Skip to content

feat: welcome users with Mate, a first crewmate they get to name - #18406

Merged
iamwhatever merged 2 commits into
mainfrom
feat/mate-first-crewmate
Oct 9, 2026
Merged

iamwhatever merged 2 commits into
mainfrom
feat/mate-first-crewmate

Conversation

@buluoray

@buluoray buluoray commented Oct 9, 2026 •

Copy link
Copy Markdown
Contributor

Problem / Motivation

Goal: Give every user one ready crewmate, Mate, that welcomes them on the Crewmates page, explains in plain words what a crewmate is, and lets them give it a name of their own.

A person who opens the Crewmates page for the first time finds no crewmate at all: nothing speaks first, and they have to know what to ask and what a crewmate is before anything happens. The decision is recorded in docs/request-for-change/rfc-crewmate-guides-and-mate.md (accepted, merged in #18326). This is the first of three PRs that ship that RFC, in order: Mate (this one), then guided tours (#18301), then change cards.

Why it matters

The first minutes decide whether someone keeps using Kiro Crew. A crewmate that is already there, speaks first and asks to be named is the first thing a newcomer does with crewmates, and seeing the name change at once shows that crewmates are theirs to shape.

Not a goal

  • A special or protected assistant. Mate is an ordinary crewmate: renamable, rebindable and deletable like any other, and a deleted one is never re-created.
  • Releasing the Crewmates page. It stays behind its Feature Preview; with it off, nothing names Mate.
  • Guided tours and change cards. Mate's welcome offers to help in words; pointing at controls and prepared changes arrive in the next two PRs.
  • Changing where a new user lands when the app opens.
  • Greeting on every open, or a second greeting for crewmates the user creates: main's create-flow greeting stays the only one.
  • Turning the default member into Mate: it stays the shared fallback crew.

What changed (motivation → approach → change)

Approach: reuse the ordinary crewmate create path and the member event log the page already folds; add a once-only create, a once-only hidden first turn, and one self-rename tool on a new platform server that the next two PRs extend.

  • Mate (agent_materialization/first_crewmate.py): every install, fresh or upgraded, creates one ordinary crewmate once (key mate, shown as Mate while it has no display name) on the ordinary crewmate template with its own private memory. A marker records the attempt, so a deleted Mate never comes back, and a row the user already keeps under mate is left alone. It is created the way every crewmate is (in-memory row, provision_member_memory, persist_member_config(create=True)), so it is never on Global memory; a failed creation leaves nothing and is tried again at next start.
  • Welcome (dashboard/mate_welcome.py, handlers/members.py POST /api/members/{slug}/greet, pages/members/useFirstGreeting.ts): the first time Mate's thread opens empty, the gateway starts one hidden kickoff turn, claimed with an O_CREAT|O_EXCL marker so two tabs or a retry never greet twice. The kickoff states principles; the model writes the welcome: what a crewmate is, that it can help set Kiro Crew up, fix problems or show the user around, and what the user would like to call it. A user who already has crewmates hears it introduce itself as one more teammate. A greeting request that fails shows nothing: the chat works without it, and the next open asks again.
  • kirocrew-guide server and live rename (mcp_guide.py, dashboard/handlers/guide.py, src/kiro_crew/members.py rename_member_display): a new always-mounted platform server, granted by exact tool name through the allowedTools ceiling (agent._GUIDE_AUTO_GRANTS), once on an existing spec (agent_materialization/guide_platform.py). Its one tool, rename_self, renames only the caller's own crewmate thread. The shared admission _resolve_agent_caller admits only a turn the user sent from the dashboard, reading the opener provenance chat_runner.py now records (_turn_channel_origin, _turn_user_sent); a channel message, schedule, subagent, app or the hidden kickoff gets 403. The rename is a member/config event, so header and roster update at once.
  • Crewmates page (MembersPage.tsx, rosterFilter.ts, hooks/useMeetCrewmatesGate.ts): with the preview on, a never-chatted Mate opens first and Meet CrewMates stays shut over it; afterwards the page lands as on main. With the preview off, the agent pickers, Customize and a template's crew list leave Mate out (hideMateWithoutPreview in lib/assistantMember.ts). Every place that names a crewmate (both agent pickers, Customize, the Crewmates header) shows its display name only, never its ID; the picker still matches a search on either.

