaboration model (Discover, Define, Deliver).

March 5, 2026

Features

Landing Page Refresh

Refreshed the landing page with a new hero headline (“Know what to build before you build it”), an updated interactive product demo that reflects the current UI across all 7 screens (Products, Home, Signals, Evidence, Synthesize, Intents, Export), and a tighter page structure — down from 11 sections to 9. Removed redundant diagram and Living Specs sections. Moved the Context Layer section higher and Metrics closer to the CTA.

Philosophy Page Rewrite

Rewrote the Our Approach page with a horizontal pipeline flow, three core convictions, and streamlined five-stage walkthrough. Replaced the confusing vertical diagram and removed the redundant product demo embed. Every section is now scannable in under 5 seconds.

Improvements

•Navigation link renamed from "Our Philosophy" to "Our Approach"
•Hero demo now shows workspace vs. product sidebar modes
•Each demo step includes a description bar explaining the screen
•Removed unused IntentEngineDiagram and WhyHowWhenPositioning imports from landing page

March 3, 2026

Features

Gemini 3 Flash & Global AI Endpoint

All AI features now run on Gemini 3 Flash, Google's latest model, via the global Vertex AI endpoint. This means better reasoning across context interviews, spec generation, and evidence analysis. The AI Processing Region setting lets workspaces choose between Global (best availability, latest models) and European Union (data residency compliance in europe-west1).

March 1, 2026

Features

Living Specs — The Spec Lifecycle

Specs are now living documents. Every spec in Pathmode is grounded in evidence, reviewed by humans, consumed by agents, and improved by outcomes. This release closes the loop between what you plan, what you ship, and what actually happened.

Make specs feel alive — grounded in evidence, reviewed by humans, consumed by agents, and improved by outcomes. That's the loop that makes Pathmode indispensable.

Evidence-Anchored Specs

Specs now link evidence at the section level. Each outcome, edge case, and constraint can reference the specific user evidence that justifies it. Anchors are visible inline as small badges showing the count of backing evidence, and expand to reveal the actual quotes, metrics, and observations that drove each decision.

Spec Review with Evidence Backing

The public spec review page (shared via token) now surfaces evidence anchors, problem severity, and health metrics. Reviewers see exactly which outcomes are backed by evidence and which are assumptions. Comment buttons are always visible on mobile for better touch targets. Unresolved comment count shows in the sticky header.

Reactive Signals

Signals are no longer manual. When new evidence is created, Pathmode automatically matches it against existing approved and validated intents using keyword and tag overlap. Matches appear as “Evidence Match” signals in the Signals view with one-click linking. The sidebar nav shows an active signal count badge. Five signal types are now persisted: evidence matches, freshness alerts, loose evidence, stale intents, and alignment drift.

Agent Verification Loop

Shipped specs can now be verified by AI. Click “Verify with AI” on any shipped or verified spec, provide an implementation summary, and Gemini 2.5 Flash grades every outcome, constraint, constitution rule, and edge case as pass, fail, or unclear. Results show inline with per-item reasoning, grouped by category. Failed verifications automatically create signals. Build Queue cards show verification status and score at a glance.

Improvements

•Evidence freshness indicator on specs — shows amber warning when linked evidence is older than 30 days
•Signal count badge in sidebar navigation
•Verification score badges on Build Queue kanban cards
•Verification failure signals with upsert dedup — re-running verification updates the signal instead of creating duplicates
•Updated AI verification model from Gemini 2.0 Flash to Gemini 2.5 Flash

February 22, 2026

Features

Intent Compiler for Claude Code

Shipped the Intent Compiler — a zero-config MCP tool that turns Claude Code into a Socratic product thinking partner. Describe a problem, paste a support ticket, or explain what's broken. Claude challenges vague language, asks one pointed question at a time, and builds a structured intent spec through conversation. When the spec is ready, it saves intent.md to your project and optionally exports as .cursorrules or CLAUDE.md. No API key or signup needed.

MCP Server v1.2 — Works Without API Key

The Pathmode MCP server no longer requires an API key to start. The Intent Compiler tools (compile-intent prompt, intent_save, intent_export) work entirely offline. Team features (dependency graphs, workspace strategy, governance) activate when you add an API key. Published to npm as @pathmode/mcp-server@1.2.2.

Official MCP Registry Listing

Pathmode is now listed on the Official MCP Registry as io.github.pathmodeio/mcp-server. Anyone searching for intent engineering, spec building, or product development tools can now discover and install Pathmode directly.

Intent Compiler Page

New /intent-compiler marketing page with interactive terminal demo, tabbed output preview (intent.md / .cursorrules / CLAUDE.md), 30-second setup guide, and REPL-to-Runtime comparison showing how the free compiler connects to the full Pathmode platform.

Improvements

•MCP server README rewritten to lead with Intent Compiler and zero-config setup
•API key changed from required to optional in manifest.json
•Fixed pre-existing TypeScript error in MCP server api-client.ts

February 18, 2026

Features

Cursor Directory Listing

Pathmode is now listed on the Cursor Directory. Install the Pathmode MCP server directly into Cursor with one click — giving your AI coding agent structured intent specs, dependency graph analysis, workspace strategy, and auto-generated .cursorrules from real product context.

February 17, 2026

Features

Integrations Hub

The workspace now supports four external integrations, turning Pathmode into a central hub for product signals. Connect your tools in Workspace Settings → Integrations and evidence flows into your Spaces automatically.

Google Analytics Integration

Connect your GA4 property via OAuth and sync analytics data directly into the Evidence Board. Pathmode fetches five report types — page performance, funnel analysis, event tracking, user acquisition, and engagement overview — then uses AI to interpret the raw metrics into actionable evidence items: friction signals from high bounce rates, metric data points, and behavioral observations. Daily auto-sync keeps your evidence fresh.

Zendesk Integration

Import support tickets as evidence items. Connect your Zendesk account, choose a target Space, and Pathmode syncs closed tickets through AI extraction — turning customer conversations into structured friction points, quotes, and feature requests.

GitHub Integration

Link your GitHub organization to connect repositories with your workspace. Pull requests can be linked to Intent Specs, closing the loop from product intent to shipped code.

Linear Integration

Connect Linear to link issues and projects with Intent Specs. Track how product decisions flow from evidence through specs into engineering execution.

February 14, 2026

Features

Claude Code Integration Page

New dedicated Claude Code page explaining how Pathmode connects to Claude Code via the MCP server. The page covers the full closed-loop system — from evidence collection through intent specs to phase-gated agent execution and verification. Includes interactive terminal mockups, the complete 12-tool MCP reference, and a step-by-step setup guide.

February 7, 2026

Features

Hero Animation as Video with Remotion

Built a shareable product demo video directly from our React hero animation — no screen recording needed. Using Remotion, we converted the existing phase-based hero component into a frame-perfect 1920×1080 video composition. All setTimeout chains became frame calculations, all useEffect animations became interpolate() calls — same JSX, same Tailwind, same SVGs. The result is a 37-second MP4 showing the full Pathmode workflow: Friction Inbox → Journey Map → AI Spec Generation → Export to Agent.

Improvements

•Added Remotion video rendering pipeline with preview and export scripts
•Downloaded and bundled Inter and Merriweather fonts for server-side video rendering

February 6, 2026

Features

IntentSpec Open Standard

Launched IntentSpec as an open standard for spec-driven development. The new IntentSpec page documents the schema, integration guides for AI agents (Cursor, Windsurf, Claude Code, GitHub Copilot), and a GitHub Action for CI/CD validation.

About Page

Added a dedicated About page with the Pathmode story, company values, and founder bio.

Philosophy & Product Demo

New Philosophy page with an animated product walkthrough. The hero demo visualizes the full workflow: friction signals flying in from support tickets, analytics, and research — AI synthesizing them into structured Intent Specs — and one-click export to your coding agent.

Playbook Guides

Expanded the Playbook with new guides including "Support Ticket to Shipped Feature" and "Why Specs Fail". Added embeddable components for flow diagrams, comparison tables, and chaos wall visualizations directly inside guide content.

Export to Agent

Enhanced the Export to Agent modal with contextual data loading. Specs now include Constitution Rules, Journey Map context, and Workspace Strategy — giving AI agents the full picture when generating code.

Improvements

•Landing Page: Redesigned hero section with interactive product demo and updated value proposition
•Solutions Pages: Added solution-specific landing pages for Journey Management, Product Intent Infrastructure, and Enterprise
•Footer: Added links to About, Philosophy, Playbook, and Comparison pages
•Terms of Service: Updated publicity rights and terms
•Intent Editor: Multiple stability and UX fixes across the spec editing workflow
•Blog: Published new articles on intent-driven development and AI-native teams
•Invitation Flow: Improved account handling and workspace invitation acceptance

February 1, 2026

Features

Intent Spec Workflow

We've completely revamped the "Living Agent Loop". You can now go from a Journey Map friction point directly to a generated Intent Spec. The AI researches context, drafts the spec, and lets you refine it with natural language.

January 29, 2026

Features

Inline Friction Highlighting

Replaced list-based friction display with Google Docs-style inline highlights directly within research documents. Click a highlight to see the finding, severity, and synthesize intents.

Document Highlighter Component

New DocumentHighlighter with viewport-aware popover positioning that never gets cut off by the browser window.

