Skip to content

feat(members): show a Bot's conversation starters on its empty chat - #37246

Open
javenciu wants to merge 1 commit into
mainfrom
feat/bot-conversation-starters
Open

javenciu wants to merge 1 commit into
mainfrom
feat/bot-conversation-starters

Conversation

@javenciu

@javenciu javenciu commented Nov 14, 2026 •

Copy link
Copy Markdown
Contributor

Problem / Motivation

Goal: a Bot's owner can pin up to four short prompts to the Bot, and its empty chat offers them as buttons, so a new person sees what the Bot is for and starts with one press.

A Bot's chat opens on nothing until the Bot has said something. git grep -i -E "conversation.?starter|suggested.?prompts" over website/src finds only unrelated hits.

Why it matters

The empty chat is the first thing a person sees after choosing a Bot. A prompt the owner wrote costs one press.

Not a goal

  • Starters on a public page or widget, in a .kirobot file (received text would need its own review card row), or editable by the Bot itself.
  • Rotating, scoring or templating prompts.

What changed

The Bot record has no field for prompts, so the chat has nothing to show.

  • src/kiro_crew/member_starters.py (new): clean_starters returns the starters that pass, a warning for each value left out, and its place and reason code (too_long, repeated, over_limit, credential, not_text, empty). Each is cleaned as a display name is (members._strip_controls). A value is left out when it is not text, empty, over 120 characters, credential-shaped (external_text_requires_redaction), a repeat, or past the fourth.
  • config/sections_session.py, config/loader_pipeline.py: the Bot record gains starters, default empty. The loader cleans a hand-edited value, so a bad value never fails a load. config-baseline.json is regenerated.
  • dashboard/agent_admin/crew_update.py: the update route accepts starters. It saves what passes and answers with the kept starters, starters_left_out (place and reason) and warnings. A non-list keeps the saved value and warns; never a 400.
  • dashboard/handlers/members_roster.py, core_config_view.py: the roster row carries starters, and GET /api/config/kirocrew masks agents.*.starters item by item. A starter the mask would change is left out, not shown as a sentinel. A ratchet holds every list[str] Bot field to the mask.
  • website/: CrewmateStarters.tsx shows the buttons while a loaded Bot chat holds no row and no turn runs, one per row (max-two-buttons-per-row); a press calls doSend. CrewStartersGroup.tsx is a box in Profile settings, one starter per line. After a save it shows the starters kept, and a note under Save lists each line left out with its reason. A failed save shows an ErrorNotice with a No hand-off comment naming the unsaved draft (errors-use-error-notice). The box reports its unsaved draft and a save in flight to the Profile page's leave guards, the ones Goals and Export use: Back, Escape, close and a route leave ask before they drop a draft, and wait while a save runs. The box judges its draft against the list the gateway last confirmed (the first load, a roster read with a new list, or its own save answer), so stale roster props after a save never undo it; a changed list from elsewhere keeps the draft and says so.
  • security_posture.py: member_starters.py joins NON_EGRESS_REDACTION_MODULES; it returns no redacted text.
  • Catalogs and docs: the Profile strings and one chat label in the twelve catalogs (each key a whole sentence), pluralKeys.json, docs/crew-members.md and the docs/feature-map/README.md row.

Rounds: r1 to r10 layout, save answer, reason codes, mask, rebases, failed-save state, Back guard; r11 a draft survives a refresh; r12 a save survives stale roster props.

Backwards compatibility

A Bot with no starters key reads as an empty list, so starters are off by default and its empty chat is drawn as before. A save that does not name starters answers exactly as on main (a test passes on main). A full config save writes "starters": [] per Bot, like every other Bot field.

Tests

  • Before: on main, 14 of the 23 tests then in test_member_starters.py and two in ChatPane.crewmate.test.tsx fail. Each later fix has a test that failed on the previous source; r11's three refresh cases and r12's stale-props case in CrewStartersGroup.test.tsx fail on the old head.
  • After, on this head: test_member_starters.py, test_config_agent_fields_redaction.py, the two roster contract tests, test_config_baseline.py, test_facade_mirror_guards.py and test_security_posture.py, and the seven Static Ratchet Gate files give 429 passed. In website/, the members pages and ChatPane.crewmate.test.tsx give 1,264 passed; the i18n suites give 1,261 passed and npm run i18n:check exits 0. eslint, tsc, black, flake8, isort, mypy, the ratchet gates and the lane CI checks pass.
  • Reproduce: python -m pytest test/test_member_starters.py -n0; in website/: npx vitest run src/test/ChatPane.crewmate.test.tsx src/pages/members/CrewStartersGroup.test.tsx src/pages/members/MembersPage.schedules.test.tsx
  • Proof: PROVEN test/test_config_agent_fields_redaction.py::test_every_list_of_text_field_is_masked_item_by_item, test/test_member_starters.py::TestUpdateEndpoint::test_a_save_stores_what_passes_and_names_what_does_not, test/test_config_baseline.py, website/src/test/ChatPane.crewmate.test.tsx, website/src/pages/members/CrewStartersGroup.test.tsx, website/src/pages/members/MembersPage.schedules.test.tsx (head caa63a912052)

Manual verification

Real components on loopback. The failed-save pair keeps the draft under "Could not save the starters. Try again."; the last pair shows the note when the saved list changed elsewhere.

Four starters, light
Four starters, dark
No starters, light
No starters, dark
Box edited, light
Box edited, dark
After save, left-out note, light
After save, left-out note, dark
Failed save notice, light
Failed save notice, dark
Changed elsewhere note, light
Changed elsewhere note, dark

Sibling sweep

Sibling: members_parts/self_edit.py -- not affected -- starters is not in SELF_EDITABLE_FIELDS
Sibling: crewmate_bundle.py, crewmate_bundle_export.py -- not affected -- no starters in the share file
Sibling: website/src/pages/chat/page/ChatEmptyState.tsx -- not affected -- the main chat's empty state is not a Bot's

Pattern harvest

Rule candidate: review-prompt
Pattern: owner-written text that a button later sends is bounded and cleaned where it enters, dropped with a warning rather than refused, and named in a warning by its place, never quoted.

Related Issues

None.

