Skip to content

feat(chat): show which skills Jev picked, under the reply - #11994

Merged
iamwhatever merged 1 commit into
mainfrom
feat/decisions-strip-ui
Sep 20, 2026
Merged

iamwhatever merged 1 commit into
mainfrom
feat/decisions-strip-ui

Conversation

@iamwhatever

@iamwhatever iamwhatever commented Sep 19, 2026 •

Copy link
Copy Markdown
Contributor

Problem / Motivation

The Decisions (Jev) preview can pick your skills for you. Nothing shows you what it picked.

A reader flips the switch in Settings › Developer › Feature Previews, and from then on some of their turns load skills Jev chose instead of the skills the shipped word-matching rule would have chosen. The reply looks the same either way. The only record is a line in a log file on the machine running Kiro Crew, so the one question a reader of that reply actually has — did the model get the right skills for this? — has no answer on screen, and no way to say yes or no.

Why it matters

A preview nobody can see is a preview nobody can judge. The switch asks a reader to send their message text off the machine, and gives them nothing back they can weigh against that. Without a per-turn receipt there is also no feedback: the people who can tell a good skill pick from a bad one are reading the reply, and they had no way to say so.

What changed (motivation → approach → change)

Under an assistant reply whose skills Jev chose, there is now one line saying what happened, with two thumbs to rate it.

Collapsed it reads like this, and truncates rather than wrapping:

skills · word matching: brazil, crux-code-reviews · Jev: brazil · (0.81) · saved 3.2K tok

When the two sides picked the same skills there is nothing to compare, so the line prints that set once — same pick · Jev and word match: brazil — naming both sides, because otherwise nothing on an agreed line says Jev was involved at all. When they differ it names each with its own list, because the divergence is the only thing a reader can act on.

Moving part What it does
The line The decision point, who picked what, Jev's own confidence, prompt tokens the narrower set saved
Open it Skills offered, in how many batches, characters of history the question carried, what was left out, which answers the gate refused and with what score, and why the decision failed if it did
Thumbs Right or wrong, POSTed per side, each pair labelled with the side it rates (Jev, Word match). Jev's pair is on the line; the word-matching rule gets its own pair when open, so a reader can say the old rule was the better one
One condition The row carries a record. The Decisions switch is not consulted
flowchart LR
  subgraph after["After"]
    direction LR
    R2["assistant reply"] --> S["skills · who picked what · 0.81 · saved 3.2K tok"]
    S --> T["👍 👎"]
    S --> E["open: candidates · batches · refused · error"]
  end
  subgraph before["Before"]
    direction LR
    R1["assistant reply"] --> L["a line in a log file"]
  end
  class R1 ctx
  class R2 ctx
  class L ctx
  class S added
  class T added
  class E added
  classDef added fill:#DCFCE7,stroke:#16A34A
  classDef changed fill:#FEF3C7,stroke:#D97706
  classDef removed fill:#FEE2E2,stroke:#DC2626,stroke-dasharray:4 3
  classDef ctx fill:#E0F2FE,stroke:#0284C7
Loading

🟩 added · 🟨 changed · 🟥 removed · 🟦 unchanged

A reader of the reply now sees which skills were picked and by which side, and can say whether that was right.

Three choices keep it honest.

It cannot print a check over a real divergence. pages/chat/decisionRecord.ts validates every field and draws nothing for a record it cannot check — including one whose two skill lists are not both readable whole, because two lists broken in different ways would each reduce to empty and then compare equal. Agreement is derived from the two name lists, never read from the wire, so a record claiming the sides agreed while naming different skills cannot hide that behind a check mark.

The contract is written down. docs/system-specs/modules/decisions.md pins the record shape and the feedback wire, and src/test/decisionStripContract.test.ts parses that fixture out of the spec rather than restating it. A spec edit that the reader does not follow reds a test here instead of making the strip quietly never appear — which matters because drawing nothing is also what a release with no record looks like. The producer side is not bound to it: the base stamps meta.decisions_strip from chat_runner.py and build_outcome decides the keys, and that producer asserts no fixture. Three keys the § 8 fixture names -- batches, dropped and error -- have no stamping site at all, and agree is stamped although the fixture drops it. Reconciling the two sides is the next change on this branch, and the reader's tolerance of an absent key is what keeps the gap cosmetic meanwhile.