Smaller pieces: memory_stores.py creates a member's memory/ directory owner-only (0700); agent.py, agent_files.py, agent_materialization/__init__.py and service_agents.py mount the server and keep it off background agents; cli.py, cli_doctor.py, src/kiro_crew/doctor_checks, mcp_core.py, mcp_discovery.py, src/kiro_crew/mcp_gateway, mcp_cleanup.py, src/kiro_crew/onboarding_sources, security_posture.py, validation.py and config/defaults.json register it, its argument schema and its grant; dashboard/server.py, server_runtime/mcp_routes.py, dashboard/state.py, routes/agents.py and handlers/__init__.py wire the routes and the provenance fields; website/src/api, website/src/store, website/src/app-sdk, website/src/components (ChatPane, AgentSelector, crew editor, chat bubbles), pages/chat/*, pages/KiroCrewAgentsPage.tsx, hooks/useAgents.ts and website/src/lib carry Mate's label, its narration folding and the preview gate; website/src/i18n adds two strings in 12 locales; src/kiro_crew/docs and docs/ cover MCP, MCP shareability, crew mode, memory and the feature map; test/gui_user adds a scenario.

Backwards compatibility

Compatible: existing installs keep their members and conversations. An upgraded install gains Mate once as an added member; nothing existing is renamed, rebound or reserved, and the new server's routes accept nothing main accepted before.

Reader: website/src/hooks/useMeetCrewmatesGate.ts:useMeetCrewmatesGate -- first run and crew page -- with the Crewmates preview on, neither the tour-end nor the page-entry auto-open fires while Mate exists with no chat history; with Mate chatted with, deleted, the roster unreadable, or the preview off, both fire as on main (useMeetCrewmatesGate.test.tsx)
Reader: website/src/hooks/useAgents.ts:useAgents -- chat composer, Customize -- with the Crewmates preview off, the row keyed mate is left out of the agent lists; every other row is listed as on main (useAgents.hideMate.test.tsx)

Tests

  • Backend: test_first_crewmate_ordinary.py, test_welcome_owed.py, test_mate_welcome.py, test_mate_rename_self.py (renames only the caller, refuses a non-crewmate thread, the default crew and a turn the user did not send, cleans the name; the kickoff asks for Mate's name and states principles), test_guide_server.py (always mounted, per-tool grant, governance ceiling, one-time grant, background agents without it, first-crewmate creation, off-dashboard wording), test_guide_admission.py (real admission: channel turns, channel steers, non-user turns refused; the runner records the provenance), plus updated composition, spec-byte, MCP registry and doctor contracts.
  • Frontend: useFirstGreeting.test.tsx, ChatPane.mateChat.test.tsx, useAgents.hideMate.test.tsx, useMeetCrewmatesGate.test.tsx, assistantMember.test.ts, rosterFilter.test.ts, Members page and crew-editor cases.
  • Mutation checks: emptying _USER_TURN_OPERATIONS reddens test_a_turn_the_user_did_not_send_is_refused and test_only_a_turn_the_user_sent_can_rename; removing the channel check reddens test_a_channel_turn_on_the_mirrored_slot_is_refused and test_a_dashboard_turn_narrowed_by_a_channel_steer_is_refused.

Manual verification

On a preview gateway built from this head (fresh data home, real model turns): the roster held default and mate; Mate's row carried its own member_id and private store, and members/mate/ was mode 0700. With the Crewmates preview off, the chat agent picker listed no Mate. After turning the preview on in Settings → Developer, opening Mate's thread showed it busy and empty while the hidden kickoff ran, then Mate greeted first ("Hi. I'm Mate. I'm your first crewmate…", three offers, ending on "What would you like to call me?"). Replying "Call you Pebble" got "I go by Pebble now." and the header and composer showed Pebble with no reload. Mate's replies draw as ordinary bubbles, the same as every crewmate's. With the preview on, the chat agent picker lists the renamed crewmate as "Pebble" (that run was on the previous head, which still drew mate beside it; the current head draws the name alone, pinned by AgentDropdownList.test.tsx and MembersPage.test.tsx).4k tests) and vitest (about 2k tests). Not exercised: Windows and macOS hosts.