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

@javenciu
javenciu requested review from a team and KiroCrewCaptain as code owners November 14, 2026 15:16
@javenciu
javenciu enabled auto-merge (squash) November 14, 2026 15:16
@github-actions github-actions Bot added the readiness: checking Automated validation is still running label Nov 14, 2026
@github-actions

github-actions Bot commented Nov 14, 2026 •

Copy link
Copy Markdown
Contributor

Design Review (Opus 5.5) — ✅ PASS

Design-level review of 3ab9354166d006c63cdba0ffde917ccb75e52f1b — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS does not.

Design-Verdict: PASS

Nothing blocks: the field is additive, defaults to empty, a revert walks it back cleanly, and screenshots of this revision are attached.

[DESIGN-REVIEWED] 3ab9354

Description read

[DESCRIPTION-READ] c474780093ee2ccef356f59738c2e024c0ac2f7f10455764e91559cd6a08573b

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 Nov 14, 2026 •

Copy link
Copy Markdown
Contributor

First Principles Review (Opus 5.5) — ✅ PASS

Premise-level review of 3ab9354166d006c63cdba0ffde917ccb75e52f1b — why this exists and whether the shipped surface is the smallest honest version. Updated in place on each push. A BLOCK verdict blocks PR readiness; PASS does not.

First-Principles-Verdict: PASS

Starters are empty by default, so nobody's chat changes unless the owner sets them. Each new piece has its own job. Only a duplicated drop-filter is worth trimming.

Notes

  • Item 4 — duplicate of core_config_view._mask_agent_free_text_list: the roster's inline [s for s in … if _roster_mask(s) == s] is the same filter as [item for item in value if _mask_agent_free_text(item) == item], written twice (grepped == s\]|== item\] in the two files: 1 site each). Subtraction: delete the inline comprehension in members_roster.py and call the one helper, so the starters have one masking rule.

What this change ships

Inventory (8 items) — 7 justified

Intent: a Bot's owner pins a few prompts so a newcomer can start the Bot's empty chat with one press. This is an ADDITION.

  1. The Bot's empty chat shows up to four starter buttons, and a press sends one — justified
  2. Profile → Settings gets a "Conversation starters" box, one starter per line — justified
  3. After a save, the box shows what was kept and lists each line left out with its reason — justified
  4. The roster row and the config view leave out a starter that looks like a credential — duplicate of core_config_view._mask_agent_free_text_list
  5. New config key agents.<name>.starters, empty by default; a bad hand-edited value is dropped with a log warning — justified
  6. PUT /api/agents/{name} accepts starters and answers with starters_left_out; a non-list is a warning, never a 400 — justified
  7. Back, Escape, close and route leave now ask before dropping an unsaved starters draft — justified
  8. member_starters.py is added to NON_EGRESS_REDACTION_MODULES — justified

[FIRST-PRINCIPLES-REVIEWED] 3ab9354

Description read

[DESCRIPTION-READ] 32acc3897378655cbdeb0cf01f1e7a45aa8adaa683eb5fc0c215fbdbb2d54444

That sha256 names the PR description this verdict read. Recompute it with .github/scripts/pr-description-capture.sh; if it differs, the description changed after this verdict was formed and any finding drawn from the description is unproven.

@github-actions

github-actions Bot commented Nov 14, 2026 •

Copy link
Copy Markdown
Contributor

UX Review (Opus 5.5) — ✅ PASS

UX-level review of 3ab9354166d006c63cdba0ffde917ccb75e52f1b — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS does not.

UX-Verdict: PASS

No primary control was misread. The blind reader correctly identified the starter buttons, the starters box and Save, and every new state has a screenshot.

Notes

  • The starter buttons in CrewmateStarters.tsx have no visible heading, so the reader was left guessing whether a press sends right away ("Nothing on this screen says which"). Add a short visible caption above the stack, such as "Tap to send". The group's aria-label is not visible.
  • After a save, onSuccess replaces the text in the box with only the kept starters, so a line dropped as "Too long" disappears and has to be retyped. Keep the dropped lines in the textarea, or give each one a "Restore to edit" action.
  • profile_starters_hint says "A press sends it…". Change it to "Each one is sent as {{name}}'s first message when pressed."

[UX-REVIEWED] 3ab9354

Description read

[DESCRIPTION-READ] 8e3b6e0fb02a5b2383ba953c1c789d04f4f5030fb832150be9e36e44811bed73

That sha256 names everything this verdict read: the PR description, plus the 15 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 Nov 14, 2026 •

Copy link
Copy Markdown
Contributor

Opus 5.5 Review — ✅ no blocking findings

Reviewed 3ab9354166d006c63cdba0ffde917ccb75e52f1b — this comment is updated in place on each push.

Review details

No findings.

[OPUS-REVIEWED] 3ab9354

Verdict parsed from the review's SHA-scoped output markers for commit 3ab9354166d006c63cdba0ffde917ccb75e52f1b.

False positive or not applicable? A repository writer can comment:
/ai-review override fable 3ab9354166d006c63cdba0ffde917ccb75e52f1b false-positive: <reason, 20 to 500 characters>

@github-actions

github-actions Bot commented Nov 14, 2026 •

Copy link
Copy Markdown
Contributor

GPT 6.1 Review — ✅ no blocking findings

GPT 6.1 completed its review of 3ab9354166d006c63cdba0ffde917ccb75e52f1b and found no blocking issues.

This comment is updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] 3ab9354

False positive or not applicable? A repository writer can comment:
/ai-review override gpt 3ab9354166d006c63cdba0ffde917ccb75e52f1b false-positive: <reason, 20 to 500 characters>

@KiroCrewCaptain

KiroCrewCaptain commented Nov 14, 2026 •

Copy link
Copy Markdown
Collaborator

Code change approver recommendation: A person should decide

Why: the decider thinks the change is probably fine but estimates a 14% chance that this change introduces a defect, above the 12% limit for approving on its own.

Adds an owner-only agents.<name>.starters list that appears as buttons on a Bot's empty chat. No blocking defects found. Writes: only the owner's PUT /api/agents/{name} and a hand edit of config.json can set starters. The crewmate.update card lets a card touch only six named fields, and starters is not one of them. GET /api/agents leaves starters out, so the crew editor never sends them back. _carries_mask also checks lists. The prune cleanup keeps a row that has starters, because it no longer matches a fresh synced row. Reads: the roster row and the config view check each starter against the mask and leave out any credential-shaped one. No config write path accepts agents.* keys. The teammate and app views carry fixed field lists that leave starters out. A new test fails if a future list[str] Bot field is added without the item-by-item mask. The loader cleans the value without the redactor (enterprise boots load before it exists) and never fails a load. The frontend draft box reports to every leave guard. The profile panel is keyed by Bot name, so a Bot switch remounts the box. The box compares its draft against the last list the gateway confirmed, so stale roster props after

Tier: 2, held for a person: design-doc-gate: D.