Improvements

•Document View: Added read/edit toggle via eye/pencil icons
•Friction Pipeline: Replaced "Specs Shipped" stat with "Evidence Docs" count
•Interactions: Changed popovers to click-based triggering for better stability
•Navigation: Fixed "View All Friction Points" button behavior
•Persistence: Upgraded storage engine to support rich friction metadata (findings, tags)

January 26, 2026

Improvements

•New Journey Experience: Redesigned modal to prioritize "Start from Friction" workflow
•Journeys List: Refined list view layout to reduce visual noise
•Welcome Email: Updated messaging to align with "Intent Layer" philosophy
•Build Queue: Upgraded to Kanban board interface for better tracking

January 22, 2026

Features

Intent Graph

Introduced interactive graph visualization for intent dependencies and strategic alignment. Visualize how your intents connect and ensure they align with your workspace strategy.

January 18, 2026

Features

From Journey Maps to Intent Layer

We started building Pathmode as a journey mapping tool. The vision was simple: help teams visualize customer experiences and connect user research to real workflows.

But as AI transformed how teams build software, we noticed something. Journey maps weren't just documentation—they were becoming the source of truth that AI agents needed to ship the right features. The map wasn't the output. It was the input.

Today, Pathmode is the Intent Layer for modern product teams. We don't just visualize journeys—we synthesize structured Intent Specs from real user friction, then push them to the tools that build. Cursor, Linear, Jira. The journey map is now a coordinate system for your entire product.

Intent is the new code. Specs are the new handoff. The map is the new constitution.

January 11, 2026

Improvements

•Admin Notifications: Implemented email alerts for new sign-ups
•UI/UX: Redesigned Blog Index page with responsive grid and author metadata
•Architecture: Decomposed MainApp for Next.js App Router migration
•Docs: Updated competitive landscape analysis
•Fixed: Resolved modal navigation issues on public pages

January 8, 2026

Features

Anonymization Firewall

We've added a powerful privacy layer for research data. You can now enable the Anonymization Firewall in your workspace settings to automatically detect and redact PII (Personally Identifiable Information) from your interviews and documents.

This feature uses a zero-retention AI pipeline to replace names, emails, and phone numbers with generic placeholders before data is ever saved to our database, ensuring full compliance with strict privacy requirements.

Improvements

•Added PII detection toggle in workspace settings
•Updated data ingestion pipeline with pre-processing hook

January 1, 2026

Features

Interactive Live Demos

Explore fully-generated journey maps without signing up. We added interactive scenarios like "Autonomous Ride-Sharing" and "B2B SaaS Onboarding" to the landing page, demonstrating the full capability of AI-generated insights and specifications.

Improvements

•Smart Caching: Implemented intelligent local storage caching for public viewers, ensuring instant load times for AI artifacts like Storyboards and Proto Specs
•Optimized Public View: Revised staleness detection logic to prioritize cached content for read-only users, preventing unnecessary regeneration loops
•Robust AI Pipeline: Enhanced the generation engine with better payload management to reliably handle complex, data-heavy journey maps

December 29, 2025

Features

Integrations View & Live Embeds

Showcase your tool ecosystem with the new Integrations view. Embed live Figma prototypes directly into journey steps to keep designs synchronized.

Improvements

•Added "Integrations" to main navigation
•Added "Design Prototype" panel to Step properties for embedding support

December 28, 2025

Features

Specs View: Triage & Kanban

Redesigned the Feature Specs interface to split workflows into two distinct views:

Triage View: Review and approve new opportunities.
Kanban Board: Track specs from "Draft" to "Shipped".

Added an Archive workflow to hide deprecated or irrelevant specs.

Compact Research Dashboard

Refined the Research Dashboard for density. Key performance indicators are now grouped into a single row, providing an immediate overview of library health.

Hybrid Rendering Architecture

Re-architected the application core. Marketing pages now use Server-Side Rendering (SSR) for SEO and initial load performance, while the workspace remains a client-side SPA for interactivity.

Improvements

•Updated Research Dashboard visual hierarchy
•Implemented `useAppNavigation` for consistent routing state
•Resolved authentication race conditions with optimistic session hydration
•Fixed session persistence issues on refresh
• eliminated "Map Not Found" navigation errors
•Resolved client-side hydration mismatches
•Fixed React hook ordering stability issues
•Centralized modal state management

December 26, 2025

Features

Brand Identity Update

Refined the logo for better legibility at small sizes. The new mark retains the spiral geometry with cleaner lines and balanced weight.

Before

After

Improvements

•Updated favicon and app icons across the platform
•Unified spacing standards in navigation bars

December 25, 2025

Features

Feature Specs

Turn your journey maps into shipping software with the new Specs view. Automatically aggregate insights and journey steps into live feature specifications, manage acceptance criteria, and maintain full traceability back to your user research.

Journey Editor Refinements

We've redesigned the "Add Stage" experience for better usability. The button has moved from the bottom of the canvas to a dedicated vertical column at the far right of the timeline, styled with vertical text to match the grid structure. Plus, adding a new stage now automatically creates an initial "New Step" so you don't have to click twice to get started.

Improvements

•Privacy Overhaul: Removed Google Tag Manager and installed privacy-first GA4 with strict cookie consent
•Updated default workspace logo to use the Pathmode brand icon
•Various performance tunings and layout fixes in the editor canvas

December 24, 2025

Features

New "Product Teams" Landing Page

A dedicated page for high-velocity product teams. We've broken down the value of Pathmode for Engineers (Specs), Product Managers (Docs), Service Designers (Orchestration), and Product Designers (Research). Check it out →

Improvements

•Updated front page tagline: "Stop managing journeys. Start building them."
•Added new "For Product Teams" navigation link in footer

December 22, 2025

Features

Dynamic Hero Video

The landing page now features an immersive autoplay video in the hero section, replacing the static image. This provides an instant, dynamic preview of the Pathmode interface in action.

Improvements

•Optimized landing page hero performance with video assets

December 21, 2025

Features

New "Pathmode vs The Rest" Comparison

Completely revamped the comparison page. Instead of generic tables, we now have deep-dive breakdowns comparing Pathmode specifically against TheyDo (Management vs. Execution), Smaply (Service Design vs. Product), and Miro (Brainstorming vs. Structure). Compare tools →

Performance Tuning

Significant optimizations for the Journey Editor. We implemented intelligent debouncing for saves (1s) and throttled cursor tracking (100ms), making the editing experience buttery smooth even with active collaboration.

Improvements

•Fixed "zombie session" issue ensuring immediate and permanent sign-out
•Cleaned up security section iconography on landing page
•Updated site-wide navigation to new comparison breakdown

December 19, 2025

Features

Analysis Workbench

Introducing a powerful new way to connect research to your journeys. Open the research sidebar to view documents side-by-side with your map, and drag highlights directly onto steps to create insights, opportunities, or evidence instantly.

Vibrant Cartoon Storyboards

We've updated the storyboard generator to use a vibrant, flat cartoon style! Your customer journeys are now visualized with colorful, engaging illustrations that make storytelling even more fun and effective.

Improvements

•Updated AI prompt to request "vibrant cartoon style" for storyboard panels
•Enhanced image generation parameters for consistent flat design aesthetics

December 16, 2025

Features

Competitor Comparison Page

Added a detailed comparison page to help users evaluate Pathmode against other journey mapping tools. This transparency helps teams make informed decisions about their journey management toolkit.

Improvements

•Restored missing comparison page routing and functionality
•Fixed an issue where the /compare link was resulting in a blank page

December 14, 2025

Features

Redesigned Front Page with AI-Style Illustrations

Completely redesigned the landing page with minimalist, AI-style illustrations to create a clean and modern visual experience. The "Problem We Solve" section now features full-size pictograms for each problem area, making the challenges more visually engaging and easier to understand at a glance.

Unified Design Workflow Visual Updates

Enhanced the "Unified Design Workflow" section with step-by-step illustrations. Each of the three steps (User Research, Automated Journey Map, and Prototype Specification) now features a prominent illustration that visually communicates the workflow process.

Enhanced Trust & Security Page

Redesigned the Trust & Security page with a new hero section featuring a security illustration. The minimalist design creates a professional and trustworthy first impression, with the illustration prominently displayed above the "Security at Pathmode" title.

Dynamic CTA Section with Rotating Words

Added an animated rotating word effect to the CTA section headline. The phrase "The future of design is..." now cycles through "unified", "intelligent", and "seamless" every second, creating an engaging and dynamic user experience.

Improvements

•Added pictograms to "The Problem We Solve" section: Research Silos, Design Chasm, and Costly Drift
•Added workflow illustrations to all three steps in the "Unified Design Workflow" section
•Redesigned Trust Center hero with security illustration and improved visual hierarchy
•Improved visual consistency across landing page sections with flat, minimalist design
•Enhanced image presentation with full-size pictograms that fill their container elements
•Updated color scheme to use theme colors (subtle gradients) for better brand consistency
•Added smooth word rotation animation to CTA section without layout shifts

December 13, 2025

Features

Enhanced Landing Page with Visual Journey Flow

Added a visual journey flow diagram to the landing page to help customers better understand what Pathmode is about. The image showcases the complete workflow from customer insights through journey mapping to prototype generation, making the value proposition immediately clear to visitors.

Improvements

•Improved visual communication of Pathmode's core value proposition on the front page
•Streamlined image display to use a single, clear visual representation of the journey mapping process

December 12, 2025

Features

Improved Workspace Settings Access

Workspace settings are now more accessible with a direct link below the workspace name in the sidebar header. Settings now use workspace-specific URLs (e.g., /workspace-name/settings) for better organization and navigation consistency.