A receipt is history, so the switch does not hide it. The strip never reads consent. A stamped record already sits on this machine, so drawing it sends nothing, and the switch governs whether a future turn may ask Jev. Gating on it would cost an owner who tried the preview and turned it off the record of which past replies were Jev-picked.

The rest is plumbing. pages/chat/AssistantMessage.tsx validates the record and draws the strip directly under the bubble, above the file chips; pages/ChatPage.tsx and app-sdk/messageRenderers.tsx each pass one property; types/index.ts declares where the record rides; api/client.ts gains the feedback call and its two wire vocabularies; every catalog under website/src/i18n/locales gains the same 21 strings in its own language, and i18n/en.context.json gains one translator note for the three-letter Jev label; src/test holds the suite, the spec-contract test and one guard on file-name casing; scripts/capture-decision-strip.mjs and its package.json entry are the screenshot harness (no dependency added or removed); docs/feature-map/README.md gains the row.

Backwards compatibility

Compatible: every field is optional and a row without the record renders exactly as a row does without one -- which is every turn the seam did not decide, the ordinary case, since the preview is off by default. The strip is additive and adds no required input anywhere.

Tests

website/src/test/DecisionStrip.test.tsx, 27 cases, 98.6% statement and 96.3% branch coverage of the two new files.

  • The reader keeps a full record, and refuses one with no turn id — including the absent field, which is the shipping path.
  • It refuses a record whose skill lists are not both readable whole, and still accepts an empty list, since "no skill applies" is a real answer.
  • Agreement is derived from the lists: a wire flag claiming agreement over two different lists reads as a divergence; order alone is not one.
  • A probability outside 0–1 prints nothing; counts are floored and a negative or non-numeric one reads as 0; unreadable names and refusal entries are dropped while a readable key with an unreadable score is kept.
  • Collapsed names both sides, the score and the saving, and carries none of the expanded detail; agreement prints one list under one check; an empty pick says so rather than showing a blank.
  • Opening it reveals the counts, the refusals and the word-matching rule's own thumbs; the toggle carries aria-expanded, and both thumbs carry a label.
  • A thumb posts the side it belongs to, a second press retracts, a refused post lights nothing and remembers nothing, and an accepted one survives the virtualised row being recycled.
  • The assistant row renders unchanged with no record, renders the strip with one, and renders nothing for a record that fails validation.
  • The receipt draws whether the switch is on, off or unread, and no consent request is made at all.
  • The spec promises neither agree nor ts — nothing reads either — and a key the reader does not know is ignored, so a producer may add fields.
  • Each thumbs pair names its own side (Jev, Word match) in visible text and in a tooltip, and the two labels differ, so a thumb cannot be read as counting for both.
  • The agreed line carries the visible words same pick and names both sides; the counts read Selection rounds, Context sent and Skills cut from the offer rather than mechanism words.

website/src/test/fileNameCasing.test.ts guards a cross-platform trap this PR walked into. A module whose name differs from a sibling only in case resolves to ONE module on macOS and Windows, and TypeScript then refuses every import of both — while a Linux checkout compiles clean, so the first failure is the slowest job in CI. The reader module is named decisionRecord.ts for that reason, and the gate now fails on the pair rather than the desktop matrix. It reaches only files a specifier can name: a test or spec file is collected by glob, never resolved, which is why two of those already pair this way on the default branch with the macOS leg green.

Manual verification

Captured from the real built SPA through scripts/capture-decision-strip.mjs (the shared transcript harness: real website/dist, every /api/** call from fixtures, no gateway and no Jev), and the rendered text read back to confirm the copy resolves rather than printing key paths.