An automated recommendation, not a review: a maintainer still decides. It approves on its own only when the estimated chance of a defect is at most 12% and the change has at most 400 changed lines. Decider v3c (217fe607) · commit 2c1c5829ea · 1275 changed lines. If this recommendation is wrong, reply here: replies are how the approver is evaluated.

@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 Nov 14, 2026
@javenciu
javenciu force-pushed the feat/bot-conversation-starters branch from c15bf4a to a8a89ae Compare November 14, 2026 17:24
@javenciu

Copy link
Copy Markdown
Contributor Author

max-two-buttons-per-row at website/src/components/chat/CrewmateStarters.tsx:18 (GPT, span 3488ea7a0d20). Fixed in a8a89ae. The starter buttons now stand one per row in a vertical stack (flex-col, each button full width), so four short starters never sit side by side. website/src/test/ChatPane.crewmate.test.tsx "stand one per row, a stack rather than a row of peers, even when every prompt is short" renders "Help", "Status", "Plan" and "Summarize" and fails against the previous CrewmateStarters.tsx. The screenshots under temp-screenshots/bot-starters/ are re-rendered from the real ChatPane.

errors-use-error-notice at website/src/pages/members/CrewStartersGroup.tsx:68 (GPT, span acb97e76bb38). Fixed in a8a89ae. The hand-off stays off, and a {/* No hand-off: ... */} comment above the notice names the unsaved starter draft it protects, the form website/AGENTS.md prescribes. website/src/pages/members/CrewStartersGroup.test.tsx "offers no hand-off on a failed save, so the unsaved draft stays in the box, and says why in the source" checks three things: the failed-save notice has no "Ask the agent" button, the draft is still in the box, and the declaration sits next to the notice. It fails against the previous CrewStartersGroup.tsx.

UX: no screenshot of the owner's editor. Added in a8a89ae. The description now shows the real Profile settings box in light and dark, once after an edit with Save enabled and once after a save that left a fifth line out, with the "left out" note. This also covers the Design note asking for an image of the box. A press already sends at once, and the box's hint says so ("a press sends that prompt"). The note about naming dropped lines in normal text is not changed in this PR. The note stays the catalog string so it is translated, and the gateway's warnings are English text only.

Code change approver: member_starters.py missing from the redaction-module lists. Fixed in a8a89ae. member_starters.py now appears in NON_EGRESS_REDACTION_MODULES (src/kiro_crew/security_posture.py) as gate-side, because it asks whether a starter would need redaction only to leave the starter out, and it returns no redacted text. test/test_security_posture.py::TestOmissionDetection::test_every_redactor_call_site_is_a_registered_sink_or_allowlisted, which the Static Ratchet Gate ran, fails against the previous security_posture.py and passes now.

@github-actions

github-actions Bot commented Nov 14, 2026 •

Copy link
Copy Markdown
Contributor

Sensitive Change Review

Sensitive files changed:

  • src/kiro_crew/security_posture.py

  • @javenciu posted a complete reasoning template for 3ab9354.

@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 readiness: checking Automated validation is still running labels Nov 14, 2026
@javenciu
javenciu force-pushed the feat/bot-conversation-starters branch from a8a89ae to 74cb03c Compare November 14, 2026 19:15
@javenciu

Copy link
Copy Markdown
Contributor Author

UX: after a partial save, the box shows text that was not saved (span 52dd2a4a9d68). Fixed in 74cb03c. The save's answer now carries the kept starters and starters_left_out, the place of each line left out. On success the box is set to the kept starters, so it resets even when they equal the ones already saved, and Save turns off. The note moved inside the card, right under Save, in the warning color, and lists each line that was left out by its text. The editor already held that text, so it never travels back from the gateway. CrewStartersGroup.test.tsx "after a partial save shows only what was kept, and names each line left out under Save" sends four lines whose kept pair equals the saved pair. It fails on the previous component (1 failed, 5 passed) and passes now. The notes are answered too: the hint says "starters", the chat's button group is labelled "Conversation starters", and the two left-out screenshots show this exact case, re-rendered.

Opus: the feature-map row fills two of six cells (docs/feature-map/README.md:208). Fixed in 74cb03c. The row now has six cells. Reach: /members?member=<name> → Profile → Settings → Conversation starters. Pages: CrewStartersGroup.tsx, CrewProfileSettings.tsx, CrewmateStarters.tsx, ChatPane.tsx. Handlers: member_starters.py, agent_admin/crew_update.py, handlers/members_roster.py, config/loader_pipeline.py. Endpoints: PUT /api/agents/{name} with its answer fields, and GET /api/members.

GPT: saved starters stop an enterprise or LTS start (src/kiro_crew/config/loader_pipeline.py:1031). Fixed in 74cb03c. clean_starters takes check_credentials, and the loader passes False, so a config load never reaches external_text_requires_redaction → redact_via_context → current_context. The save route still runs the check after boot and leaves a credential-shaped line out. The roster row still leaves one out (_roster_mask), so the chat never shows one. test_member_starters.py::TestConfigField::test_the_load_needs_no_platform_context_for_a_saved_starter makes current_context raise PlatformCompositionError, as an unbooted non-standalone host does, and loads a config that has saved starters. On the previous loader it fails with PlatformCompositionError: no installed context, and now it passes. TestCleanStarters::test_without_the_credential_check_it_never_asks_the_redactor pins the switch.

CI: Backend Tests (3.12, 2), test_config_baseline.py::TestCommittedBaselineParity::test_committed_snapshot_matches_generator. Fixed in 74cb03c. The new agents.*.starters key was missing from config-baseline.json, which is now regenerated with scripts/generate_config_baseline.py (849 entries). The test passes.