Screenshots / video

One continuous run on the preview gateway above, on this head, fresh data home: Mate's thread while its first greeting runs; the welcome; after "Call you Pebble"; the chat agent picker naming it Pebble (captured on the previous head, so the grey mate ID beside it is gone on this head); Pebble's editor in Customize.

1-waiting

2-welcome

3-renamed

4-picker

5-editor

Related Issues

Closes #18315

Part of the RFC in #18326; followed by #18301 (guided tours) and a change-cards PR.

Pattern harvest

Not generalizable: a feature PR.

Checklist

  • At most two commits (one is the norm), with a Conventional Commits title (feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)
  • Existing tests pass and new tests added for new functionality
  • Self-review completed; code follows project style guidelines
  • Documentation updated (if applicable)
  • No secrets, credentials, or internal references in the diff

Contribution License Agreement

Contributions are accepted under the project's Apache-2.0 license; see the Licensing section of CONTRIBUTING.md.

@buluoray
buluoray requested review from a team and KiroCrewCaptain as code owners October 9, 2026 08:46
@buluoray
buluoray requested a review from chenmingwei23 October 9, 2026 08:46
@github-actions github-actions Bot added the readiness: checking Automated validation is still running label Oct 9, 2026
@github-actions

github-actions Bot commented Oct 9, 2026 •

Copy link
Copy Markdown
Contributor

Sensitive Change Review

Sensitive files changed:

  • src/kiro_crew/security_posture.py

  • @buluoray posted a complete reasoning template for 5f1a0e4.

@github-actions

github-actions Bot commented Oct 9, 2026 •

Copy link
Copy Markdown
Contributor

GPT 6.1 Review — ✅ no blocking findings

GPT 6.1 completed its review of 5f1a0e49b690813bca3ae42b0d3dbda82e6ca11a and found no blocking issues.

This comment is updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] 5f1a0e4

False positive or not applicable? A repository writer can comment:
/ai-review override gpt 5f1a0e49b690813bca3ae42b0d3dbda82e6ca11a: <one-sentence reason>

@github-actions

github-actions Bot commented Oct 9, 2026 •

Copy link
Copy Markdown
Contributor

Design Review (Opus 5.5) — 🟡 CONCERNS

Design-level review of 5f1a0e49b690813bca3ae42b0d3dbda82e6ca11a — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

Design-Verdict: CONCERNS

Crewmate IDs are now hidden in every picker, which reverses a test that said they must stay visible. That change sits outside the Goal and is bundled into the Mate PR.

Watch

  • Hiding the ID reverses a recorded rule without answering its reason. Commit 5f1a0e4 deletes the pin // A label that COVERS the ID keeps the ID line: routes and crons address it., the agent_id_tooltip strings in 12 locales, and the AgentSelector comment saying a picker that hid the ID "would strand anyone wiring those up". Crons, spawn_run(crew=...) and the CLI still use the ID. Display names are not unique, and rename_self lets the model pick a label. So once a crewmate is renamed, its ID no longer appears in the pickers or the Crewmates header, and two crewmates with the same label look identical. The Goal ("lets them give it a name") does not need any of this.
    Clears when: the ID change moves to its own PR that answers the cron/spawn/CLI rationale, or the ID stays reachable (tooltip or secondary line) when a label covers it.
  • Mate is created even with the preview off, so "nothing names Mate" only holds in the dashboard. create_first_crewmate_once() runs at every boot whatever the Feature Preview says. Every upgraded install therefore gets a mate row in config.json and a private memory store. Only website pickers filter it out (hideMateWithoutPreview). Non-dashboard listings such as the CLI, channel commands and the error that names available crews still show mate to users who never turned the preview on.
    Clears when: creation is gated on the preview (or on the first time the Crewmates page opens), or the description narrows the claim to the dashboard pickers.