Screenshots / video

Diverged and agreed, collapsed:

Collapsed strip, the two sides picked different skills, light
Collapsed strip, the two sides agreed, light

Open, with a refusal and a failed decision:

Expanded strip, light

Dark

Collapsed strip, the two sides picked different skills, dark
Collapsed strip, the two sides agreed, dark
Expanded strip, dark

Related Issues

#11997 -- this is the frontend half of the Decisions (Jev) skills-select seam. The backend half is in the base: a sampled turn's outcome rides its reply as meta.decisions_strip and POST /api/decisions/feedback records a verdict, both from #11993. So the strip draws for a decided turn as soon as this lands, and draws nothing on every turn the seam did not decide, which is the state every test above pins.

Checklist

  • At most two commits (one is the norm), with a Conventional Commits title (feat|fix|docs|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

@iamwhatever
iamwhatever requested review from a team and pepmach September 19, 2026 09:14
@github-actions github-actions Bot added the readiness: checking Automated validation is still running label Sep 19, 2026
@iamwhatever

Copy link
Copy Markdown
Contributor Author

Intent: Give a reader of an assistant reply a receipt for the skill pick behind it — who picked what, how sure, what it saved — and a way to say whether that pick was right, for either side.

Not a goal: the backend that stamps the record or serves the feedback route (a sibling branch); any surface for the aggregate /api/decisions/summary; changing which skills actually load; touching the Settings consent switch.

@github-actions

github-actions Bot commented Sep 19, 2026 •

Copy link
Copy Markdown
Contributor

Design Review (Fable 5) — 🟡 CONCERNS

Design-level review of 54cb7014e17c05ee0e01a43ed7e573d09c616ccd — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

Design-Verdict: CONCERNS

Spec fixture promises batches/dropped/error the producer never stamps, so every real expanded strip prints a false "Selection rounds: 0" today.

Watch

The contract is one-sided: decisionStripContract.test.ts pins the reader to the § 8 fixture, but build_outcome (skills_select.py:636) stamps no batches, dropped or error and asserts no fixture — and DecisionStrip.tsx renders batches unconditionally, so asCount(undefined) prints "Selection rounds: 0" on every production record, a wrong number rather than an absent row. The description calls this gap "cosmetic"; a printed false count is not.
Clears when: the producer stamps the three keys with its own fixture-parity test against § 8, or the strip hides batches (like dropped/error) until it does.

Suggestions

  • A page reload forgets the module-level verdict map, so already-rated thumbs render unpressed and a re-press appends a duplicate feedback row; stamping the last recorded verdict into the record (or a hydrate read) would close it.

[DESIGN-REVIEWED] 54cb701

@github-actions

github-actions Bot commented Sep 19, 2026 •

Copy link
Copy Markdown
Contributor

First Principles Review (Fable 5) — 🟡 CONCERNS

Premise-level review of 54cb7014e17c05ee0e01a43ed7e573d09c616ccd — 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/CONCERNS are advisory.

First-Principles-Verdict: CONCERNS

§ 8 pins a top-level decisions_strip spelling and three record keys (batches, dropped, error) that no producer stamps — contract surface ahead of any writer.

Not justified as shipped

  • Item 6 — zero consumers: decisionStripFieldOf reads msg.decisions_strip ?? msg.meta?.decisions_strip, but every stamping site (chat_runner.py:4087,4288,7964) writes via meta= only — grepped decisions_strip across src/, 0 top-level producers, ever; the sole support is "the split kind above has too", and kind's split had two real producers.
    Subtraction: delete the msg.decisions_strip ?? branch, the decisions_strip field on ChatMessage in types/index.ts, and the spec sentence licensing it — read meta.decisions_strip alone.
    Clears when: a counted producer stamps the top-level spelling, or the branch is deleted.
  • Item 2 — zero consumers (declared, deferred): build_outcome (skills_select.py:651) stamps 9 keys; the § 8 fixture promises batches, dropped, error with no stamping site (count: 0), so the expanded batches Detail prints a permanent false 0 on every real record.
    Subtraction: defer the three keys from the § 8 fixture and drop the batches Detail row until the producer stamps them.
    Clears when: the described "next change on this branch" lands the stamping sites, or the keys are removed from fixture and UI.