@javenciu

Copy link
Copy Markdown
Contributor Author

Sensitive change review

  • Head commit: 74cb03c
  • What rule changed: src/kiro_crew/member_starters.py is added to NON_EGRESS_REDACTION_MODULES in security_posture.py.
  • Before -> after (what was allowed/blocked, now): before, the omission scan had no entry for the new module and failed; after, the module is listed as gate-side. It asks whether a conversation starter would need redaction only to leave that starter out, and it returns no redacted text. No redaction sink or rule changed.
  • Worst case if wrong (breaks commands / leaks secret / locks user out): a credential-shaped starter could reach the roster row. That is covered separately: the roster row leaves out any starter _roster_mask would change (test_roster_rows_carry_the_starters_minus_anything_it_would_mask).
  • Evidence checked (which sandbox result proves it): test/test_security_posture.py and test/test_member_starters.py pass on this head, inside 1,277 passed in the touched suites.
  • No regression:
    • Tools still work (which tools/commands were run, result): PUT /api/agents/{name} and GET /api/members route tests in test/test_member_starters.py pass. A save without starters answers exactly as on main.
    • Backward compatible (old config / old data / old callers still work, how checked): a Bot with no starters key loads as an empty list (test_a_record_without_the_key_has_none), and a hand-edited bad value never fails the load.
    • Existing tests (which suites passed on this head): 204 related test files gave 8,316 passed. The 4 failures in test_cli_doctor_config_sections.py also fail on the merge base on this machine.
  • How to undo: revert the commit. The starters key is ignored by older builds.

@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 readiness: checking Automated validation is still running labels Nov 14, 2026
@javenciu
javenciu force-pushed the feat/bot-conversation-starters branch from 74cb03c to eb06a92 Compare November 14, 2026 22:16
@javenciu

Copy link
Copy Markdown
Contributor Author

UX: the partial-save note opens with "Not saved." and lists every rule (span 2756ef9602f9). Fixed in eb06a92. Each value clean_starters leaves out now carries a reason code with its place, and the save's answer sends both (starters_left_out: [{"place": 3, "reason": "too_long"}]). The note under Save opens with what happened ("Saved 4 starters. Each line left out is listed with its reason.", plural-aware in every catalog) and gives each left-out line its own reason in front of the line: "Too long, over 120 characters: ", "Repeated: ", "Over the limit of 4: ", "Looks like a password or key: ". The paragraph that listed every rule is gone. The hint is now the one line the note suggested: "Up to 4, one per line, 120 characters each. A press sends it as the first message in Radar's chat." CrewStartersGroup.test.tsx "after a partial save shows only what was kept, says so first, and gives each line left out its reason" and "words one saved starter in the singular, and lists a line left out for an unknown reason without one" fail on the previous component (2 failed, 5 passed) and pass now. The reason codes are pinned in test_member_starters.py (test_each_value_left_out_is_named_by_its_place_and_reason_and_never_quoted, test_a_save_stores_what_passes_and_names_what_does_not). Copy change in this head: the i18n check rejected the earlier strings because each ended in a colon and the line followed in another key. The summary is now a whole sentence, and each reason is one key that takes the line as {{line}}, so a translator can reorder it; every catalog is updated and npm run i18n:check exits 0. The two left-out screenshots (light and dark) are re-rendered. The left-out pair now saves six lines: one 125 characters long and one repeated, so no demo line reads as an error message.

Opus: GET /api/config/kirocrew returns agents.*.starters unmasked (src/kiro_crew/config/loader_pipeline.py:1034). Fixed in eb06a92. _masked_config_dict now masks starters item by item. Any item _mask_agent_free_text would change is left out of the list. It is not replaced by the sentinel, because a press sends the text, which is the same rule the roster row follows. The field is listed in a new _AGENT_UNTRUSTED_TEXT_LIST_FIELDS. test_every_list_of_text_field_is_masked_item_by_item holds every list[str] field of KiroCrewAgentConfig to that list, as the existing ratchet holds every str field, so the safeguard gap the approver named is closed too. TestConfigEndpointWire::test_get_response_leaves_out_a_credential_shaped_starter loads a hand-edited config with a credential-shaped starter between two plain ones and calls the real route. On the previous core_config_view.py it fails because the value is in the body. Now the body holds the two plain starters in order, and the stored file still holds all three. No route, bind or auth check changed.

@javenciu

Copy link
Copy Markdown
Contributor Author

Sensitive change review

  • Head commit: eb06a92
  • What rule changed: src/kiro_crew/member_starters.py is added to NON_EGRESS_REDACTION_MODULES in security_posture.py. This head also makes GET /api/config/kirocrew leave out any agents.*.starters item the agent free-text mask would change (core_config_view.py).
  • Before -> after (what was allowed/blocked, now): before, the omission scan had no entry for the new module and failed; after, the module is listed as gate-side. It asks whether a conversation starter would need redaction only to leave that starter out, and it returns no redacted text. No redaction sink or rule changed. The config view's masking gains one list field and loses nothing.
  • Worst case if wrong (breaks commands / leaks secret / locks user out): a credential-shaped starter could reach a browser. Both read paths now leave it out: the roster row (test_roster_rows_carry_the_starters_minus_anything_it_would_mask) and the config view (test_get_response_leaves_out_a_credential_shaped_starter). A ratchet (test_every_list_of_text_field_is_masked_item_by_item) fails if a new list-of-text field of the Bot record is not masked.
  • Evidence checked (which sandbox result proves it): test/test_security_posture.py, test/test_member_starters.py and test/test_config_agent_fields_redaction.py pass on this head, inside 1,886 passed in the 35 touched test files, run on the head rebased onto 89c02b16cd.
  • No regression:
    • Tools still work (which tools/commands were run, result): the PUT /api/agents/{name}, GET /api/members and GET /api/config/kirocrew route tests pass. A save without starters answers exactly as on main. The 94 further test files that call the config routes or the agent update route give 4,130 passed.
    • Backward compatible (old config / old data / old callers still work, how checked): a Bot with no starters key loads as an empty list (test_a_record_without_the_key_has_none), and a hand-edited bad value never fails the load.
    • Existing tests (which suites passed on this head): 1,886 passed in the touched files and 4,130 passed in the route callers. Frontend: 80 files and 1,255 tests under src/i18n, and npm run i18n:check exits 0.
  • How to undo: revert the commit. The starters key is ignored by older builds.