Improvements

•Moved workspace settings link from bottom navigation to header below workspace name
•Settings now use workspace URL key format for consistent routing
•Simplified "Back to Workspace" button to just "Back" in settings view

December 11, 2025

Features

Enhanced Storyboard Generator with High-Quality AI Images

The storyboard generator now uses Nano Banana (Gemini 2.5 Flash Image) to produce high-quality, photorealistic images for your journey visualizations. Images are now generated as detailed PNG files instead of simple SVG placeholders, bringing your customer journey stories to life with professional-quality visuals.

Improvements

•Fixed image generation to properly extract images from API response structure
•Improved image quality from SVG placeholders to high-resolution PNG images
•Added authentication check - unauthenticated users see placeholder images with sign-up prompts

December 10, 2025

Improvements

•Improved Google sign-in experience - you'll now see "Pathmode" when signing in with Google for a more professional and secure authentication process

December 8, 2025

Features

Workspace Settings Navigation

Moved Workspace Settings to the main navigation panel for easier access. Settings now use workspace-specific URLs (e.g., /workspace-name/settings) for better organization and navigation consistency.

Improvements

•Fixed Activity feed navigation link
•Improved workspace invitation permissions - workspace owners can now create and manage invitations
•Streamlined Users & Team view by removing redundant Settings button
•Updated RLS policies to support workspace owner role for invitation management

December 6, 2025

Features

Trust & Security Center

Comprehensive security and privacy information page showcasing our security measures and compliance standards.

Team Management & Activity Feed

Complete overhaul of team management with email invitations, granular roles (Owner, Admin, Editor, Commenter, Viewer), and a new activity feed to track workspace actions.

Cookie Consent Management

GDPR-compliant cookie consent banner with granular controls for optional cookies and analytics.

Improvements

•Improved footer navigation across all pages
•Enhanced landing page with security information

November 27, 2025

Features

Beta Launch

Pathmode is now in public beta! We're excited to share our AI-powered journey mapping platform with designers and product teams.

API Rate Limiting

Implemented distributed rate limiting with Upstash Redis to prevent API abuse and ensure fair usage across all endpoints.

GDPR Compliance

Added comprehensive GDPR compliance features including cookie consent banner, privacy policy, terms of service, and self-service account deletion.

Profile Page

New profile page accessible from the sidebar where users can manage their account settings and delete their account.

Performance Monitoring

Integrated Vercel Speed Insights to monitor application performance and identify optimization opportunities.

Improvements

•Enhanced security headers and Content Security Policy
•Improved Trust & Security page with detailed compliance information
•Added links to legal pages throughout the application
•Launched changelog to keep users informed of updates

Want to stay updated? Visit our Trust & Security page for the latest on security and compliance.

Changelog

Updates and improvements to Pathmode. We ship regularly.

October 18, 2026

Changed

•An implementation review passes only when the reviewer marks every outcome the intent states as met. The bar used to be a majority of outcomes. On the labelled examples we check the grader against, the two bars gave the same verdict every time: incomplete changes met half or fewer of their outcomes, and correct changes met all of them. The stricter rule closes one way a partial change could pass. It does not make a green check proof that the behaviour works: on those same examples the reviewer still credited outcomes whose deciding code was outside the diff. Read a pass as “nothing shown contradicts the intent”, and review what the diff does not show yourself.

Fixed

•Your agent now sees all of Pathmode’s instructions. Claude Code cuts MCP server instructions at 2,048 characters, and the connected instructions were about four times that, so the rules it read ended before preflight-before-code for an existing intent.md, the adopt flow, and what to do when the server rejects a key. Both the connected and the keyless instructions now fit under that limit with those rules intact (@pathmode/mcp-server 1.43.2, plugin 0.1.53).
•When a product choice is still open, your agent’s readiness check leads with the not-ready verdict and the choice to resolve, even when all six field checks pass. The field checks used to appear first, and a waived check could read as ready while a decision was outstanding (@pathmode/mcp-server 1.43.2).
•After your agent confirms an intent field, the reply tells it to run the readiness check again before further edits or a summary. The confirmation used to read as the end of the step, so an agent could report ready from the confirmation alone (@pathmode/mcp-server 1.43.2).
•The daily agent-activity email is held when the workspace’s queue address cannot be looked up, and is retried once it can. It used to be sent with a link to the wrong page.

October 17, 2026

Features

Review pull requests from your own GitHub workflow

Removed

•New workspaces no longer start with a "Connect your coding agent" intent, and existing ones have been removed. It sat in your intent list beside your own work and described a setup flow that has since changed. Connect a repository from Home or from the start screen after sign-up instead.

Improvements

•We now check implementation reviews against a fixed set of examples labelled before the grader sees them. The examples cover incomplete changes, misleading citations, and rules that do not apply to a change. We compare proposed grader changes with the current behavior, track false passes and false blocks separately, and reserve unseen examples for a final check. Recorded responses can be replayed to test the review rules without another model call.

October 14, 2026

Features

Try a review from both sides

October 13, 2026

Features

Your agent works from the version you approved

See what Pathmode handed your agent

Fixed

•Your agent’s readiness check now applies the app’s approval rule exactly. A repository intent needs its exact version approved, whoever wrote it, and an unfinished intake review blocks building.
•When your agent saves a new version while the review page is open, the page moves to that version so you can review and approve it. It used to stay on the earlier version behind an error that Try again could not clear.
•A workspace invitation now names the address it was sent to before you choose how to sign in. If you sign in with a different address, Sign Out and Use Correct Email takes you to account entry with the invited address filled in, instead of a password sign-in that could not create an account. To join with the address you used instead, the screen asks you to have the invitation sent to that address.

Improvements

•While a version waits for your approval, your agent starts listening for it and ends its turn instead of checking back on a timer. The commands your agent is told to run match the server version it uses, so a pinned plugin setup is not switched to the latest release.
•A connected save leaves intent.md matching what the workspace stored, and says when the workspace kept something differently. For example, a verification sent without a checks list keeps its stored checks.

October 12, 2026

Features

Review a proposal without knowing how the software works

Illustrative example (33 seconds): four product decisions hiding in a cancellation request, reviewed before the agent builds. Watch on YouTube

Ask for changes in your own words

October 11, 2026

Fixed

•Connecting a repository intent to your workspace dropped any code block from its Implementation Context. Code examples in that section now reach the hosted intent intact; they still do not count as outcomes, checks or other requirements. Agents proposing a correction to an outcome are now told to send the complete outcome, so their first proposal is accepted instead of failing on a bare sentence (@pathmode/mcp-server 1.41.3, plugin 0.1.50).

October 10, 2026

Features

Say whether evidence supports or challenges a claim

Read the original passage beside the finding

Fixed

•Security updates to Next.js and to the libraries bundled into the MCP server, including the MCP SDK (1.31.0). None of the advisories applied to how Pathmode uses them, so no action is needed on your side; updating removes them from your security scanners (@pathmode/mcp-server 1.41.2, plugin 0.1.49).

October 9, 2026

Features

Your agent builds when you authorize, if you asked it to

Your agent listens for your review answers

Withdraw an answer, answer again, or ask your agent to retry

Fixed

•In a connected workspace, your agent builds only from a spec that passes the execution checks. The implementation prompt still offered a way to proceed after you accepted a failing check, and that path led to a prompt telling your agent not to implement. Accepting a named risk still works in keyless mode (@pathmode/mcp-server 1.41.1, plugin 0.1.48).
•When a repository has no intent.md, the grill, verify, split and handoff skills ask which saved intent you mean instead of picking the workspace’s most recent one, so their edits and notes land on the intent you chose. The workflow interview no longer starts by connecting the repository (plugin 0.1.48).
•The compile-intent conversation checks its draft against the six preflight gates, records behavior you left undecided as open product choices, and saves a new spec as a local draft until you adopt it. A failing check is shown but never blocks the save. Its checklist used to accept a deferred verification and out-of-scope lines in place of a constraint, both of which preflight rejects (@pathmode/mcp-server 1.41.1).
•Adopting a second repository into the same workspace against a local or staging server no longer re-points the other repositories on that machine. The one credential file per workspace used to be overwritten, so the first repository’s agent traffic went to the new server with the new key, and nothing warned. Each repository now holds the credential for the API origin it was adopted at, installs from earlier versions keep working, and a repository binding that names an origin with no stored credential connects nowhere rather than borrowing another one (@pathmode/mcp-server 1.41.0).
•Connecting a second repository on a free plan reuses the key your first repository holds. Adopt presented only a configured API key, never the key an earlier adoption stored, so the approval offered to replace the workspace’s only key, and collecting it deleted the key the first repository used. Adopt now presents keys stored on this machine for the same server, never for another one, and a browser-started adoption code tries each instead of stopping at the key limit (@pathmode/mcp-server 1.41.0).
•A message too long for an intent’s discussion is no longer saved as the assistant’s reply. Pasting a long document used to add "Invalid intent interview request" to the conversation, and a short follow-up could be refused the same way. The message box now keeps your text and says how much to cut: one message holds up to 20,000 characters. Nothing is added to the conversation, and a message refused for any other reason comes back to the box too.
•When import redaction is enabled, a research file whose redaction fails is skipped with a retry message. It used to be saved with its original content. Privacy settings now explain which imports redaction covers and which content it leaves unchanged.
•Workspace privacy and integration controls are read-only for people who are not owners or admins. Connection details remain visible, and the settings explain who can change them.
•You can save the signing secret from your Linear webhook in workspace settings, including on an existing connection without entering its API key again. The connection used to generate an unrelated secret, leaving Linear’s signed updates unable to verify.
•MCP setup instructions only fill in a newly created key when it has full access. A key limited to evidence ingestion no longer appears in a setup command that needs broader access.

Improvements

•While an answer is with your agent, the review page checks for the revised proposal and loads it without a reload. If you have unsaved changes, it waits until they are saved, and it tries again if loading fails.
•Your agent gets a clearer account of the Pathmode tools. The execution prompt says when it refuses and what it carries, every product-choice field is described, and recording a finding points to proposing the exact correction. Pressure wording in the prompts and skills is replaced with plain statements (@pathmode/mcp-server 1.41.1, plugin 0.1.48).

October 8, 2026

Fixed

•An answer you sent to your agent no longer reads "Answer requested", which looked like a request for the same answer again. The row says "Answer sent" and whether your agent has read it, a proposal that depends on an answered choice says it waits for your agent, and the header tells you to go back to your agent and ask it to apply your answers.
•Your agent can apply your answer to a product choice when the recommendation adds a requirement the intent already has. The apply used to be refused, as if intent.md had moved after you answered. The existing requirement is kept as it is, and a refusal for any other reason no longer says intent.md moved.

Improvements

•Your own decisions stand out in the product choices: the "Your decision" label is marked as complete, and the decision itself reads darker than the page’s own labels.

October 7, 2026

Fixed

•Workspace home no longer asks your role every time you return to it. An unanswered setup question used to come back on each visit; one person saw the role question 13 times. It now waits a day before asking again, and answering or skipping it still settles it for good.
•If you said after signup whether you will run the coding agent yourself, workspace home no longer asks again after your role. The second answer used to overwrite the first.

October 2, 2026

Features

Delete an intent from its page

Changed

•Your agent can record an answer to an open question that has no recommendation, such as one from the workspace assisted intake. Cost and reopen trigger are optional for those answers. adopt refuses an intent that still holds assisted questions or answers, and says to keep it local or use the workspace intake (@pathmode/mcp-server 1.39.0).

Fixed

•After you approve a connection, the waiting page says when your agent has stopped checking for it, and tells you to go back and ask it to continue before the approval expires. It used to promise that the page would move on by itself.
•Rerunning adopt after its approval request expired starts a new request straight away. It no longer prints the expired code first, or reports an expired request as denied in the browser (@pathmode/mcp-server 1.39.0).
•intent.md saves preserve confirmations in files with Windows line endings, and repository roots supplied as Windows file URLs resolve to the correct directory (@pathmode/mcp-server 1.39.0).
•The email telling you your agent added an intent links to that intent. It used to open the workspace and leave you to find it; the review queue now has a link of its own.

October 1, 2026

Features

Start an intent with a brief, then edit the draft

Choose between the ways your brief says it is done today

Ask a colleague to propose an answer

Prepare for implementation on the draft page

Fixed

•Evidence counts and searches continue beyond 1,000 records instead of silently stopping at the first page.
•Queued intent autosaves run in order, preventing two saves from competing against the same revision. Verification checks retain unique IDs when older and newer check formats coexist.
•Undoing a tracker import preserves evidence linked to an intent, including a link added during the undo. Repeated URLs in one import no longer create duplicate evidence items.
•Failed Linear, Jira, and Slack webhook processing remains retryable instead of being treated as already completed. Retried shipment updates do not duplicate the shipped note.
•Account deletion reports cleanup failures instead of claiming success. Content belonging to a surviving workspace is preserved.
•Unreadable product-choice records keep readiness blocked instead of disappearing from the decision count.

Improvements

•The original brief shows which sentences have no current source selection. Clear automatic roles removes untouched automatic additions while preserving manual edits and separately marked requirements.
•Brief titles use a complete sentence or a short generated title instead of cutting the opening sentence in half. A generated title never replaces one you have edited.

September 28, 2026

Changed

•Preflight grades intents written in Finnish. The objective, outcomes, title and constraint checks read Finnish wording as strictly as its English translation, so a Finnish spec no longer fails on its language alone. Other languages are still read but left unconfirmed (@pathmode/mcp-server 1.38.0, @pathmode/cli 2.7.0, plugin 0.1.45).
•Preflight counts a refusal as a measurable outcome when it says which case is refused or what the refusal names: "An import with a duplicate ID is refused and names the existing row" now passes. A bare refusal such as "Bad requests are rejected" does not. Blocked and prevented follow the same rule, so "Bad requests are blocked" no longer passes either (@pathmode/mcp-server 1.38.0, @pathmode/cli 2.7.0, plugin 0.1.45).

September 27, 2026

Features

Your agent starts the connection, you approve it

Changed

•After you save an intent.md without a workspace, your agent points team review to adopt instead of the website and a manual API key. setup, the compile-intent prompt, and the Claude Code plugin say the same, and the plugin’s optional key can stay empty. An existing pm_live_ key still works for workspace access without a local intent (@pathmode/mcp-server 1.36.1, plugin 0.1.43).
•Preflight reads a `## Why` section as the objective when there is no Objective or Problem section. A `## What` section is not read as outcomes, because it usually describes the solution. When outcomes are missing, the verdict says the What section was not evaluated and that success criteria belong under Outcomes (@pathmode/mcp-server 1.37.0, @pathmode/cli 2.6.0, plugin 0.1.44).

Fixed

•Workspace editors can finish connecting a repository. The last step used to refuse them after the browser had generated their command.
•If a connection response is lost, or a temporary server error interrupts adoption, rerunning adopt resumes the same connection. It keeps the saved credential and avoids creating another key or uploading the same spec twice (@pathmode/mcp-server 1.36.0).
•setup stores your API key in a file only you can read, and checking the key no longer counts as using it. It refuses anything that is not a pm_live_ key before writing any config, including the settings placeholder, which used to report success (@pathmode/mcp-server 1.36.0).
•Saving intent.md no longer deletes a Problem section written beside Objective, or a Proposed outcome beside Outcomes. Preflight reads only Objective or Outcomes when both are there, and the save used to treat the other section as spec and drop it without saying so. It is now kept, and the save message lists it. If the save leaves that field empty, the kept section would be read in its place, so the save removes it and names it (@pathmode/mcp-server 1.37.0, plugin 0.1.44).
•Saving intent.md also keeps a prose Proposed outcome section when the file lists its outcomes in frontmatter. A list under that heading replaces the frontmatter outcomes; prose does not, so the prose is now kept as your text instead of being deleted (@pathmode/mcp-server 1.37.0, plugin 0.1.44).
•The web preflight now treats an empty Objective, Problem or Outcomes heading the way the CLI does: the field stays empty instead of being filled from a Background or Summary section, so the same file gets the same verdict in both.
•Preflight no longer calls an objective too vague when it found none. An intent.md with no objective section now gets "No objective found", and the vague message is kept for text it actually read (@pathmode/mcp-server 1.37.0, @pathmode/cli 2.6.0, plugin 0.1.44).

September 26, 2026

Features

Every check passed. Wrong duck.

An illustrated example: the same question, answered before the build or after.

Answer product choices without a workspace

A warning before an answer can fail the outcomes check

Changed

•The Ask an AI section no longer calls its answer independent. The prompt is ours, and the answer draws largely on what we wrote. It stays on the comparison pages and is gone from pricing.

Fixed

•Changing a choice from open to accepted in intent.md, with nothing else, no longer passes preflight. Without a workspace, an accepted choice counts only when the spec includes what it promised. This keeps the file consistent. It does not verify who decided (@pathmode/mcp-server 1.35.0, @pathmode/cli 2.5.0).
•The preflight.pathmode.io paste box reads product choices, scope, and decisions. It used to call a spec with four open choices ready.
•Decisions whose ruled-out text contains a closing parenthesis are no longer dropped when a local save rewrites intent.md (@pathmode/mcp-server 1.35.0).
•Two accepted choices that promise the same outcome, edge case, or constraint add it to the spec once. It used to appear twice (@pathmode/mcp-server 1.35.2).
•Without a workspace, the verdict no longer sends you to a proposal review you don’t have. It says a person answers or excludes each open choice before implementing. After a save, the note on connecting a workspace names what it adds: private evidence, a reviewer who answers and authorizes, and staleness across repositories (@pathmode/mcp-server 1.35.2, @pathmode/cli 2.5.2).
•Answering a choice with its own recommendation, to reword an outcome the warning flagged, keeps the choice’s alternative. It used to replace the alternative with the answer, so the decision read as ruling out the option it chose. Same fix in the workspace review (@pathmode/mcp-server 1.35.3).
•Intent.md writes from your agent run one at a time, so two answers sent together are both recorded. The second used to overwrite the first while both reported success (@pathmode/mcp-server 1.35.4).
•Without a workspace, your agent reads intent.md from the repository your editor has open. It used to read the folder the MCP server was started in, so after switching checkouts it kept serving the old spec until the server was restarted (@pathmode/mcp-server 1.35.5).

Improvements

•The pricing page starts with a choice between the keyless repository workflow and the Free workspace, and the first price shows without scrolling.
•intentspec.org/intent-md checks an intent.md written in the format of Anthropic’s AI-native SDLC playbook: the verdict it gets as written, and the same file after three repairs. pathmode.io/intent-md redirects there.
•intentspec.org documents the intent/<feature>/intent.md layout beside a root intent.md, and which one preflight reads.
•The CLI and MCP READMEs open with a recorded first run you can reproduce with npx -y @pathmode/cli preflight.

September 25, 2026

Features

Green CI does not validate a test the same change edited

Check the build against the decisions you made

Preflight finds intent.md in an intent/ folder

Mark a choice that doesn’t need a product decision

Changed

•Green CI requires a complete changed-file list compared against the PR base commit IDs captured in its event. Runs without those IDs, including runs before a PR exists, leave checks held. A failed run still records failing checks.
•The site navigation shows Log in and Sign up. Installation stays on the homepage and the Developers page.
•Headings inside a field, such as a ## heading in implementation context, are written one level below the field’s own heading, so they stay part of it (@pathmode/mcp-server 1.34.0).