What this change ships

Inventory (10 items) — 8 justified

Intent: let a reader of a reply see which skills Jev picked versus the word-matching rule, and rate the pick — an ADDITION.

  1. One-line receipt under a Jev-decided reply naming both sides' picks, confidence, tokens saved — justified
  2. Expanded detail shows candidates, batches, history chars, refusals, error — zero consumers: batches/dropped/error have no stamping site
  3. Thumbs on the line rate Jev's pick, posting per side, second press retracts — justified
  4. A second thumbs pair for the word-matching side when expanded — justified
  5. Receipt draws whether the Decisions switch is on, off or unread — justified
  6. Record also read from the row's top level, not just meta — zero consumers: no producer stamps it there
  7. Spec § 8 contract plus a test parsing the fixture out of the spec — justified
  8. Repo-wide guard against case-only file-name collisions — justified
  9. Screenshot harness verify:decision-strip for both strip shapes — justified
  10. 21 strip strings added to every locale catalog, with translator notes — justified

[FIRST-PRINCIPLES-REVIEWED] 54cb701

@github-actions

github-actions Bot commented Sep 19, 2026 •

Copy link
Copy Markdown
Contributor

Opus 4.8 Review — ✅ no blocking findings

Reviewed 54cb7014e17c05ee0e01a43ed7e573d09c616ccd — this comment is updated in place on each push.

Review details

No findings.

[OPUS-REVIEWED] 54cb701

Verdict parsed from the review's SHA-scoped output markers for commit 54cb7014e17c05ee0e01a43ed7e573d09c616ccd.

False positive or not applicable? A repository writer can comment:
/ai-review override fable 54cb7014e17c05ee0e01a43ed7e573d09c616ccd: <one-sentence reason>

@github-actions

github-actions Bot commented Sep 19, 2026 •

Copy link
Copy Markdown
Contributor

UX Review (Fable 5) — 🟡 CONCERNS

UX-level review of 54cb7014e17c05ee0e01a43ed7e573d09c616ccd — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

UX-Verdict: CONCERNS

Two identical thumb pairs, each labelled with one faint word; the blind reader couldn't tell which pick each rates and "could easily press the wrong one."

Watch

  • Thumb attribution: the pairs are told apart only by muted labels "Jev" / "Word match" (rate_jev, rate_baseline), and the reader — who otherwise parsed the strip well — said "nothing on screen says so plainly, and I could easily press the wrong one", plus "no idea" / "I would not click" for the grey header "Jev". Wrong-side presses corrupt the exact feedback signal this PR exists to collect, on every strip, silently. Smallest fix: make the visible labels name the act ("Rate Jev" / "Rate word match"), still short enough for the truncating line.
  • "Jev" is never anchored on the surface: the reader "ha[d] no idea what or who 'Jev' is" and whether "word match" is a method or a competitor. Owners did opt into "Decisions (Jev)", but that was in Settings, possibly weeks earlier. Fix: a title on jev_label/jev_named tying the name back to the Decisions preview.
  • Verdicts live in a page-session Map ("Thumbs already pressed in this page session", decisionRecord.ts), so after reload every rated thumb renders unlit — a reader can't tell they already answered and may re-rate or assume it was lost. Fix: persist per-turn verdicts (or hydrate them from the server with the record).

Suggestions

  • The pressed-thumb, in-flight, and "Could not save that" / "Decision failed" states appear in no screenshot; capture-decision-strip.mjs already exists — add those states to it.

[UX-REVIEWED] 54cb701

@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 Sep 19, 2026
@github-actions

github-actions Bot commented Sep 19, 2026 •

Copy link
Copy Markdown
Contributor

GPT 5.6 Review — ✅ no blocking findings (all downgraded on adjudication)

GPT 5.6 flagged blocking issues on 54cb7014e17c05ee0e01a43ed7e573d09c616ccd; Opus 4.8 adjudication downgraded every one of them to advisory. Adjudication downgraded all 1 blocking finding(s) to advisory: the remedy each one requires is disproportionate to the harm it prevents. Read them as advice, not as merge conditions.