@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 Nov 16, 2026
@javenciu
javenciu force-pushed the feat/bot-conversation-starters branch from de3e9ee to 9054744 Compare November 16, 2026 03:57
@javenciu

Copy link
Copy Markdown
Contributor Author

Rebased onto current main (65833af152) as 905474457d. The conflicts were in config-baseline.json, config/loader_pipeline.py, config/sections_session.py, dashboard/handlers/members_roster.py, test/test_members_roster_relay_teammate.py and website/src/api/client/agents.ts. In each, main's new show_sessions_in_list field stays and this PR's starters sits beside it, in the loader, the Bot record, the roster row, the roster contract list and the client type. The baseline was regenerated with scripts/generate_config_baseline.py and gains only the two agents.*.starters entries. The 260 test files that touch the changed modules pass (11,588), apart from four test_cli_doctor_config_sections.py cases that fail identically on plain main.

Settings Back silently discarding unsaved starters is fixed in 905474457d. The starters box now reports its unsaved draft and a save in flight to the Profile page's existing leave guards, the same ones Goals and Export use, so Back and Escape (through requestProfileBack), closing the card, a crewmate switch and a route leave all ask before dropping a draft, and all refuse while a save runs. Three cases in MembersPage.schedules.test.tsx fail on the old head: Back with a draft asks and keeps the draft on Cancel, Back during a save waits with no question, and a draft arms the reload hold and the route veto. A fourth case, Back with nothing unsaved, passes on both heads and pins main's behaviour. CrewStartersGroup.test.tsx adds one case for the published state.

@javenciu

Copy link
Copy Markdown
Contributor Author

Sensitive change review

  • Head commit: 9054744
  • What rule changed: src/kiro_crew/member_starters.py is added to NON_EGRESS_REDACTION_MODULES in security_posture.py. This PR also makes GET /api/config/kirocrew leave out any agents.*.starters item the agent free-text mask would change (core_config_view.py). The dashboard.handlers.core facade's type-only manifest lists the two config-view names, which binds nothing at run time, and the starter tests use a synthetic GitHub token as their secret. This head is rebased onto 65833af152; the conflicts in the loader, the Bot record, the roster row, the roster contract test, the API client type and config-baseline.json keep every change main made beside this PR's starters line, and the baseline is regenerated with scripts/generate_config_baseline.py. It also makes the Profile Back, Escape, close and route-leave guards ask before a starters draft is dropped and wait while its save runs, which touches no sensitive path.
  • Before -> after (what was allowed/blocked, now): before, the omission scan had no entry for the new module and failed; after, the module is listed as gate-side. It asks whether a conversation starter would need redaction only to leave that starter out, and it returns no redacted text. No redaction sink or rule changed. The config view's masking gains one list field and loses nothing.
  • Worst case if wrong (breaks commands / leaks secret / locks user out): a credential-shaped starter could reach a browser. Both read paths leave it out: the roster row (test_roster_rows_carry_the_starters_minus_anything_it_would_mask) and the config view (test_get_response_leaves_out_a_credential_shaped_starter). A ratchet (test_every_list_of_text_field_is_masked_item_by_item) fails if a new list-of-text field of the Bot record is not masked.
  • Evidence checked (which sandbox result proves it): on this head, the 260 test files that touch the changed modules, with test/test_config_baseline.py, test/test_facade_mirror_guards.py and test/test_sections_facade.py, give 11,588 passed and 3 skipped. The 4 failures in test/test_cli_doctor_config_sections.py fail the same way on plain origin/main.
  • No regression:
    • Tools still work (which tools/commands were run, result): the PUT /api/agents/{name}, GET /api/members and GET /api/config/kirocrew route tests pass. A save without starters answers exactly as on main.
    • Backward compatible (old config / old data / old callers still work, how checked): a Bot with no starters key loads as an empty list (test_a_record_without_the_key_has_none), and a hand-edited bad value never fails the load.
    • Existing tests (which suites passed on this head): the Python run above; in website/, 67 test files (1,258 tests) pass across the members pages and the chat pane, the i18n suites give 1,261 passed, and npm run i18n:check exits 0.
  • How to undo: revert the commit. The starters key is ignored by older builds.

@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 Nov 16, 2026
@javenciu
javenciu force-pushed the feat/bot-conversation-starters branch from 9054744 to 2d650e4 Compare November 16, 2026 05:51
@javenciu

Copy link
Copy Markdown
Contributor Author

The starters box no longer replaces an unsaved draft when the roster refreshes (2d650e4e98, website/src/pages/members/CrewStartersGroup.tsx). The box now remembers the saved list it was last in line with. A refresh that brings the same list leaves the draft alone. A refresh that brings a changed list replaces the box only when nothing was typed. When a draft is open, the draft stays, Save stays on, the Profile leave guards still see it, and a note under Save says the saved list changed elsewhere and that Save replaces it (new key profile_starters_changed_elsewhere in the twelve catalogs). A save from this editor sets the new baseline and clears the note, and the note also goes once the box matches the saved list again. Three cases in CrewStartersGroup.test.tsx fail on the old head: the draft kept with the note when the saved list changes, the baseline and note after this editor's own save, and the note clearing when the draft matches the saved list. Two further cases, the same saved list and a clean box following a changed list, pass on both heads and pin main's behavior. The branch is rebased onto current main (bffb0b6eda) and the Static Ratchet Gate files pass.

@javenciu

Copy link
Copy Markdown
Contributor Author

Sensitive change review

  • Head commit: 2d650e4
  • What rule changed: src/kiro_crew/member_starters.py is added to NON_EGRESS_REDACTION_MODULES in security_posture.py. This PR also makes GET /api/config/kirocrew leave out any agents.*.starters item the agent free-text mask would change (core_config_view.py). The dashboard.handlers.core facade's type-only manifest lists the two config-view names, which binds nothing at run time, and the starter tests use a synthetic GitHub token as their secret. This head is rebased onto bffb0b6eda; the conflicts in the loader, the Bot record, the roster row, the roster contract test, the API client type and config-baseline.json keep every change main made beside this PR's starters line, and the baseline is regenerated with scripts/generate_config_baseline.py. It also makes the Profile Back, Escape, close and route-leave guards ask before a starters draft is dropped and wait while its save runs, and keeps an unsaved starters draft when a roster refresh brings a changed saved list (a note under Save says so). Neither touches a sensitive path.
  • Before -> after (what was allowed/blocked, now): before, the omission scan had no entry for the new module and failed; after, the module is listed as gate-side. It asks whether a conversation starter would need redaction only to leave that starter out, and it returns no redacted text. No redaction sink or rule changed. The config view's masking gains one list field and loses nothing.
  • Worst case if wrong (breaks commands / leaks secret / locks user out): a credential-shaped starter could reach a browser. Both read paths leave it out: the roster row (test_roster_rows_carry_the_starters_minus_anything_it_would_mask) and the config view (test_get_response_leaves_out_a_credential_shaped_starter). A ratchet (test_every_list_of_text_field_is_masked_item_by_item) fails if a new list-of-text field of the Bot record is not masked.
  • Evidence checked (which sandbox result proves it): on this head, test/test_member_starters.py, test/test_config_agent_fields_redaction.py, the roster contract tests, test/test_config_baseline.py, test/test_facade_mirror_guards.py, test/test_sections_facade.py and the seven Static Ratchet Gate files give 429 passed.
  • No regression:
    • Tools still work (which tools/commands were run, result): the PUT /api/agents/{name}, GET /api/members and GET /api/config/kirocrew route tests pass. A save without starters answers exactly as on main.
    • Backward compatible (old config / old data / old callers still work, how checked): a Bot with no starters key loads as an empty list (test_a_record_without_the_key_has_none), and a hand-edited bad value never fails the load.
    • Existing tests (which suites passed on this head): the Python run above; in website/, 67 test files (1,263 tests) pass across the members pages and the chat pane, the i18n suites give 1,261 passed, and npm run i18n:check exits 0.
  • How to undo: revert the commit. The starters key is ignored by older builds.