[DESIGN-REVIEWED] 5f1a0e4

Description read

[DESCRIPTION-READ] fd1a394a8ccc89a6b823f535bf12a3ddc1f559a08389c08d28e64759cc9eec8e

That sha256 names everything this verdict read: the PR description, plus the 1 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with .github/scripts/pr-description-capture.sh over the same inputs; if it differs, the description or that evidence changed after this verdict was formed and any finding drawn from either is unproven.

@github-actions

github-actions Bot commented Oct 9, 2026 •

Copy link
Copy Markdown
Contributor

First Principles Review (Opus 5.5) — ✅ human override accepted

@buluoray overrode this lane for 5f1a0e49b690813bca3ae42b0d3dbda82e6ca11a via /ai-review override (this commit only).

@github-actions

github-actions Bot commented Oct 9, 2026 •

Copy link
Copy Markdown
Contributor

UX Review (Opus 5.5) — ✅ human override accepted

@buluoray overrode this lane for 5f1a0e49b690813bca3ae42b0d3dbda82e6ca11a via /ai-review override (this commit only).

@github-actions

github-actions Bot commented Oct 9, 2026 •

Copy link
Copy Markdown
Contributor

Opus 5.5 Review — ✅ no blocking findings

Reviewed 5f1a0e49b690813bca3ae42b0d3dbda82e6ca11a — this comment is updated in place on each push.

Review details

No findings.

[OPUS-REVIEWED] 5f1a0e4

Verdict parsed from the review's SHA-scoped output markers for commit 5f1a0e49b690813bca3ae42b0d3dbda82e6ca11a.

False positive or not applicable? A repository writer can comment:
/ai-review override fable 5f1a0e49b690813bca3ae42b0d3dbda82e6ca11a: <one-sentence reason>

@github-actions github-actions Bot added readiness: action required A blocking check or review needs attention and removed readiness: checking Automated validation is still running labels Oct 9, 2026
@github-actions github-actions Bot added readiness: checking Automated validation is still running and removed readiness: action required A blocking check or review needs attention labels Oct 9, 2026
@github-actions github-actions Bot added the merge conflict Branch has merge conflicts with its base — author must resolve before merge label Oct 9, 2026
@buluoray
buluoray force-pushed the feat/mate-first-crewmate branch from ed5b74c to 6dd9d77 Compare October 9, 2026 10:30
@github-actions github-actions Bot added readiness: action required A blocking check or review needs attention and removed readiness: checking Automated validation is still running labels Oct 9, 2026
@buluoray
buluoray force-pushed the feat/mate-first-crewmate branch from 6dd9d77 to ab68b31 Compare October 9, 2026 10:43
@github-actions github-actions Bot added readiness: checking Automated validation is still running readiness: action required A blocking check or review needs attention and removed readiness: action required A blocking check or review needs attention merge conflict Branch has merge conflicts with its base — author must resolve before merge readiness: checking Automated validation is still running labels Oct 9, 2026
@buluoray
buluoray force-pushed the feat/mate-first-crewmate branch from ab68b31 to 2fdd164 Compare October 9, 2026 10:59
@github-actions github-actions Bot removed the readiness: action required A blocking check or review needs attention label Oct 9, 2026
@github-actions

github-actions Bot commented Oct 9, 2026

Copy link
Copy Markdown
Contributor

AI-review override not recorded: keep the reason to 500 characters or fewer.

@buluoray

buluoray commented Oct 9, 2026

Copy link
Copy Markdown
Contributor Author

/ai-review override first-principles 40c35f1: Product owner decision: crewmates are shown by display name only. The crew ID is a slug of the first name typed, so beside the name it repeats it, goes stale after a rename (Pebble shown with mate), or is a hash for CJK/emoji names; none of that helps a user. Agents that address a crewmate by ID in crons, spawn params or the CLI resolve it through tools rather than the UI. A read-only ID in the Customize editor for hand-written CLI/cron use is a follow-up, not part of this PR.