This comment is updated in place on each push.

Review details

BLOCKING -- website/src/pages/chat/AssistantMessage.tsx:544 -- Variant switching misattributes the decision receipt
{decisionRecord && (
Regenerate reply -> switch to an older variant -> content changes but row metadata does not -> the newer decision strip appears beneath the older reply.
Anchor: feature-map-correctness
Fix: Associate decision metadata with each variant, or hide the strip when displaying a variant without matching metadata.
[BLOCK-MERGE-DOWNGRADED] 54cb701
[GPT-REVIEWED] 54cb701

Adjudication (Opus 4.8) — is blocking on each finding proportionate?

I've confirmed the mechanism. Let me verify the recovery path — that the strip re-reads the prop on every render.

The finding is clear from the code I've opened:

  • Strip renders whenever decisionRecord is truthy (AssistantMessage.tsx:544).
  • decisionRecord derives ONLY from the decisionsStrip prop, which is a single turn-level value independent of the displayed variant (AssistantMessage.tsx:323).
  • Displayed content switches with the variant via effectiveContent/localIdx/onSwitchVariant (AssistantMessage.tsx:198-199), while decisionsStrip does not.

So switching to an older variant changes the reply prose but leaves the decision strip describing the primary (newest) turn.

Weighing:

  • Harm rung: LOW. The strip is an informational receipt ("which skills Jev picked"). A mismatch is a cosmetic misattribution under a rarely-viewed older variant — visible, causes no action or data effect, and is fully reversible: decisionRecord is recomputed from decisionsStrip on every render (:323), so returning to the latest variant restores the correct strip. Condition combination (regenerate to create variants → switch back to an older one) is uncommon.
  • Remedy cost. The real fix is either threading per-variant decision metadata through the variants model across the parent that builds variants and the message shape (new cross-module data plumbing for a receipt that may not exist per-variant), or adding a branch that hides the strip whenever a non-primary variant is displayed. Either exceeds a LOW, self-correcting, purely-informational display glitch.

This is the LOW rung where disproportionate-remedy belongs.

[ADJUDICATION] 54cb7014e17c05ee0e01a43ed7e573d09c616ccd total=1 uphold=0 downgrade=1
DOWNGRADE F1 website/src/pages/chat/AssistantMessage.tsx:544 reason=disproportionate-remedy
[GPT-ADJUDICATED] 54cb7014e17c05ee0e01a43ed7e573d09c616ccd

False positive or not applicable? A repository writer can comment:
/ai-review override gpt 54cb7014e17c05ee0e01a43ed7e573d09c616ccd: <one-sentence reason>

@iamwhatever
iamwhatever force-pushed the feat/decisions-strip-ui branch from 7decd22 to 5d39db4 Compare September 19, 2026 09:37
@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 Sep 19, 2026
@iamwhatever
iamwhatever force-pushed the feat/decisions-strip-ui branch from 5d39db4 to 6d83486 Compare September 19, 2026 10:04
@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 Sep 19, 2026
@iamwhatever

Copy link
Copy Markdown
Contributor Author

span=7804e6a24f0e
self-added: yes
mechanism: VerdictThumbs now owns a required visible label, so no call site can render an unlabelled pair

  • The feedback thumbs — half the PR's stated value — are unlabeled to sighted users

Fixed in 6d834865f0db. The asymmetry was real and it was mine: the baseline pair got a visible label from an external span while Jev's pair got only an aria-label, and IconButton renders neither text nor a tooltip.
The label moved INSIDE VerdictThumbs as a required prop, so the asymmetry cannot come back by adding a third call site: Jev's pair now reads "Rate Jev's pick" on the collapsed line and the baseline pair reads "Rate word matching" when open, both from the same element.
Each button also carries its aria-label string as a title, so the right/wrong meaning is reachable on hover and not only by screen reader.
Pinned by says what each thumbs pair rates, in text a sighted reader can see and carries each thumb’s meaning as a tooltip too. Both labels are in the re-attached screenshots.