@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 readiness: checking Automated validation is still running labels Nov 16, 2026
@javenciu
javenciu force-pushed the feat/bot-conversation-starters branch from 2d650e4 to caa63a9 Compare November 16, 2026 07:48
@javenciu

Copy link
Copy Markdown
Contributor Author

Stale roster props undo a partial save (caa63a9120, website/src/pages/members/CrewStartersGroup.tsx). The box now keeps the list the gateway last confirmed (the first load, a roster read that brought a new list, or this editor's own save answer) apart from the roster props. Dirty is judged against that list, and the props last seen are remembered, so a re-render with unchanged props does nothing. A roster list that differs from both is applied only when the box is not dirty (or the draft already equals it); otherwise the draft stays with the "changed elsewhere" note. After saving two identical lines the kept line stays while the props still hold the old list, Save stays off, and the later refetch and a later change from elsewhere are followed. The new case "keeps the list a save kept while the roster props still hold the old list, then follows the refetch" fails on the previous head (the box came back empty after the save) and passes now; the refresh and Back cases still pass (14 tests in the file).

The "changed elsewhere" note had no screenshot. Light and dark captures of the real box with a draft open after another window saved a different list are added to the description (settings-changed-elsewhere-light.png, settings-changed-elsewhere-dark.png).

@javenciu

Copy link
Copy Markdown
Contributor Author

Sensitive change review

  • Head commit: caa63a9
  • What rule changed: src/kiro_crew/member_starters.py is added to NON_EGRESS_REDACTION_MODULES in security_posture.py. This PR also makes GET /api/config/kirocrew leave out any agents.*.starters item the agent free-text mask would change (core_config_view.py). The dashboard.handlers.core facade's type-only manifest lists the two config-view names, which binds nothing at run time, and the starter tests use a synthetic GitHub token as their secret. This head is rebased onto 91aa0e199c; the conflicts in the loader, the Bot record, the roster row, the roster contract test, the API client type and config-baseline.json keep every change main made beside this PR's starters line, and the baseline is regenerated with scripts/generate_config_baseline.py. It also makes the Profile Back, Escape, close and route-leave guards ask before a starters draft is dropped and wait while its save runs, keeps an unsaved starters draft when a roster refresh brings a changed saved list (a note under Save says so), and judges the box against the list the gateway last confirmed, so stale roster props after a save no longer undo it. Neither touches a sensitive path.
  • Before -> after (what was allowed/blocked, now): before, the omission scan had no entry for the new module and failed; after, the module is listed as gate-side. It asks whether a conversation starter would need redaction only to leave that starter out, and it returns no redacted text. No redaction sink or rule changed. The config view's masking gains one list field and loses nothing.
  • Worst case if wrong (breaks commands / leaks secret / locks user out): a credential-shaped starter could reach a browser. Both read paths leave it out: the roster row (test_roster_rows_carry_the_starters_minus_anything_it_would_mask) and the config view (test_get_response_leaves_out_a_credential_shaped_starter). A ratchet (test_every_list_of_text_field_is_masked_item_by_item) fails if a new list-of-text field of the Bot record is not masked.
  • Evidence checked (which sandbox result proves it): on this head, test/test_member_starters.py, test/test_config_agent_fields_redaction.py, the roster contract tests, test/test_config_baseline.py, test/test_facade_mirror_guards.py, test/test_sections_facade.py and the seven Static Ratchet Gate files give 429 passed.
  • No regression:
    • Tools still work (which tools/commands were run, result): the PUT /api/agents/{name}, GET /api/members and GET /api/config/kirocrew route tests pass. A save without starters answers exactly as on main.
    • Backward compatible (old config / old data / old callers still work, how checked): a Bot with no starters key loads as an empty list (test_a_record_without_the_key_has_none), and a hand-edited bad value never fails the load.
    • Existing tests (which suites passed on this head): the Python run above; in website/, 67 test files (1,264 tests) pass across the members pages and the chat pane, the i18n suites give 1,261 passed, and npm run i18n:check exits 0.
  • How to undo: revert the commit. The starters key is ignored by older builds.

@github-actions github-actions Bot added readiness: checking Automated validation is still running readiness: passed Eligible automated validation passed for the current revision merge conflict Branch has merge conflicts with its base — author must resolve before merge and removed readiness: action required A blocking check or review needs attention readiness: checking Automated validation is still running labels Nov 16, 2026
@javenciu