@buluoray

buluoray commented Oct 9, 2026

Copy link
Copy Markdown
Contributor Author

/ai-review override ux 40c35f1: Product owner decision to land this head without recapturing. The three missing stills (picker without the ID, Customize crew cards, picker with the preview off) are evidence gaps, not defects: the ID-less picker and header are pinned by AgentDropdownList.test.tsx and MembersPage.test.tsx (both redden when the ID span is restored), and the preview-off picker was captured on the previous head with no change to hideMateWithoutPreview since. The Watch item on the Customize fallback wording for Mate is noted as a follow-up.

@buluoray

buluoray commented Oct 9, 2026

Copy link
Copy Markdown
Contributor Author

Sensitive change review

  • Head commit: 40c35f1
  • What rule changed: security_posture.py registers the new kirocrew-guide MCP shim. mcp_guide.py joins the list of modules that redact only the refusal text they hand back to the model (the same class as mcp_panel.py; the boundary is the stdio transport), and MCP_GUIDE_SCHEMAS joins the tool-schema registries the posture inventory enumerates. An existing spec gains @kirocrew-guide and the one pre-approval @kirocrew-guide/rename_self once, and that addition now writes an SEL mcp_tools_added record like every other writer that adds a reference or a pre-approval. The first crewmate's memory directory is created owner-only (mkdir(mode=0o700)), and its members/<slug> directory is restricted to the owner. No deny rule, sandbox scope, credential path or approval default changes.
  • Before -> after (what was allowed/blocked, now): before, no kirocrew-guide server existed. After, it exists with one tool, rename_self, whose route admits only a turn the user sent from the dashboard and renames only the caller's own crewmate, on the memory store its thread runs on; a turn that a non-user text was steered into is refused like any turn the user did not send. Nothing that was blocked before is allowed now.
  • Worst case if wrong (breaks commands / leaks secret / locks user out): a refusal string from the gateway reaching the model unredacted. The shim passes every error through redact_via_context before returning it, and the gateway's refusal text names no secret.
  • Evidence checked (which sandbox result proves it): test_security_posture.py (inventory and redaction-module coverage) and test_mcp_call_site_auth_coverage.py pass on this head; the route's admission is pinned by test_guide_admission.py and test_mate_rename_self.py.
  • No regression:
    • Tools still work (which tools/commands were run, result): on a preview gateway built from this branch, rename_self renamed Mate live ("Call you Pebble" → "Pebble it is. I go by Pebble now."); the default spec mounted kirocrew-guide with exactly @kirocrew-guide/rename_self granted.
    • Backward compatible (old config / old data / old callers still work, how checked): existing specs gain the ref and grant once (guide_platform_granted.json), covered by test_guide_server.py; no existing route or schema changed.
    • Existing tests (which suites passed on this head): test_security_posture.py, test_mcp_call_site_auth_coverage.py, test_guide_server.py, test_guide_admission.py, test_mate_*.py, composition and spec-byte contracts (spec-byte audit digests regenerated for the new SEL record), test/integration/test_owner_only_boot.py (only the pre-existing workspace/md-notebook line, same as on main); mypy, flake8, tsc, build, i18n.
  • Memory: Mate is created through the ordinary member-create path (provision_member_memory then persist_member_config(create=True)), never on Global memory; memory_stores.py differs from main only by the 0o700 mode. A failed creation leaves no row, store or marker.
  • How to undo: revert the commit; the two list entries and the shim go with it, and the one-time grant marker only records that the grant ran.

@github-actions

github-actions Bot commented Oct 9, 2026

Copy link
Copy Markdown
Contributor

Human judgment recorded

@buluoray marked the first-principles AI finding as false positive, not applicable, or explicitly accepted for 40c35f17d8c352e72d0972d2c92e57e8a4342f05.

Product owner decision: crewmates are shown by display name only. The crew ID is a slug of the first name typed, so beside the name it repeats it, goes stale after a rename (Pebble shown with mate), or is a hash for CJK/emoji names; none of that helps a user. Agents that address a crewmate by ID in crons, spawn params or the CLI resolve it through tools rather than the UI. A read-only ID in the Customize editor for hand-written CLI/cron use is a follow-up, not part of this PR.

This decision applies only to this commit. A new push requires a new judgment.