@iamwhatever

Copy link
Copy Markdown
Contributor Author

span=e57ec16f1aff
self-added: yes
mechanism: the check's meaning is on a title as well as in sr-only text

  • Agreement is sr-only: the check reads as "something succeeded"

Fixed in 6d834865f0db. The glyph carried its meaning only in sr-only text, which is exactly the misread reported.
The check is now wrapped in a title="Both picked the same skills", so a sighted reader reaches the meaning on hover. The sr-only text stays rather than being replaced: a title is not reliably announced, so removing it would trade one audience for the other.
Visible text was the other option offered and was not taken: the collapsed line is width-constrained and already truncates the skill list, so a sentence there would push the names out of view — the thing the line exists to show.
Pinned by puts the agreement check’s meaning where a sighted reader can reach it.

@iamwhatever

Copy link
Copy Markdown
Contributor Author

span=9e54d236132b
self-added: yes
mechanism: two catalog strings renamed in all 12 locales; no code path changed

  • Expanded labels are mechanism vocabulary

Fixed in 6d834865f0db, taking both of the suggested names.
Batches → Selection rounds. Left out → Skills cut from the offer, which says what the number counts.
On "doesn't say which two": the record carries truncated as a COUNT only, with no names in it — so naming them is not something this surface can do without a wire change. The label now says what was counted instead of implying a list is being withheld.
Both renames are in all 12 catalogs plus the regenerated pseudolocale, and pinned by names the counts in words a reader can act on, not mechanism words, which asserts the old words are absent as well as the new ones present.

@iamwhatever

Copy link
Copy Markdown
Contributor Author

span=4e811c121ed1
self-added: yes
mechanism: one catalog string per locale

  • saved_tokens: "tok" → "tokens"

Fixed in 6d834865f0db. tok appeared in no other product string, so it was jargon this one surface invented.
Now saved {{tokens}} tokens in English, with each locale carrying its own word for a token rather than the English abbreviation.
Pinned by spells the saving in a word the product uses elsewhere, which asserts tokens is present AND that the bare tok is gone.

@iamwhatever

Copy link
Copy Markdown
Contributor Author

span=fd4154fb3ec8
self-added: no
mechanism: new spec section pinning the record + feedback vocabulary, consumed by a frontend contract test

  • The reader's fail-safe is byte-identical to the healthy pre-backend state

Fixed in 6d834865f0db, in the form a frontend-only PR can reach.
docs/system-specs/modules/decisions.md gains a section pinning the decisions_strip record shape and the POST /api/decisions/feedback vocabulary, with the canonical record as a fenced JSON fixture.
website/src/test/decisionStripContract.test.ts parses that fixture OUT of the spec — it does not restate it, because a second copy would be free to drift from the one the gateway author reads — and asserts the reader accepts it field for field, that the spec still promises every key the reader consumes, and that the feedback vocabulary still spells jev/baseline/right/wrong/null.
So a field renamed on either side now reds a test instead of silently making the strip never appear. That closes the "no test on either side goes red" half from this side; the gateway half can consume the same fenced fixture.

@iamwhatever

Copy link
Copy Markdown
Contributor Author

span=320bff29ec2b
self-added: no
mechanism: none — the gate is as the work item specified

  • Gate only the feedback thumbs on live consent, not display of an already-stamped record

Needs a maintainer decision, and the code is unchanged meanwhile.
The observation is sound and I am not rebutting it: a receipt records what happened, so hiding it when the switch flips off does make past Jev-chosen turns read as baseline-chosen — the invisibility this PR exists to remove, reintroduced for history.
I am not taking it unilaterally because the AND is the behaviour this work item specified ("rendered when message.decisions_strip exists AND the Decisions switch is on"), and splitting the gate is a user-visible change to what a withdrawn consent means. Turning consent off and still seeing per-turn Jev receipts is a reasonable thing for an owner to be surprised by, and that judgment is not mine.
The split is one line if a maintainer wants it: keep the display ungated on consent and pass consentEnabled to VerdictThumbs as a disabled/hidden condition.