Copy link
Copy Markdown
Contributor Author

Sensitive change review

  • Head commit: 2c1c582
  • What rule changed: src/kiro_crew/member_starters.py is added to NON_EGRESS_REDACTION_MODULES in security_posture.py. This PR also makes GET /api/config/kirocrew leave out any agents.*.starters item the agent free-text mask would change (core_config_view.py). The dashboard.handlers.core facade's type-only manifest lists the two config-view names, which binds nothing at run time, and the starter tests use a synthetic GitHub token as their secret. This head is rebased onto 91aa0e199c; the conflicts in the loader, the Bot record, the roster row, the roster contract test, the API client type and config-baseline.json keep every change main made beside this PR's starters line, and the baseline is regenerated with scripts/generate_config_baseline.py. It also makes the Profile Back, Escape, close and route-leave guards ask before a starters draft is dropped and wait while its save runs, keeps an unsaved starters draft when a roster refresh brings a changed saved list (a note under Save says so), and judges the box against the list the gateway last confirmed, so stale roster props after a save no longer undo it. Neither touches a sensitive path.
  • Before -> after (what was allowed/blocked, now): before, the omission scan had no entry for the new module and failed; after, the module is listed as gate-side. It asks whether a conversation starter would need redaction only to leave that starter out, and it returns no redacted text. No redaction sink or rule changed. The config view's masking gains one list field and loses nothing.
  • Worst case if wrong (breaks commands / leaks secret / locks user out): a credential-shaped starter could reach a browser. Both read paths leave it out: the roster row (test_roster_rows_carry_the_starters_minus_anything_it_would_mask) and the config view (test_get_response_leaves_out_a_credential_shaped_starter). A ratchet (test_every_list_of_text_field_is_masked_item_by_item) fails if a new list-of-text field of the Bot record is not masked.
  • Evidence checked (which sandbox result proves it): on this head, test/test_member_starters.py, test/test_config_agent_fields_redaction.py, the roster contract tests, test/test_config_baseline.py, test/test_facade_mirror_guards.py, test/test_sections_facade.py and the seven Static Ratchet Gate files give 429 passed.
  • No regression:
    • Tools still work (which tools/commands were run, result): the PUT /api/agents/{name}, GET /api/members and GET /api/config/kirocrew route tests pass. A save without starters answers exactly as on main.
    • Backward compatible (old config / old data / old callers still work, how checked): a Bot with no starters key loads as an empty list (test_a_record_without_the_key_has_none), and a hand-edited bad value never fails the load.
    • Existing tests (which suites passed on this head): a rebase onto main that merged one list (website/src/i18n/pluralKeys.json: main's keys plus this PR's one); 533 Python tests pass across the starters, roster, config, config-baseline, security-posture and facade suites; in website/, 84 test files (1,360 tests) pass across the members pages, the chat pane and the i18n suites; the language style suites with the base ref give 126 passed; npm run i18n:check exits 0; lane_ci_checks.sh prints CI-CHECKS PASS.
  • How to undo: revert the commit. The starters key is ignored by older builds.

@javenciu

Copy link
Copy Markdown
Contributor Author

Sensitive change review

  • Head commit: b528a56
  • What rule changed: src/kiro_crew/member_starters.py is added to NON_EGRESS_REDACTION_MODULES in security_posture.py. This PR also makes GET /api/config/kirocrew leave out any agents.*.starters item the agent free-text mask would change (core_config_view.py). The dashboard.handlers.core facade's type-only manifest lists the two config-view names, which binds nothing at run time, and the starter tests use a synthetic GitHub token as their secret. This head is rebased onto 8d4129dfb0; the one conflict, in the roster relay test, takes main's version because main removed the teammate roster that this PR's line extended, and every other file merges cleanly. It also makes the Profile Back, Escape, close and route-leave guards ask before a starters draft is dropped and wait while its save runs, keeps an unsaved starters draft when a roster refresh brings a changed saved list (a note under Save says so), and judges the box against the list the gateway last confirmed, so stale roster props after a save no longer undo it. Neither touches a sensitive path.
  • Before -> after (what was allowed/blocked, now): before, the omission scan had no entry for the new module and failed; after, the module is listed as gate-side. It asks whether a conversation starter would need redaction only to leave that starter out, and it returns no redacted text. No redaction sink or rule changed. The config view's masking gains one list field and loses nothing.
  • Worst case if wrong (breaks commands / leaks secret / locks user out): a credential-shaped starter could reach a browser. Both read paths leave it out: the roster row (test_roster_rows_carry_the_starters_minus_anything_it_would_mask) and the config view (test_get_response_leaves_out_a_credential_shaped_starter). A ratchet (test_every_list_of_text_field_is_masked_item_by_item) fails if a new list-of-text field of the Bot record is not masked.
  • Evidence checked (which sandbox result proves it): on this head, test/test_member_starters.py, the roster relay and roster contract tests, test/test_config_agent_fields_redaction.py, test/test_config_baseline.py, test/test_facade_mirror_guards.py, test/test_sections_facade.py and test/test_security_posture.py give 291 passed, and lane_ci_checks.sh (the seven Static Ratchet Gate files) prints CI-CHECKS PASS.
  • No regression:
    • Tools still work (which tools/commands were run, result): the PUT /api/agents/{name}, GET /api/members and GET /api/config/kirocrew route tests pass. A save without starters answers exactly as on main.
    • Backward compatible (old config / old data / old callers still work, how checked): a Bot with no starters key loads as an empty list (test_a_record_without_the_key_has_none), and a hand-edited bad value never fails the load.
    • Existing tests (which suites passed on this head): the 291 Python tests above and the flake and wait ratchets (125 passed); in website/, 68 test files (1,280 tests) pass across the members pages and the chat pane, 81 files (1,262 tests) across the i18n suites, npm run i18n:check exits 0, eslint and tsc -p tsconfig.app.json exit 0.
  • How to undo: revert the commit. The starters key is ignored by older builds.