@buluoray

buluoray commented Oct 9, 2026

Copy link
Copy Markdown
Contributor Author

GPT 6.1 on 40c35f1

  • MembersPage.test.tsx:630 — use the existing ceiling for chained waits — fixed in 5232113. The greeting wait at line 630 and the roster-retry and thread waits at lines 4275–4276 now pass PANE_READY (the file's 5000 ms ceiling), so the roster → thread → greeting chain no longer times out at the default 1 s under load. MembersPage.test.tsx passes (231 tests). Only those three waits changed.

@buluoray

buluoray commented Oct 9, 2026

Copy link
Copy Markdown
Contributor Author

/ai-review override first-principles 5232113: Product owner decision: crewmates are shown by display name only. The crew ID is a slug of the first name typed, so beside the name it repeats it, goes stale after a rename (Pebble shown with mate), or is a hash for CJK/emoji names; none of that helps a user. Agents that address a crewmate by ID in crons, spawn params or the CLI resolve it through tools rather than the UI. A read-only ID in the Customize editor for hand-written CLI/cron use is a follow-up, not part of this PR.

@buluoray

buluoray commented Oct 9, 2026

Copy link
Copy Markdown
Contributor Author

/ai-review override ux 5232113: Product owner decision to land this head without recapturing. The three missing stills (picker without the ID, Customize crew cards, picker with the preview off) are evidence gaps, not defects: the ID-less picker and header are pinned by AgentDropdownList.test.tsx and MembersPage.test.tsx (both redden when the ID span is restored), and the preview-off picker was captured on the previous head with no change to hideMateWithoutPreview since. The Watch item on the Customize fallback wording for Mate is noted as a follow-up.

@buluoray

buluoray commented Oct 9, 2026

Copy link
Copy Markdown
Contributor Author

Sensitive change review

  • Head commit: 5232113
  • What rule changed: security_posture.py registers the new kirocrew-guide MCP shim. mcp_guide.py joins the list of modules that redact only the refusal text they hand back to the model (the same class as mcp_panel.py; the boundary is the stdio transport), and MCP_GUIDE_SCHEMAS joins the tool-schema registries the posture inventory enumerates. An existing spec gains @kirocrew-guide and the one pre-approval @kirocrew-guide/rename_self once, and that addition now writes an SEL mcp_tools_added record like every other writer that adds a reference or a pre-approval. The first crewmate's memory directory is created owner-only (mkdir(mode=0o700)), and its members/<slug> directory is restricted to the owner. No deny rule, sandbox scope, credential path or approval default changes.
  • Before -> after (what was allowed/blocked, now): before, no kirocrew-guide server existed. After, it exists with one tool, rename_self, whose route admits only a turn the user sent from the dashboard and renames only the caller's own crewmate, on the memory store its thread runs on; a turn that a non-user text was steered into is refused like any turn the user did not send. Nothing that was blocked before is allowed now.
  • Worst case if wrong (breaks commands / leaks secret / locks user out): a refusal string from the gateway reaching the model unredacted. The shim passes every error through redact_via_context before returning it, and the gateway's refusal text names no secret.
  • Evidence checked (which sandbox result proves it): test_security_posture.py (inventory and redaction-module coverage) and test_mcp_call_site_auth_coverage.py pass on this head; the route's admission is pinned by test_guide_admission.py and test_mate_rename_self.py.
  • No regression:
    • Tools still work (which tools/commands were run, result): on a preview gateway built from this branch, rename_self renamed Mate live ("Call you Pebble" → "Pebble it is. I go by Pebble now."); the default spec mounted kirocrew-guide with exactly @kirocrew-guide/rename_self granted.
    • Backward compatible (old config / old data / old callers still work, how checked): existing specs gain the ref and grant once (guide_platform_granted.json), covered by test_guide_server.py; no existing route or schema changed.
    • Existing tests (which suites passed on this head): test_security_posture.py, test_mcp_call_site_auth_coverage.py, test_guide_server.py, test_guide_admission.py, test_mate_*.py, composition and spec-byte contracts (spec-byte audit digests regenerated for the new SEL record), test/integration/test_owner_only_boot.py (only the pre-existing workspace/md-notebook line, same as on main); mypy, flake8, tsc, build, i18n.
  • Memory: Mate is created through the ordinary member-create path (provision_member_memory then persist_member_config(create=True)), never on Global memory; memory_stores.py differs from main only by the 0o700 mode. A failed creation leaves no row, store or marker.
  • How to undo: revert the commit; the two list entries and the shim go with it, and the one-time grant marker only records that the grant ran.