@iamwhatever

Copy link
Copy Markdown
Contributor Author

span=0547f77c2a45
self-added: no
mechanism: none

  • Item 1 — zero consumers: no gateway produces decisions_strip

Rebutted as disproportional to this PR's shape, with the fact accepted.
The count is right: grep decisions_strip src/kiro_crew is 0, and the body says so in its own words rather than implying otherwise.
The two halves cannot land as one diff. This work item is the frontend half by construction and the gateway half is a sibling branch (feat/decisions-strip), named in the body; whichever lands second is the one that makes the feature visible. Holding this PR until then does not make the gateway change arrive sooner, and it does make the two reviews serial instead of parallel.
Nothing merged here is reachable-broken meanwhile: the absent field is the reader's null path, so every row on main renders exactly as today — pinned by the assistant row renders exactly as today when the row carries no decision record.

@iamwhatever

Copy link
Copy Markdown
Contributor Author

span=59a99a6e1758
self-added: no
mechanism: none

  • Item 2 — zero consumers: the feedback route is registered nowhere

Rebutted as disproportional, with the fact accepted.
Correct that no route is registered. But "every press ships as a failure notice by design" cannot happen before the record exists: the thumbs render only inside a strip, and the strip renders only for a validated decisions_strip. No gateway stamps one, so there is no reachable press to fail.
The two arrive together or not at all — the same gateway change that stamps the record registers the route, which is why this item and item 1 share one "Clears when".
The failure path is nonetheless the honest one rather than a silent no-op: a refused press lights nothing, remembers nothing and says so through the shared error surface, pinned by reports a refused verdict instead of pretending it landed.

@iamwhatever

Copy link
Copy Markdown
Contributor Author

span=158c71389c34
self-added: no
mechanism: the wire vocabulary now lives in the owning spec, not only in the client

  • Item 3 — the wire vocabulary is fixed unilaterally with no server to agree with it

Fixed in 6d834865f0db.
This was the actionable one of the three. The vocabulary is no longer only in TypeScript: docs/system-specs/modules/decisions.md now states the record shape and the POST /api/decisions/feedback body, and decisionStripContract.test.ts holds the client to that document.
So the gateway PR conforms to a reviewed spec section rather than to a client nobody read, and a disagreement between the two is a red test on this side instead of a strip that never appears.

@github-actions github-actions Bot added the readiness: passed Eligible automated validation passed for the current revision label Sep 19, 2026
@KiroCrewCaptain
KiroCrewCaptain enabled auto-merge (squash) September 19, 2026 23:04
buluoray
buluoray previously approved these changes Sep 19, 2026

@buluoray buluoray left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Approving at head 44a71339a3f2fafae45152d9181c5e193193637c. No blocking findings. Verified at this head; I did not take the disposition comments on trust.

The inertness the body discloses is true rather than understated: git grep decisions_strip -- src/kiro_crew is 0 and git grep "decisions/feedback" -- src/kiro_crew is 0 at this head. I am not re-litigating the split — the sibling gateway PR #11993 is the producer and I reviewed it in parallel.

The reader cannot draw a wrong statement

This is the property that matters most for a receipt, and it holds. asNames (website/src/pages/chat/decisionRecord.ts:101-105) returns null unless the value is an array whose every element is a non-empty string — it does not filter bad elements out. readDecisionStrip (:144-146) then refuses the whole record on baseline === null || jev === null before building anything, so sameSet (:108-114) is only ever reached with two lists that were read whole. Two differently-broken lists can no longer both collapse to [] and print "same pick · no skills" about a record the reader never understood. The GPT finding at span 11c7a563720a is genuinely closed, and closed in the stronger form (refuse the record) rather than by patching the comparison.

agree is recomputed from the two lists at :166 rather than read off the wire, so a hostile or stale flag cannot put a check mark over a real divergence. p outside 0–1 reads as null (:150).