@javenciu

Copy link
Copy Markdown
Contributor Author

Sensitive change review

  • Head commit: 099ba8b
  • What rule changed: src/kiro_crew/member_starters.py is added to NON_EGRESS_REDACTION_MODULES in security_posture.py. This PR also makes GET /api/config/kirocrew leave out any agents.*.starters item the agent free-text mask would change (core_config_view.py). The dashboard.handlers.core facade's type-only manifest lists the two config-view names, which binds nothing at run time, and the starter tests use a synthetic GitHub token as their secret. This head is rebased onto 0cc0336a3c; the one conflict, in website/src/i18n/pluralKeys.json, keeps main's new key and adds this PR's one, and every other file merges cleanly (an earlier rebase took main's version of the roster relay test, because main removed the teammate roster that this PR's line extended). It also makes the Profile Back, Escape, close and route-leave guards ask before a starters draft is dropped and wait while its save runs, keeps an unsaved starters draft when a roster refresh brings a changed saved list (a note under Save says so), and judges the box against the list the gateway last confirmed, so stale roster props after a save no longer undo it. Neither touches a sensitive path.
  • Before -> after (what was allowed/blocked, now): before, the omission scan had no entry for the new module and failed; after, the module is listed as gate-side. It asks whether a conversation starter would need redaction only to leave that starter out, and it returns no redacted text. No redaction sink or rule changed. The config view's masking gains one list field and loses nothing.
  • Worst case if wrong (breaks commands / leaks secret / locks user out): a credential-shaped starter could reach a browser. Both read paths leave it out: the roster row (test_roster_rows_carry_the_starters_minus_anything_it_would_mask) and the config view (test_get_response_leaves_out_a_credential_shaped_starter). A ratchet (test_every_list_of_text_field_is_masked_item_by_item) fails if a new list-of-text field of the Bot record is not masked.
  • Evidence checked (which sandbox result proves it): on this head, test/test_member_starters.py, the roster relay and roster contract tests, test/test_config_agent_fields_redaction.py, test/test_config_baseline.py, test/test_facade_mirror_guards.py, test/test_sections_facade.py and test/test_security_posture.py give 291 passed, and lane_ci_checks.sh (the seven Static Ratchet Gate files) prints CI-CHECKS PASS.
  • No regression:
    • Tools still work (which tools/commands were run, result): the PUT /api/agents/{name}, GET /api/members and GET /api/config/kirocrew route tests pass. A save without starters answers exactly as on main.
    • Backward compatible (old config / old data / old callers still work, how checked): a Bot with no starters key loads as an empty list (test_a_record_without_the_key_has_none), and a hand-edited bad value never fails the load.
    • Existing tests (which suites passed on this head): the 291 Python tests above and the flake and wait ratchets (125 passed); in website/, 149 test files (2,543 tests) pass across the members pages, the chat pane and the i18n suites, npm run i18n:check exits 0, eslint and tsc -p tsconfig.app.json exit 0.
  • How to undo: revert the commit. The starters key is ignored by older builds.

A new Bot's chat opens on nothing, and a person has no hint of what it is for.
The owner can now pin up to four one-line prompts to a Bot, and its empty chat
shows them as buttons. Pressing one sends that prompt as the first message
through the same send path as typed text.

The list is stored as agents.<name>.starters and set from the Bot's Profile
settings through the existing agent update route. Each starter is bounded to
120 characters and four per Bot. A value that is blank, too long, repeated,
past the fourth, not text, holds a control or format character, or looks like a
credential is left out, named in the answer's warnings and given a reason the
editor shows beside the line, and the rest are kept; a hand-edited config drops
the same values with a log warning and never fails to load. A credential-shaped value is not quoted in any warning.

The roster row carries the list and leaves out anything the roster would mask,
and the config view leaves out any starter its agent free-text mask would change.
A Bot cannot edit its own starters and the share file does not carry them.
Strings are added to every catalog.
@javenciu

Copy link
Copy Markdown
Contributor Author

Sensitive change review

  • Head commit: 3ab9354
  • What rule changed: src/kiro_crew/member_starters.py is added to NON_EGRESS_REDACTION_MODULES in security_posture.py. This PR also makes GET /api/config/kirocrew leave out any agents.*.starters item the agent free-text mask would change (core_config_view.py). The dashboard.handlers.core facade's type-only manifest lists the two config-view names, which binds nothing at run time, and the starter tests use a synthetic GitHub token as their secret. This head is rebased onto 0f8ad1491e (main with feat(members): open a Bot's sessions in its side panel, drop the Show-in-list switch #37915, which removed the Show-in-list switch and show_sessions_in_list). Five files conflicted. loader_pipeline.py, sections_session.py, members_roster.py and agents.ts keep main's removal and add only this PR's starters lines; config-baseline.json was regenerated with scripts/generate_config_baseline.py (three starters entries, no show_sessions_in_list). Nothing of the removed switch comes back. It also makes the Profile Back, Escape, close and route-leave guards ask before a starters draft is dropped and wait while its save runs, keeps an unsaved starters draft when a roster refresh brings a changed saved list (a note under Save says so), and judges the box against the list the gateway last confirmed, so stale roster props after a save no longer undo it. Neither touches a sensitive path.
  • Before -> after (what was allowed/blocked, now): before, the omission scan had no entry for the new module and failed; after, the module is listed as gate-side. It asks whether a conversation starter would need redaction only to leave that starter out, and it returns no redacted text. No redaction sink or rule changed. The config view's masking gains one list field and loses nothing.
  • Worst case if wrong (breaks commands / leaks secret / locks user out): a credential-shaped starter could reach a browser. Both read paths leave it out: the roster row (test_roster_rows_carry_the_starters_minus_anything_it_would_mask) and the config view (test_get_response_leaves_out_a_credential_shaped_starter). A ratchet (test_every_list_of_text_field_is_masked_item_by_item) fails if a new list-of-text field of the Bot record is not masked.
  • Evidence checked (which sandbox result proves it): on this head, test/test_member_starters.py, the roster relay and roster contract tests, test/test_config_agent_fields_redaction.py, test/test_config_baseline.py, test/test_facade_mirror_guards.py, test/test_sections_facade.py and test/test_security_posture.py and the tests main's feat(members): open a Bot's sessions in its side panel, drop the Show-in-list switch #37915 changed (test_chat_slots_crew_view.py, test_config_live_server_appliers.py, test_crew_show_sessions_setting.py) give 330 passed, and lane_ci_checks.sh (the seven Static Ratchet Gate files) prints CI-CHECKS PASS.
  • No regression:
    • Tools still work (which tools/commands were run, result): the PUT /api/agents/{name}, GET /api/members and GET /api/config/kirocrew route tests pass. A save without starters answers exactly as on main.
    • Backward compatible (old config / old data / old callers still work, how checked): a Bot with no starters key loads as an empty list (test_a_record_without_the_key_has_none), and a hand-edited bad value never fails the load.
    • Existing tests (which suites passed on this head): the 330 Python tests above and the flake and wait ratchets (125 passed); in website/, the members pages, the chat pane and the two files feat(members): open a Bot's sessions in its side panel, drop the Show-in-list switch #37915 changed give 71 files (1,340 tests) and the i18n suites give 82 files (1,269 tests), npm run i18n:check exits 0, eslint and tsc -p tsconfig.app.json exit 0.
  • How to undo: revert the commit. The starters key is ignored by older builds.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

merge conflict Branch has merge conflicts with its base — author must resolve before merge readiness: passed Eligible automated validation passed for the current revision

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants