Repository navigation
feat(chat): write the warm welcome as a plain four-part recap in the gateway - #33805
Conversation
Sensitive Change ReviewSensitive files changed:
|
First Principles Review (Opus 5.5) — ✅ human override accepted@iamwhatever overrode this lane for |
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS Nothing blocks. Moving every fact into gateway code and every judgement into a ledger record removes the false model sentences the audit found. Rendered evidence is attached, and no reader outside the patch uses the removed model-brief path. Notes
[DESIGN-REVIEWED] 7a50589 Description read[DESCRIPTION-READ] 12230faf81c26516dbfac3975360340e45220ae393cb206c4f10e4245d4542c3 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 |
UX Review (Opus 5.5) — ✅ PASSUX-level review of UX-Verdict: PASS Nothing blocks. Screenshots show every changed surface, and the blind reader read the welcome, the preparing bubble and the Board PR tags correctly. Notes
[UX-REVIEWED] 7a50589 Description read[DESCRIPTION-READ] 43ac794bda68ba34ad236f8a2021038de087dee4a1d96db714fd168d3e002dfe That sha256 names everything this verdict read: the PR description, plus the 16 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 |
Opus 5.5 Review — ✅ no blocking findingsReviewed Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
GPT 6.1 Review — ✅ no blocking findings (all downgraded on adjudication)GPT 6.1 flagged blocking issues on This comment is updated in place on each push. Review detailsBLOCKING --
Adjudication (Opus 5.5) — is blocking on each finding proportionate?F1 — LOW. For this to happen, the owner has to dismiss the card in the few milliseconds the gateway spends in threaded memory reads ( False positive or not applicable? A repository writer can comment: |
8f97a35 to
39096ac
Compare
Sensitive change review
|
39096ac to
eb06e66
Compare
Sensitive change review
|
eb06e66 to
54b3850
Compare
54b3850 to
86eb770
Compare
Sensitive change review
|
8fb9e51 to
c4c47a4
Compare
Sensitive change review
|
Sensitive change review
|
Sensitive change review
|
Sensitive change review
|
|
/ai-review override first-principles e6d911e owner-decision: Owner accepts the product-shape finding; RFC change ships in this PR by owner direction. Decision: #33805 (comment) |
Human judgment recorded@iamwhatever recorded a owner-decision release (class
This decision applies only to this commit. A new push requires a new judgment. |
Sensitive change review
|
Sensitive change review
|
Sensitive change review
|
|
UX Review on
|
Sensitive change review
|
Sensitive change review
|
Sensitive change review
|
Sensitive change review
|
Sensitive change review
|
…gateway The warm welcome is now a short, plain recap the gateway writes in the owner's language, with no cards, pictures, widgets or model call: the goal; what needs the owner, each option explained with what it leads to and the pick with its reason; what is being done now with its state and check time; and what comes after. The judgement comes from a new conductor action, work_ledger_record action=ask, which decide and close clear. Each live pull-request read is recorded as object/observed (producer warm_brief) into a new objects slot fold that the Board draws. The preparing hint sits inside the reply bubble in both chat surfaces. RFC rfc-delayed-task-brief.md section 3.11 records the owner's ruling of 2026-11-09. Replaces closed #33032.
Sensitive change review
|
|
/ai-review override first-principles 7a50589 owner-decision: Owner accepts the product-shape finding; RFC change ships in this PR by owner direction. Decision: #33805 (comment) |
Human judgment recorded@iamwhatever recorded a owner-decision release (class
This decision applies only to this commit. A new push requires a new judgment. |
Problem / Motivation
Goal: On return, the warm welcome is a short, plain recap in the owner's language that states only facts gateway code read, and explains what the owner must decide in words.
Owner-directed shape. This shape is the owner's own ruling of 2026-11-09. It replaces the closed #33032, whose cards, live hiding and model framing the owner did not take. The owner directed ONE pull request holding both the RFC change (
docs/request-for-change/rfc-delayed-task-brief.md, new dated section 3.11) and the code. Design and First Principles: the RFC change here is the owner's ruling, not an agent's proposal.A truth audit of 71 welcomes (2026-11-02 to 2026-11-07) found the gateway's own pull-request rows right 34 of 34 times and the model's sentences false 11 of 58 times. The owner asked a follow-up question in 32 of 55 replies.
Why it matters
The welcome is the first thing an owner reads on return. When one model sentence in five is false, the owner re-checks everything and the welcome costs time instead of saving it.
Not a goal
[OPTIONS:]chip row stays the click path, unchanged.What changed (motivation → approach → change)
Root cause: the model wrote prose about facts and checks tried to catch it afterwards. The fix moves every fact to gateway code and every judgement to a ledger record the conductor writes.