Every install, fresh or upgraded, gains one ordinary crewmate, Mate,
created once. With the Crewmates preview on, the first visit opens
Mate, which speaks first: one hidden kickoff turn asks the model to
explain in plain words what a crewmate is, what it can help with, and
what the user would like to call it. When the user answers, Mate calls
the new kirocrew-guide tool rename_self, and the header and roster
show the new name at once. With the preview off, Mate is hidden from
every list that names crew.

kirocrew-guide is a new always-mounted platform server. This change
carries its registration, the one-time grant on existing specs, and the
gateway admission its tools share: only a turn the user sent from the
dashboard may call it, never a channel message, schedule, subagent,
app or the hidden welcome kickoff.

Mate is deletable and renamable like any crewmate, never re-created
after deletion, and a crewmate the user already keeps under the key
mate is left alone. The decision is recorded in
docs/request-for-change/rfc-crewmate-guides-and-mate.md.
@buluoray

buluoray commented Oct 9, 2026

Copy link
Copy Markdown
Contributor Author

/ai-review override first-principles 5f1a0e4: Product owner decision: crewmates are shown by display name only. The crew ID is a slug of the first name typed, so beside the name it repeats it, goes stale after a rename (Pebble shown with mate), or is a hash for CJK/emoji names; none of that helps a user. Agents that address a crewmate by ID in crons, spawn params or the CLI resolve it through tools rather than the UI. A read-only ID in the Customize editor for hand-written CLI/cron use is a follow-up, not part of this PR.

@buluoray

buluoray commented Oct 9, 2026

Copy link
Copy Markdown
Contributor Author

/ai-review override ux 5f1a0e4: Product owner decision to land this head without recapturing. The three missing stills (picker without the ID, Customize crew cards, picker with the preview off) are evidence gaps, not defects: the ID-less picker and header are pinned by AgentDropdownList.test.tsx and MembersPage.test.tsx (both redden when the ID span is restored), and the preview-off picker was captured on the previous head with no change to hideMateWithoutPreview since. The Watch item on the Customize fallback wording for Mate is noted as a follow-up.

@buluoray

buluoray commented Oct 9, 2026

Copy link
Copy Markdown
Contributor Author

Sensitive change review

  • Head commit: 5f1a0e4
  • What rule changed: security_posture.py registers the new kirocrew-guide MCP shim. mcp_guide.py joins the list of modules that redact only the refusal text they hand back to the model (the same class as mcp_panel.py; the boundary is the stdio transport), and MCP_GUIDE_SCHEMAS joins the tool-schema registries the posture inventory enumerates. An existing spec gains @kirocrew-guide and the one pre-approval @kirocrew-guide/rename_self once, and that addition now writes an SEL mcp_tools_added record like every other writer that adds a reference or a pre-approval. The first crewmate's memory directory is created owner-only (mkdir(mode=0o700)), and its members/<slug> directory is restricted to the owner. No deny rule, sandbox scope, credential path or approval default changes.
  • Before -> after (what was allowed/blocked, now): before, no kirocrew-guide server existed. After, it exists with one tool, rename_self, whose route admits only a turn the user sent from the dashboard and renames only the caller's own crewmate, on the memory store its thread runs on; a turn that a non-user text was steered into is refused like any turn the user did not send. Nothing that was blocked before is allowed now.
  • Worst case if wrong (breaks commands / leaks secret / locks user out): a refusal string from the gateway reaching the model unredacted. The shim passes every error through redact_via_context before returning it, and the gateway's refusal text names no secret.
  • Evidence checked (which sandbox result proves it): test_security_posture.py (inventory and redaction-module coverage) and test_mcp_call_site_auth_coverage.py pass on this head; the route's admission is pinned by test_guide_admission.py and test_mate_rename_self.py.
  • No regression:
    • Tools still work (which tools/commands were run, result): on a preview gateway built from this branch, rename_self renamed Mate live ("Call you Pebble" → "Pebble it is. I go by Pebble now."); the default spec mounted kirocrew-guide with exactly @kirocrew-guide/rename_self granted.
    • Backward compatible (old config / old data / old callers still work, how checked): existing specs gain the ref and grant once (guide_platform_granted.json), covered by test_guide_server.py; no existing route or schema changed.
    • Existing tests (which suites passed on this head): test_security_posture.py, test_mcp_call_site_auth_coverage.py, test_guide_server.py, test_guide_admission.py, test_mate_*.py, composition and spec-byte contracts (spec-byte audit digests regenerated for the new SEL record), test/integration/test_owner_only_boot.py (only the pre-existing workspace/md-notebook line, same as on main); mypy, flake8, tsc, build, i18n.
  • Memory: Mate is created through the ordinary member-create path (provision_member_memory then persist_member_config(create=True)), never on Global memory; memory_stores.py differs from main only by the 0o700 mode. A failed creation leaves no row, store or marker.
  • How to undo: revert the commit; the two list entries and the shim go with it, and the one-time grant marker only records that the grant ran.

@github-actions

github-actions Bot commented Oct 9, 2026

Copy link
Copy Markdown
Contributor

AI-review override not recorded: 52321132390932c3d38b4ff95d6c9313451ca596 is not the current PR head. Re-run the command with 5f1a0e49b690813bca3ae42b0d3dbda82e6ca11a.

@buluoray

buluoray commented Oct 9, 2026

Copy link
Copy Markdown
Contributor Author

/ai-review override ux 5f1a0e4: Product owner decision to land without recapturing. The missing stills (picker and Customize card without the ID, picker with the preview off) are evidence gaps, not defects: the ID-less picker and header are pinned by AgentDropdownList.test.tsx and MembersPage.test.tsx, which redden when the ID span returns, and hideMateWithoutPreview is unchanged since the last capture. The Customize fallback wording for Mate is a follow-up.

@github-actions

github-actions Bot commented Oct 9, 2026

Copy link
Copy Markdown
Contributor

Human judgment recorded

@buluoray marked the ux AI finding as false positive, not applicable, or explicitly accepted for 5f1a0e49b690813bca3ae42b0d3dbda82e6ca11a.

Product owner decision to land without recapturing. The missing stills (picker and Customize card without the ID, picker with the preview off) are evidence gaps, not defects: the ID-less picker and header are pinned by AgentDropdownList.test.tsx and MembersPage.test.tsx, which redden when the ID span returns, and hideMateWithoutPreview is unchanged since the last capture. The Customize fallback wording for Mate is a follow-up.

This decision applies only to this commit. A new push requires a new judgment.

@buluoray

buluoray commented Oct 9, 2026

Copy link
Copy Markdown
Contributor Author

/ai-review override first-principles 5f1a0e4: Product owner decision: crewmates are shown by display name only. The crew ID is a slug of the first name typed, so beside the name it repeats it, goes stale after a rename (Pebble shown with mate), or is a hash for CJK/emoji names; none of that helps a user. Agents that address a crewmate by ID in crons, spawn params or the CLI resolve it through tools rather than the UI. A read-only ID in the Customize editor for hand-written CLI/cron use is a follow-up, not part of this PR.

@github-actions

github-actions Bot commented Oct 9, 2026

Copy link
Copy Markdown
Contributor

Human judgment recorded

@buluoray marked the first-principles AI finding as false positive, not applicable, or explicitly accepted for 5f1a0e49b690813bca3ae42b0d3dbda82e6ca11a.

Product owner decision: crewmates are shown by display name only. The crew ID is a slug of the first name typed, so beside the name it repeats it, goes stale after a rename (Pebble shown with mate), or is a hash for CJK/emoji names; none of that helps a user. Agents that address a crewmate by ID in crons, spawn params or the CLI resolve it through tools rather than the UI. A read-only ID in the Customize editor for hand-written CLI/cron use is a follow-up, not part of this PR.

This decision applies only to this commit. A new push requires a new judgment.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Welcome new users with Mate, a first crewmate they name

3 participants