Removed

•The full intent editor, including Open advanced editor. Intents open in the proposal view, and conversation continues in its discussion drawer.

Fixed

•Changed claims no longer display evidence backing from an earlier revision while new records load or after a failed fetch.
•An unrelated workspace folder no longer blocks a save to the repository being reviewed (@pathmode/mcp-server 1.34.0).
•Undo in a brief no longer drops its decisions.
•Onboarding recognizes when you are already signed in.
•Invitation and membership changes complete as one step, so simultaneous requests can’t leave a workspace half-updated.
•Saving a hand-written intent.md keeps the text under its title and every section Pathmode doesn’t read, such as Open questions. A save used to delete them (@pathmode/mcp-server 1.34.0).
•A save writes only to the repository preflight checked, and never outside it, including through a symbolic link (@pathmode/mcp-server 1.34.0).

Improvements

•Returning to a revised proposal starts with "What changed since your last decision", above the remaining choices.
•Repository proposals wait briefly for review records before appearing, and same-content refreshes keep existing records on screen.
•Cursor setup starts with a keyless configuration and Preflight. Connecting a workspace is a separate step.
•Edit a claim in place on the proposal.
•When verification blocks handoff, "Add a verification check" saves a repeatable check to run. Saving does not record a pass.
•Blockers other than open choices are listed under the choices as "Also blocking handoff".
•"Answer the open choice" brings that choice to the top of the page and marks it for a moment.
•Invitation sign-up shows the invited email address and workspace.
•check_intent_readiness takes repoRoot, so agents in clients without workspace roots, such as Cursor and Codex, can name their repository (@pathmode/mcp-server 1.32.0).
•Preflight reads specs with "## Problem" and "## Proposed outcome" headings, with or without frontmatter (@pathmode/mcp-server 1.32.1, @pathmode/cli 2.3.1).

September 23, 2026

Features

Confirm or reject one claim in an agent’s proposal

Discuss a product choice before you answer it

Invite a new reviewer to one feature

The Share dialog now has Invite a new reviewer. Enter an email to create an editor invitation for that feature. Once they accept, the invitee lands on the feature they were asked to review.

Changed

•The Intents board is now a searchable list.
•New workspaces start from a brief.
•Outcomes can be recorded only after an intent ships.

Removed

•The Relationships view. Agents can still read relationships with analyze_intent_graph.

Fixed

•Shipped or authorized intents no longer ask for authorization again.
•Proposals with product choices now list their outcomes.
•Opening repository setup no longer discards unfinished attempts elsewhere.
•A brief that fails to draft is kept for manual editing.
•Generated briefs with empty optional fields now save.
•Your own brief is no longer marked as an agent suggestion.
•Keeping a single suggestion no longer fails to save.
•People who stay in onboarding now get the welcome email.
•check_intent_readiness uses the intent you name when several folders contain an intent.md (@pathmode/mcp-server 1.31.2).

Improvements

•The intent page shows one primary action, such as "Answer the open choice" or "Authorize as is".
•Each action has one name. "Request changes" always goes to the repository agent.
•The intent page states each fact once.
•The selected product choice opens inside its own row.
•Workspace home shows intents with unanswered product choices.
•Unfinished briefs are kept in your browser and can be resumed from product home.
•Repository setup is reachable from Settings, product home, and the product list.
•Evidence items show their source, and "Add evidence" opens in place.
•Intent rows show titles on up to two lines.
•The developer docs start with the review workflow.
•The About page lists key company facts.

September 20, 2026

Features

Answer product choices before authorizing the build

Recorded in the Pathmode app, in an example workspace. Fieldnote is a fictional product.

Follow a decision from the request to the next coding session

Fixed

•Settled product choices no longer trigger a competing generated review question.
•Prerequisite conflicts explain which decision needs attention instead of reporting a generic save failure.
•Applied-choice summaries show the recorded answer instead of incorrectly reporting that product choices were removed.

Improvements

•Repository reviews lead with the recorded problem and a numbered choice list showing dependencies, pending answers, and resolved decisions.
•Advanced editing is available from More options, keeping proposal review as the default page.

September 13, 2026

Features

Review what the agent understood before it builds

Keep the original request beside the proposal

Share one feature with the teammate who decides on it

Invite the product manager as an editor

Fixed

•If an agent cannot apply your correction, the review now shows why and who acts next. It tells you whether the agent can retry or you need to request the correction again against the current revision
•Failed review analyses now say they could not be completed, instead of suggesting the proposal matches your request
•A failed relationship rerun no longer shows its error beside "No new relationships found"
•Invited teammates can create an account and return to their invitation without being asked to create a separate workspace
•Linking evidence to a claim during review no longer blocks your coding agent from applying a correction
•Requests attached by an agent now appear reliably in the proposal review. Retrying the same attachment does not create a duplicate
•Failed evidence links now report an error instead of reporting success
•Incomplete AI findings are rejected before they can be saved

Improvements

•Reviewers share one question and recommendation per revision, so repeat visits use no additional AI allowance. Re-importing the ticket clears the outdated question
•Correction fields start with the current wording. Choose "Use this wording" to adopt Pathmode's suggestion before sending
•Review questions and recommendations use the language of your request. Suggested corrections use the language of the text they replace
•Workspace home lists each intent by title, with its revision status underneath, making pending proposals easier to tell apart
•When a discussion suggests a change that must go through your coding agent, you can open a correction with that wording directly from the notice

September 11, 2026

Features

Sign up with an email link

Create an account by clicking a link sent to your email, with no password to choose. Google signup is also available. Existing accounts can still sign in with their password.

Name your workspace during signup

Choose your workspace name and URL as you sign up. The URL follows the name until you edit it yourself, and you can see when a URL is already taken.

See when your coding agent is connected

New workspaces start with a Connect your coding agent intent. It completes when an agent reads it using your workspace key. If setup needs to wait, the step stays open so you can return to it later.

Follow your teammate’s setup progress

Removed

•New workspaces no longer include the fictional Acme HR product. Existing workspaces keep their sample content, and demos still include it

Fixed

•Email signup links now open an existing Google account for that address instead of creating a duplicate
•You can go back to correct your email address and request a link immediately for the corrected address
•Invalid workspace names return you to the form with an explanation and your input preserved
•Confirming signup in two tabs now opens the same workspace in both
•The email confirmation screen now describes the signup path you chose

Improvements

•The welcome email arrives after you confirm your address and explains how to get started in your workspace
•The signup page keeps pricing visible and simplifies the steps before account creation

September 9, 2026

Features

Connect a repository as you get started

Invite a teammate to connect the agent

Send an invitation that takes your teammate directly to repository setup. You can start writing while they connect, and follow the connection status from your own screen.

Try the readiness checks before signing up

Run six deterministic readiness checks in your browser on an example spec or one you paste in. Nothing is uploaded and no account is needed.

Fixed

•Signup works when the preceding decision did not name a product
•The Pathmode logo on the preflight page now returns to the homepage

Improvements

•Questions deliberately left open are distinguished from questions you have not reviewed, both in the review and the exported spec
•Pre-signup decision reviews let you accept, edit, or leave each suggestion open individually
•Continue from a product decision to signup and proposal review without a separate workspace step
•Briefs record where in the product they were started

September 8, 2026

Features

Review a proposal from your agent

Review suggested additions

Authorize the revision your agent will use

Improvements

•Workspace and Product Home prioritize proposals needing a decision, suggestions to review, and work waiting on an agent
•Intent rows distinguish proposal review, suggestion review, revised proposals, and waiting for an agent from delivery status
•The conversation editor remains available for discussion and detailed editing
•Agent-written outcomes and edge cases retain their suggestion and evidence references through ordinary saves

September 3, 2026

Features

See what backs each claim

Keep evidence and assumptions distinct in handoffs

Review pull requests against the intent before merge

Keep private evidence out of committed files

intent.md, CLAUDE.md, AGENTS.md, and .cursorrules now reference evidence by type and ID. Quotes and source names stay in Pathmode, where connected agents can retrieve them through MCP.

Removed

•The Evidence board’s Kanban, Flow, and Clusters layouts have been removed. The Inbox and evidence detail panel remain
•Evidence Browse is now a flat, searchable list. Grouping, bulk selection, and Suggest intents have been removed; review claim backing on the intent itself
•New evidence links in the signed-in UI must identify the claim they support. Existing general links remain visible, and API and MCP linking behavior is unchanged

Improvements

•Evidence records distinguish human-reviewed items, imports awaiting review, and agent-created signals
•Sample tickets and example cards are labeled as unverified assumptions
•Paste text into the composer to extract and review signals inline; use the import dialog for files

August 31, 2026

Features

Connect an existing intent.md to Pathmode

Connect repositories to separate workspaces

Improvements

•If repository adoption fails during a network step, you can resume without creating duplicate credentials
•Adoption codes are single-use, expire after 15 minutes, and have rate-limited exchange attempts

August 30, 2026

Features

Use Pathmode with OpenSpec

Contribute to the IntentSpec standard

Read IntentSpec consistently across tools

Use the repository’s intent for local checks

Pass handoff notes to the next agent

See intent status when Claude Code starts

Fixed

•The Engineering Lead and First Intent use case now opens the flow that works without an API key
•Plugin installation instructions are consistent across the homepage, preflight, intent compiler, and developer docs
•Page titles and blog metadata now describe their current content

August 29, 2026

Features

Import and export OpenSpec changes

Confirm readiness dimensions in your workspace

August 28, 2026

Fixed

•Workspace readiness checks now read edge cases correctly, fixing a false failure for specs that already defined them. Current-intent reads and outcome rubric exports also include edge cases. Fixed in MCP server 1.14.1

August 26, 2026

Features

Bring code context from your agent into the spec

Link a pull request to verify the merged work

Save agent-created specs to your workspace

Fixed

•Merged pull requests now update intents when a product has its own GitHub connection
•A merge only updates intents associated with that repository

Improvements

•Concurrent edits to a spec report a version conflict instead of silently overwriting each other
•The new record_implementation_context MCP tool lets agents submit code context

August 23, 2026

Features

Try Preflight in your browser

Authorize agent proposals before implementation

Import a Linear or Jira backlog as evidence

Verify work against the pull request diff

Continue from a finished spec to agent handoff

After you accept Finalize, the assistant offers the next step for your workspace: connect an agent, complete an unused setup, or hand over the spec. Shipped work receives no setup prompt.

Describe how the product works today

Removed

•Creating GitHub Issues from intents has been removed. GitHub repository analysis and merge verification remain available, as do issue creation in Linear and Jira

Improvements

•The setup checklist now includes connecting a repository for codebase analysis
•Pasted customer quotes and measurements retain their types and source references

August 22, 2026

Features

Find agent-created specs awaiting your review

Measure outcomes after shipping

Record decisions directly in the spec

Tailor setup and interviews to your role

Install the Claude Code plugin

Fixed

•The shipped-status gate now recognizes verification checks created in the current editor
•Unsubscribe links now require a valid signed token
•Outcome measurements require a workspace writer role; viewers retain read access

Improvements

•Connector imports now identify their integration source
•The product setup checklist detects team and agent connections and closes when complete
•Developer docs now include decisions and implementation-findings endpoints

August 21, 2026

Fixed

•Installing without an API key now automatically starts local mode and exposes the tools that work with intent.md
•Saving the same intent updates it; replacing a different intent requires an explicit overwrite
•Reading intent.md now preserves multiline sections, scope, and decisions
•The setup command now completes without an API key

June 21, 2026

Features

Draft your first intent before signup

Describe what you are building on the landing page and review a proposed objective and outcomes. Assumptions remain labeled separately from evidence. Signing up saves the draft you reviewed.

Build product context from a document

Bring implementation findings back into the spec

Update repository instructions with sync_context

The MCP server can update the Pathmode section of CLAUDE.md or AGENTS.md from your intent. Repeated runs update the existing section without adding duplicates or changing unrelated content.

Fixed

•Archived evidence is excluded from agent prompts and exports

Improvements

•The landing-page draft asks a focused follow-up question to clarify a vague description
•Product Home displays the strength of your product context
•Pasting a substantial document during signup handoff starts extraction without asking you to paste it again
•Pre-signup AI requests now use bot checks and rate limits

June 18, 2026

Features

Define scope and verification in IntentSpec v1.2

June 15, 2026

Features

Organize evidence by its properties

Group evidence by type, severity, source, or status, and switch between list and board layouts. Edit a signal’s text and properties in its detail panel, and link or unlink intents there.

Import evidence from a remote MCP server

Add a remote MCP server through Add Evidence → MCP source to extract typed signals from its resources. This first version supports manual sync of resources, and imported items show a Synced label.

See which decisions need your attention

Edit a scanned workspace before signup

Scan a URL, then edit the draft product, evidence, and header directly in the preview. Sign in from the same page to save the draft you reviewed.

Let teammates join through their company email

Ask an AI about Pathmode

Links on the homepage, pricing, comparison, and product-team pages open Claude, ChatGPT, or Perplexity with a question ready to send. Competitor pages include a comparison question.

New essay: The Cost of Being Worth Using

Explore why faster software delivery still needs product judgment to create something people use. Read the essay.

Changed

•The role-selection dialog no longer blocks the first workspace screen. Your role remains editable in Profile
•Legacy /journey-map URLs redirect to /product-teams

Fixed

•Atlassian and Jira integrations now verify installation bindings before resolving intents or assigning workspaces, preventing access across workspace boundaries
•The Anonymization Firewall holds data back if redaction cannot complete
•Google signup analytics now record correctly, and signup forms prevent duplicate submissions
•Evidence suggestions no longer trigger an infinite render loop
•A failed workspace load no longer sends you to workspace creation

Improvements

•Settings and the Intents board use simpler controls and quieter styling
•Creating a first product without a scan now opens its intent editor
•New accounts derive a display name from the signup email
•Account deletion now records an analytics event and notifies administrators

June 12, 2026

Features

Capture and review evidence inline

Type a signal in the Evidence composer for automatic classification, or use / to choose its type. Paste a support ticket to extract multiple signals, then keep or discard each one inline.

Edit evidence in place

Hover an evidence row to edit its text, type, and severity without opening another page.

Improvements

•Empty products show sample signals you can adopt to get started
•The evidence composer is available in List, Clusters, and Flow views
•Single-line observations are automatically classified like longer pastes

June 8, 2026

Features

Explore advanced Playbook guides

Three new guides cover product constitution rules, verification feedback loops, and coding-agent setup. The Playbook now separates introductory and advanced learning paths.

Read developer docs generated from the API

The developer reference now draws its tool lists, arguments, and OpenAPI schema from REST API and MCP server definitions. Automated checks help keep the documentation aligned with shipped behavior.

Fixed

•Health Metrics now appear in Markdown agent prompts as well as Claude Code and Cursor exports
•Broken Get started links in Playbook guides now open onboarding

Improvements

•Playbook guides include previous, next, and related links with clearer difficulty levels
•Playbook, glossary, and comparison pages now explain Scope and Health Metrics alongside the other IntentSpec fields

June 6, 2026

Features

Explain and enforce workspace rules

See exactly what blocks readiness

Install agent skills and export an outcome rubric

Fixed

•Intent conversations offer a retry when the AI fails
•Saves no longer race between competing copies of the spec
•Fast conversations no longer create duplicate specs, and autosave keeps each intent’s baseline separate
•AI streaming preserves updates split across response chunks
•Scope entered in conversation now appears correctly in the spec
•Returning to a conversation phase no longer repeats the greeting
•Demo requests submitted from the site are now saved correctly

Improvements

•The Spec panel places the title and key properties above the document body
•Constitution templates include enforcement levels and reasons for their rules
•Site scans have stronger URL and network checks, more write operations are rate-limited, and integration webhooks reject replay attempts
•Security updates cover dependencies and database permissions

May 27, 2026

Features

Include evidence in IntentSpec v1.1

Compare Pathmode with other tool categories

New comparison pages explain how Pathmode relates to issue trackers and spec-as-contract tools. Individual vendor comparisons link to their category, and Linear has a new comparison page.

Fixed

•Returning users reopen their existing workspace instead of creating a duplicate
•Site scans no longer reject visitors solely because a bot-check script failed to load; rate limits remain in place
•Reopening a spec no longer records past conversation turns as new activity, and autosave failures are tracked

Improvements

•You can skip the site scan or continue without it after a scan fails. New workspaces still include sample content
•Product pages now explain how evidence, decisions, and code context support work before and after shipping

May 25, 2026

Features

Ground a spec in your existing codebase

Include code context in agent handoffs

Claude Code, Cursor, and MCP agent prompts now include the intent’s Implementation Context: relevant files, existing behavior, risks, and verification suggestions.

Choose a repository for each product

A product can use a different GitHub repository from the workspace default. Set it in product settings under Code binding.

May 21, 2026

Features

Start a workspace from your website

Paste a public URL to draft a product, extract signals from your site, and get questions for your first intent. Review and edit the preview before signing up to save it.

Choose which product to scan

For sites with multiple products, choose the one you work on before Pathmode drafts its context. The scan then focuses on that product.

Describe your product when a scan cannot read it

If a site requires login or cannot be scanned, paste a description to draft the context. The original URL stays attached as a source reference.

Start your first intent from the product page

Scanned workspaces show a Draft your first Intent prompt until you create one.

Changed

•Public signup buttons and bookmarked signup links now open the site-scan flow
•The sign-in dialog links new users to the same signup flow
•Signing in without a workspace now opens site-scan onboarding

Removed

•The previous first-run welcome panel has been replaced by site-scan onboarding

Improvements

•The setup checklist reflects completed site scans and the next step of drafting an intent
•A scanned description satisfies basic product setup; vision and north star can be added later

May 19, 2026

Features

Gemini 3.5 Flash for global workspaces

Global workspaces now use Gemini 3.5 Flash for AI features, including intent conversations. EU-residency workspaces continue to use Gemini 2.5 Flash.

May 11, 2026

Features

Pathmode for Jira — Atlassian Marketplace

Push to Jira → linked from inside Jira

Bidirectional status sync

Atlassian Forge App workspace binding

Improvements

•Privacy policy gained a dedicated section covering Atlassian integration data flows — what we receive (cloud id, site URL, issue keys, account IDs), what we send (intent summary + description on push), how we authenticate, where it is stored, retention, and uninstall behavior
•New /support page with Pathmode for Jira–specific troubleshooting and security reporting contact
•Jira Test Connection now works after Save — server-side test against stored credentials instead of requiring the user to re-type the API token
•JiraIssueModal portals to document.body so it renders above the conversation panel in the Intent Editor

May 10, 2026

Features

AI Spec Review

Codex Integration Page

New Codex guide mirrors the Claude Code agent page. Both now share a single component and registry, so adding another coding agent is a data change, not a copy.

Use Cases Hub

Auto-Linked Glossary

Blog, playbook, and glossary content now auto-link the first occurrence of each glossary term. New entries propagate to old content automatically — no manual back-linking, no rot.

AI Turn Rating

Rate AI turns inline during an intent conversation. Per-turn signal flows into analytics so we can tune prompts on real data instead of vibes.

Workspace Ownership Transfer

Workspace owners can now transfer ownership to another member from the Danger Zone settings. Backed by a role-checked Postgres function so the transfer is atomic and safe.

Evidence Board 101 Playbook

New beginner playbook on running the Evidence Board as a diagnostic play — collect symptoms, spot patterns, decide where to operate — instead of a feedback dumping ground.

Blog Subscribe

Readers can now subscribe to the Pathmode blog. Email capture, confirmation, and rate-limited handling all in-app.

Changed

•Robots.txt moved from static `public/robots.txt` to dynamic `app/robots.ts` — adds disallows for /api/, /intent/, demo, and review routes, references the sitemap
•Dropped `offers` from SoftwareApplication structured data — Pathmode is SaaS, shopping-result fields do not apply

Removed

•Legacy Figma plugin package (was drifting from the current intent model)
•Legacy "Journey Mapping 101" playbook — replaced by Evidence Board 101
•Standalone `ClaudeCodePage` component — replaced by the shared `AgentIntegrationPage` used by both /claude-code and /codex

Improvements

•Spec Panel — Refactored, splitting execution hand-offs (Jira / Linear / Build Queue) and lifecycle controls (status, verification, review) into focused components. Smaller surface, fewer re-renders
•Intent Quality scoring centralized in one library so the Spec Panel, AI Spec Review, and readiness checks all use the same rubric
•Intent Interview JSON schema (with tests) replaces implicit field-name agreements between AI and client
•Glossary entries now cross-link to each other and to core terms — a small built-in knowledge graph
•New unified workspace integrations API across Jira, GitHub, Linear, and Slack with proper credential redaction
•Workspace settings — new Rename and Export Context endpoints, plus tracking for conversation-fired AI guidance rules
•Landing page hero, copy, and section structure tightened around "know what to build before you build it"

April 30, 2026

Features

First-Run Workspace Onboarding

Onboarding-Aware Intent Conversation

Improvements

•Free plan now includes 3 editor seats (up from 1) — small teams can get started without hitting the paywall
•Product creation now returns full metadata, allowing onboarding and analytics flows to tie created products to follow-up intents without extra lookups
•Paid-plan CTAs open the in-app contact modal when billing is not enabled, keeping interested teams inside the Pathmode experience
•Updated primary landing CTAs to "Start for free" and tightened homepage messaging
•Simplified product demo animation state to reduce timer/state drift
•Onboarding analytics across the full activation funnel — hero views, submissions, skips, first AI response, and first spec update

April 6, 2026

Features

Intent Readiness Scoring

Smarter Context Interviews

The product context builder runs a deeper interview flow, producing richer product context that feeds into every spec your team writes. Better context in, better specs out.

Agent Prompt Generation

The MCP server now generates full execution prompts for intents with readiness checks. AI agents get structured instructions with objectives, outcomes, constraints, and verification steps.

Plan Changes & Team Management

Workspaces can now upgrade, downgrade, and manage team members directly from billing settings. Full Lemon Squeezy integration for checkout, portal, and subscription sync.

Fixed

•Billing status hook, checkout flow, and portal redirect issues across multiple plan types

Improvements

•API Reliability — Rewrote intent API routes to share a single, tested handler layer with consistent validation
•MCP Server — Expanded tool surface: enhanced local file reading, better intent compilation, improved export formatting
•Context Health — Updated indicators so teams see at a glance which products have strong context and which need attention

March 29, 2026

Features

Functional Roles

Improvements

•Role-aware onboarding now prompts first-time users to choose their lens from workspace home
•Functional role selection is saved to the user profile and can be updated later in Profile Settings
•Added database support for functional_role with a safe default and validation constraint

March 28, 2026

Features

Evidence Inbox

AI Intent Synthesis

Structured Outcomes

Intent Scope

New scope field on IntentSpec with explicit inScope and outOfScope lists that tell AI agents what they can and cannot touch. Supported in the editor, API, and MCP server.

Intent Engineering vs Context Engineering

New blog post exploring why intent is upstream of context when building with AI. Teams optimize AI briefing before figuring out what to ask — better context just helps AI fail more efficiently.

Changed

•Removed beta banner — header now shows "Pathmode has launched!" announcement
•Restructured workspace settings with new API Keys tab

Improvements

•MCP Server — Structured outcome support, live quality checking with explicit rubric, scope field support, product strategy in exports
•Document View — Keyboard navigation in outcome lists, structured outcome rendering with priority labels, provenance view integration
•Evidence Board — New Inbox and Sources layout modes, layout selection in sidebar sub-menu, triage status counts in badges
•PostHog — Server-side event tracking with session and distinct ID headers

March 21, 2026

Features

Conversation-First Intent Editor

ThinkingQuality Scoring

4-dimensional assessment (specificity, testability, coherence, tension) scored 0-3 each, replacing the previous SpecCompleteness metric for more nuanced feedback.

Evidence-Anchored Specs

AI auto-links outcomes and edge cases to product evidence items. Specs display inline badges showing evidence links that persist across updates.

Changed

•Legacy form-based editor available via ?mode=form query parameter; conversation-first is now default

Fixed

•Intent editor save flow to properly persist spec updates and undo history
•Navigation issues between intent editor and product home

Improvements

•Intent Editor Performance — Consolidated state machine and reduced re-renders
•UI Consistency — Spec panel uses workspace design system with gray tones and no decorative hover effects

March 14, 2026

Features

Constraints Document Discoverability

Improvements

•Context Builder guardrails phase now includes a file upload button for constraints documents (PDF, DOCX, TXT, MD, CSV)
•New Intent modal shows a tip about constraints documents when none is uploaded for the product

March 11, 2026

Features

Smarter, Faster AI Conversations

Fixed

•AI would loop on the same topic or re-ask questions you already answered
•Conversations could get stuck on "Thinking..." with no way to recover — retry options now appear on failure

March 9, 2026

Features

PostHog Analytics Integration

Full-stack analytics with client-side event tracking and server-side error capture via PostHog EU, routed through a reverse proxy.

In-App Feedback System

"Send feedback" in the sidebar now opens a modal that emails the team directly via Resend (replaces mailto: link).

Product Creation Onboarding Flow

After creating a product, a guided welcome step offers "Get Started" (Context Builder) or "I'll explore on my own".

Changed

•Evidence Board redesign — Tabbed layout (Evidence / Sources), sort dropdown, more compact cards, dedicated "Suggest intent" button
•Intent Editor spec panel widened from 480px to 560px for better readability
•Product Home — Strategy section split into compact two-column card (Vision | North Star); tighter spacing throughout
•Insights Digest — Renamed from "Signals Digest"; critical items now use amber highlights; "View all" links to Evidence
•Hero demo — Streamlined from 7 steps to 6 (Signals step removed)

Removed

•Entire Signals system — Views, hooks, services, API routes, cron job, MCP tools, Slack notifications, and all UI references
•Glyphex analytics tracker — Replaced by PostHog
•Outcome quality scoring — Removed per-outcome quality badges (Testable/Decent/Vague) from the Intent Editor

March 6, 2026

Features

Hand-Drawn Illustrations

March 5, 2026

Features

Landing Page Refresh

Philosophy Page Rewrite

Improvements

•Navigation link renamed from "Our Philosophy" to "Our Approach"
•Hero demo now shows workspace vs. product sidebar modes
•Each demo step includes a description bar explaining the screen
•Removed unused IntentEngineDiagram and WhyHowWhenPositioning imports from landing page

March 3, 2026

Features

Gemini 3 Flash & Global AI Endpoint

March 1, 2026

Features

Living Specs — The Spec Lifecycle

Make specs feel alive — grounded in evidence, reviewed by humans, consumed by agents, and improved by outcomes. That's the loop that makes Pathmode indispensable.

Evidence-Anchored Specs

Spec Review with Evidence Backing

Reactive Signals

Agent Verification Loop

Improvements

•Evidence freshness indicator on specs — shows amber warning when linked evidence is older than 30 days
•Signal count badge in sidebar navigation
•Verification score badges on Build Queue kanban cards
•Verification failure signals with upsert dedup — re-running verification updates the signal instead of creating duplicates
•Updated AI verification model from Gemini 2.0 Flash to Gemini 2.5 Flash

February 22, 2026

Features

Intent Compiler for Claude Code

MCP Server v1.2 — Works Without API Key

Official MCP Registry Listing

Intent Compiler Page

Improvements

•MCP server README rewritten to lead with Intent Compiler and zero-config setup
•API key changed from required to optional in manifest.json
•Fixed pre-existing TypeScript error in MCP server api-client.ts

February 18, 2026

Features

Cursor Directory Listing

February 17, 2026

Features

Integrations Hub

Google Analytics Integration

Zendesk Integration

GitHub Integration

Link your GitHub organization to connect repositories with your workspace. Pull requests can be linked to Intent Specs, closing the loop from product intent to shipped code.

Linear Integration

Connect Linear to link issues and projects with Intent Specs. Track how product decisions flow from evidence through specs into engineering execution.

February 14, 2026

Features

Claude Code Integration Page

February 7, 2026

Features

Hero Animation as Video with Remotion

Improvements

•Added Remotion video rendering pipeline with preview and export scripts
•Downloaded and bundled Inter and Merriweather fonts for server-side video rendering

February 6, 2026

Features

IntentSpec Open Standard

About Page

Added a dedicated About page with the Pathmode story, company values, and founder bio.