dashboard/warm_task_brief_welcome.py,dashboard/warm_task_brief.py). Four parts in order, in the owner's language: Goal (the ledger goal text); Needs you (one plain line when empty; otherwise, per item the Board's ownawaits_persontest says waits, its state, its PR state when read live and "checked HH:MM", then the why, every option with what it leads to, and the pick with its reason; with no record, a plain line saying the reason and options were not written yet; question and change cards and a reply's open choices too); Being done now (one line per started item: ledger status, PR state only when read live, "checked HH:MM"); Comes after (items not started, labelled plan). No model call and no wake turn. The row is an ordinary Bot message (nometa.brief) with main's copied[OPTIONS:]line. A quiet return stays one line. After the reads, the ledger and the waiting list are read again before the post; a change renders again (at most three times).work_ledger_record action=ask). New conductor action withitem_id,what,why,options: [{label, outcome}],pick,pick_why; the store stampswritten_at, refuses a pick outside the labels, anddecide/closeclear it. Its event kind isask, notdecision, so it never reads as an answer. It rideswork/recorded(entry field, baseline, fold), so a rebuild restores it. The goal-conductor skill and the shared conductor prompt say to record it in the same turn a question is relayed to the person.object/observed(producerwarm_brief) in the thread's own unit, no extra GitHub call. A newobjectsslot fold keeps the newest read per subject and producer (64 subjects, 2 KiB facts, out-of-order reads never overwrite newer ones)./api/crew-board(owner-gated, unchanged) addspr_facts; the Crew Board page draws its tags with "read HH:MM".components/chat/PreparingReplyBubble.tsx, lazy-loaded). Raised before the first GitHub read, inside the response bubble in both the Ultra Mode chat (ChatTranscriptView) and the Bot pane (ChatPane), replaced in place by the welcome.warm_task_brief_wake.py), with their tests.RFC: rfc-delayed-task-brief.md is
partialon main; section 3.11 is added by this PR at the owner's direction.Backwards compatibility
Compatible: a stored brief keeps drawing as its card through
WarmBriefCard; an item with noaskis stored and read exactly as before (the field is written only when set);object/observedwidens a closed producer vocabulary and adds no required field; the newaskaction and event kind widen closed vocabularies.Owner ruling (2026-11-12): the
askwrite is its own deferred tool onkirocrew-work,work_ask_record(item_id, what, why, options[{label, outcome}], pick, pick_why), sending the record route's existingaskwrite. It is not inJOB_TOOLS, so it costs no schema tokens until a conductor loads it.work_ledger_recordkeeps main's description and schema, and thetest/test_job_tools.pybudgets (7,600 / 7,800) are unchanged. The goal-conductor skill and the conductor prompt say to load and call it in the turn a question is relayed to the person; the goal conductor's grants include it. The welcome and the Board read the same stored record.Reader: src/kiro_crew/crew_log/projection_work.py:_work_apply -- crew page -- applies the new
askfield and clears it on decide/close (test_work_ledger_projection.py)Reader: src/kiro_crew/work_ledger_records.py:WorkItem.from_dict -- crew page -- reads
askthrough coerce_ask; absent reads as None (test_work_ledger.py)Reader: src/kiro_crew/dashboard/warm_task_brief.py:_run_warm_task_brief -- chat -- the only caller of the removed warm_task_brief_wake module and of the model brief; it now writes the welcome itself (test_warm_task_brief_welcome.py)
Reader: website/src/components/chat/WarmBriefCard.tsx -- chat -- unchanged; new welcome rows carry no meta.brief and draw as an ordinary message (ChatPage.views.test.tsx)
Tests
test/test_warm_task_brief_welcome.py(new): the four parts and their order; the record with every option and the pick; the plain not-written line; a pick outside the labels hidden; PR named only when read, failed read "not read since"; case-folded value words, bad first; a reply's choices not listed twice; control markers, links, images, backticks and a cap inside an address all inert; a key redacted whole before the cut; every language's words and no "conductor"; a card answered during the reads re-renders; a remote worker's binding kept; theobjectsfold andwarm_briefproducer;pr_facts.test/test_work_ledger.py,test/test_work_ledger_projection.py:askstored, refused when malformed, cleared by decide and close, carried through a rebuild.test/test_warm_task_brief*.py: adapted to the gateway welcome (render seamtest/warm_brief_render_seam.py); model-path tests removed with the model path.prFactTags.test.ts,CrewBoardPage.focus.test.tsx,ChatPane.preparing.test.tsx,ChatPage.views.test.tsx.tsc, eslint, black, mypy. Local GPT 6.1 and Opus reviews ran once; their blocking findings (a cap inside an address, a ledger backtick, the remote worker binding, no re-read before post) are fixed in this commit.Proof: not provable -- other: feature change, not a fix; the tests pin new behaviour rather than a reverted defect
Manual verification
Captured against the capture harness (members page) at 1280 and 360 px with every frame asserted. The two "after" texts were rendered by the gateway's own
welcome.renderand_clean. Not run against a live gateway.Screenshots / video
Before (main): the model-written brief card
Preparing hint inside the reply bubble
After: one item needs you (why, options with outcomes, pick; chip row unchanged)
After: nothing needs you
360 px
After: a welcome that needs the owner, with the reply line just above the chips ("点下面的选项,就是发出你的回复。", committed under
temp-screenshots/warm-welcome/)Crew Board: an item row with the welcome's PR fact tags (one bad tag "has conflicts" in the warn tone, one good tag "checks 12 passed, 0 failed", and a read from an earlier day shown with its date)
When the welcome carries the copied
[OPTIONS:]row, one line just above it says a chip only sends a reply: "Tap an option below to reply." (zh "点下面的选项,就是发出你的回复。"), in all seven welcome languages.Related Issues
Fixes #32844 (the preparing hint now sits in the reply bubble). Refs #25306. Replaces #33032 (closed).
Checklist
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)Contribution License Agreement
Contributions are accepted under the project's Apache-2.0 license; see the Licensing section of CONTRIBUTING.md.