Philosophy & Product Demo

Playbook Guides

Export to Agent

Improvements

•Landing Page: Redesigned hero section with interactive product demo and updated value proposition
•Solutions Pages: Added solution-specific landing pages for Journey Management, Product Intent Infrastructure, and Enterprise
•Footer: Added links to About, Philosophy, Playbook, and Comparison pages
•Terms of Service: Updated publicity rights and terms
•Intent Editor: Multiple stability and UX fixes across the spec editing workflow
•Blog: Published new articles on intent-driven development and AI-native teams
•Invitation Flow: Improved account handling and workspace invitation acceptance

February 1, 2026

Features

Intent Spec Workflow

January 29, 2026

Features

Inline Friction Highlighting

Replaced list-based friction display with Google Docs-style inline highlights directly within research documents. Click a highlight to see the finding, severity, and synthesize intents.

Document Highlighter Component

New DocumentHighlighter with viewport-aware popover positioning that never gets cut off by the browser window.

Improvements

•Document View: Added read/edit toggle via eye/pencil icons
•Friction Pipeline: Replaced "Specs Shipped" stat with "Evidence Docs" count
•Interactions: Changed popovers to click-based triggering for better stability
•Navigation: Fixed "View All Friction Points" button behavior
•Persistence: Upgraded storage engine to support rich friction metadata (findings, tags)

January 26, 2026

Improvements

•New Journey Experience: Redesigned modal to prioritize "Start from Friction" workflow
•Journeys List: Refined list view layout to reduce visual noise
•Welcome Email: Updated messaging to align with "Intent Layer" philosophy
•Build Queue: Upgraded to Kanban board interface for better tracking

January 22, 2026

Features

Intent Graph

Introduced interactive graph visualization for intent dependencies and strategic alignment. Visualize how your intents connect and ensure they align with your workspace strategy.

January 18, 2026

Features

From Journey Maps to Intent Layer

We started building Pathmode as a journey mapping tool. The vision was simple: help teams visualize customer experiences and connect user research to real workflows.

Intent is the new code. Specs are the new handoff. The map is the new constitution.

January 11, 2026

Improvements

•Admin Notifications: Implemented email alerts for new sign-ups
•UI/UX: Redesigned Blog Index page with responsive grid and author metadata
•Architecture: Decomposed MainApp for Next.js App Router migration
•Docs: Updated competitive landscape analysis
•Fixed: Resolved modal navigation issues on public pages

January 8, 2026

Features

Anonymization Firewall

Improvements

•Added PII detection toggle in workspace settings
•Updated data ingestion pipeline with pre-processing hook

January 1, 2026

Features

Interactive Live Demos

Improvements

•Smart Caching: Implemented intelligent local storage caching for public viewers, ensuring instant load times for AI artifacts like Storyboards and Proto Specs
•Optimized Public View: Revised staleness detection logic to prioritize cached content for read-only users, preventing unnecessary regeneration loops
•Robust AI Pipeline: Enhanced the generation engine with better payload management to reliably handle complex, data-heavy journey maps

December 29, 2025

Features

Integrations View & Live Embeds

Showcase your tool ecosystem with the new Integrations view. Embed live Figma prototypes directly into journey steps to keep designs synchronized.

Improvements

•Added "Integrations" to main navigation
•Added "Design Prototype" panel to Step properties for embedding support

December 28, 2025

Features

Specs View: Triage & Kanban

Redesigned the Feature Specs interface to split workflows into two distinct views:

Triage View: Review and approve new opportunities.
Kanban Board: Track specs from "Draft" to "Shipped".

Added an Archive workflow to hide deprecated or irrelevant specs.

Compact Research Dashboard

Refined the Research Dashboard for density. Key performance indicators are now grouped into a single row, providing an immediate overview of library health.

Hybrid Rendering Architecture

Re-architected the application core. Marketing pages now use Server-Side Rendering (SSR) for SEO and initial load performance, while the workspace remains a client-side SPA for interactivity.

Improvements

•Updated Research Dashboard visual hierarchy
•Implemented `useAppNavigation` for consistent routing state
•Resolved authentication race conditions with optimistic session hydration
•Fixed session persistence issues on refresh
• eliminated "Map Not Found" navigation errors
•Resolved client-side hydration mismatches
•Fixed React hook ordering stability issues
•Centralized modal state management

December 26, 2025

Features

Brand Identity Update

Refined the logo for better legibility at small sizes. The new mark retains the spiral geometry with cleaner lines and balanced weight.

Before

After

Improvements

•Updated favicon and app icons across the platform
•Unified spacing standards in navigation bars

December 25, 2025

Features

Feature Specs

Journey Editor Refinements

Improvements

•Privacy Overhaul: Removed Google Tag Manager and installed privacy-first GA4 with strict cookie consent
•Updated default workspace logo to use the Pathmode brand icon
•Various performance tunings and layout fixes in the editor canvas

December 24, 2025

Features

New "Product Teams" Landing Page

Improvements

•Updated front page tagline: "Stop managing journeys. Start building them."
•Added new "For Product Teams" navigation link in footer

December 22, 2025

Features

Dynamic Hero Video

The landing page now features an immersive autoplay video in the hero section, replacing the static image. This provides an instant, dynamic preview of the Pathmode interface in action.

Improvements

•Optimized landing page hero performance with video assets

December 21, 2025

Features

New "Pathmode vs The Rest" Comparison

Performance Tuning

Improvements

•Fixed "zombie session" issue ensuring immediate and permanent sign-out
•Cleaned up security section iconography on landing page
•Updated site-wide navigation to new comparison breakdown

December 19, 2025

Features

Analysis Workbench

Vibrant Cartoon Storyboards

Improvements

•Updated AI prompt to request "vibrant cartoon style" for storyboard panels
•Enhanced image generation parameters for consistent flat design aesthetics

December 16, 2025

Features

Competitor Comparison Page

Added a detailed comparison page to help users evaluate Pathmode against other journey mapping tools. This transparency helps teams make informed decisions about their journey management toolkit.

Improvements

•Restored missing comparison page routing and functionality
•Fixed an issue where the /compare link was resulting in a blank page

December 14, 2025

Features

Redesigned Front Page with AI-Style Illustrations

Unified Design Workflow Visual Updates

Enhanced Trust & Security Page

Dynamic CTA Section with Rotating Words

Improvements

•Added pictograms to "The Problem We Solve" section: Research Silos, Design Chasm, and Costly Drift
•Added workflow illustrations to all three steps in the "Unified Design Workflow" section
•Redesigned Trust Center hero with security illustration and improved visual hierarchy
•Improved visual consistency across landing page sections with flat, minimalist design
•Enhanced image presentation with full-size pictograms that fill their container elements
•Updated color scheme to use theme colors (subtle gradients) for better brand consistency
•Added smooth word rotation animation to CTA section without layout shifts

December 13, 2025

Features

Enhanced Landing Page with Visual Journey Flow

Improvements

•Improved visual communication of Pathmode's core value proposition on the front page
•Streamlined image display to use a single, clear visual representation of the journey mapping process

December 12, 2025

Features

Improved Workspace Settings Access

Improvements

•Moved workspace settings link from bottom navigation to header below workspace name
•Settings now use workspace URL key format for consistent routing
•Simplified "Back to Workspace" button to just "Back" in settings view

December 11, 2025

Features

Enhanced Storyboard Generator with High-Quality AI Images

Improvements

•Fixed image generation to properly extract images from API response structure
•Improved image quality from SVG placeholders to high-resolution PNG images
•Added authentication check - unauthenticated users see placeholder images with sign-up prompts

December 10, 2025

Improvements

•Improved Google sign-in experience - you'll now see "Pathmode" when signing in with Google for a more professional and secure authentication process

December 8, 2025

Features

Workspace Settings Navigation

Moved Workspace Settings to the main navigation panel for easier access. Settings now use workspace-specific URLs (e.g., /workspace-name/settings) for better organization and navigation consistency.

Improvements

•Fixed Activity feed navigation link
•Improved workspace invitation permissions - workspace owners can now create and manage invitations
•Streamlined Users & Team view by removing redundant Settings button
•Updated RLS policies to support workspace owner role for invitation management

December 6, 2025

Features

Trust & Security Center

Comprehensive security and privacy information page showcasing our security measures and compliance standards.

Team Management & Activity Feed

Complete overhaul of team management with email invitations, granular roles (Owner, Admin, Editor, Commenter, Viewer), and a new activity feed to track workspace actions.

Cookie Consent Management

GDPR-compliant cookie consent banner with granular controls for optional cookies and analytics.

Improvements

•Improved footer navigation across all pages
•Enhanced landing page with security information

November 27, 2025

Features

Beta Launch

Pathmode is now in public beta! We're excited to share our AI-powered journey mapping platform with designers and product teams.

API Rate Limiting

Implemented distributed rate limiting with Upstash Redis to prevent API abuse and ensure fair usage across all endpoints.

GDPR Compliance

Added comprehensive GDPR compliance features including cookie consent banner, privacy policy, terms of service, and self-service account deletion.

Profile Page

New profile page accessible from the sidebar where users can manage their account settings and delete their account.

Performance Monitoring

Integrated Vercel Speed Insights to monitor application performance and identify optimization opportunities.

Improvements

•Enhanced security headers and Content Security Policy
•Improved Trust & Security page with detailed compliance information
•Added links to legal pages throughout the application
•Launched changelog to keep users informed of updates

Want to stay updated? Visit our Trust & Security page for the latest on security and compliance.