Injection, theme, i18n

No HTML sink. No dangerouslySetInnerHTML or innerHTML anywhere in DecisionStrip.tsx; every server-supplied string — skill names included — renders as a JSX text child (:213,217,219,252-253,267 via names()/fmtList), which React escapes. The repo's prior markdown-injection class has no path here because there is no markdown or HTML rendering step.

Theme pairs are intact. No hex, rgb(, bg-white, text-black or literal Tailwind palette in the component. Colors are paired tokens only: bg-card text-muted ring-border on the container (:174), text-text on values (:120,195), hover:text-text (:187). forced-colors:border is present, so high-contrast mode keeps the strip's edge. No half-set foreground/background pair.

All 13 locales are key-symmetric. I extracted the added key set from each edited catalog: en.manual, en-XA, and bn de es fr hi it ja ko pt ru zh-CN carry an identical 25-line set (24 leaves under decisionStrip plus the parent). No locale carries a key en.manual lacks, and none is missing one it has. en-XA is genuinely machine-pseudolocalized — accented glyphs with bracket padding, e.g. "agreed_short": "[şàɱè þìçķ ···]" — with {{names}} placeholders preserved, so it was regenerated rather than hand-edited.

The spec contract really binds

decisionStripContract.test.ts:32-39 reads docs/system-specs/modules/decisions.md with readFileSync, locates ## 7. The decision strip's record and feedback, and JSON.parses the first fenced json block — it is not a hand-copied literal, and specFeedbackVocabulary() (:42-47) slices live spec text the same way. I confirmed the spec actually holds what the test looks for (§7 header at line 47, the json fence at 62, POST /api/decisions/feedback at 102, the retract prose at 105-106). A field renamed or respelled in the spec reds :59-73, :83, :125 and :133. So the reader is bound to a reviewed document rather than to itself, which is what the design lane asked for.

One non-blocking note on the ride-along gate

fileNameCasing.test.ts:89-101 walks all of website/src (skipping only node_modules and __snapshots__) and fails on any directory holding a case-only filename or importable-stem collision. It is a repo-wide gate landing inside a feature PR and it will run on every future PR.

The scope risk is real but bounded, and I do not think it should be split out. NOT_A_SPECIFIER_TARGET (:33) exempts .test / .spec / .d.ts pairs — which is why the two such pairs already on main stay green — and caseOnlyStems (:66) fires only when stems differ in case, so Foo.ts beside Foo.tsx is not flagged. The only way it reds an unrelated PR is if that PR introduces a genuine case-only module collision, which is a real macOS/Windows build break rather than a spurious failure. Worth knowing it scans the tree rather than the diff, so the failure it reports may name a file the PR did not touch.

Not merging; that is the maintainer's call. Whichever of this and #11993 lands second is the one that makes the feature visible, so the §7 fixture assertion on the producer side is the thing to confirm at that point.

@github-actions github-actions Bot added the merge conflict Branch has merge conflicts with its base — author must resolve before merge label Sep 19, 2026
@iamwhatever
iamwhatever force-pushed the feat/decisions-strip-ui branch from 44a7133 to e8480fe Compare September 20, 2026 04:50
@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: passed Eligible automated validation passed for the current revision readiness: checking Automated validation is still running merge conflict Branch has merge conflicts with its base — author must resolve before merge labels Sep 20, 2026
@iamwhatever

Copy link
Copy Markdown
Contributor Author

span=2f6082500a24
self-added: yes
mechanism: the producer publishes nine fields; the strip promises four more, and reads truncated as candidates when it counts clipped history rows

  • The receipt promises diagnostics the producer does not emit

Accepted as a real finding, and NOT fixed in this round. This round is the rebase onto the base that first put both halves in one tree, which is what made the mismatch checkable at all; the shape correction is the next round's change, on this PR.
The mechanism is exactly as described, verified against the producer now under it rather than against the finding's own words. build_outcome (src/kiro_crew/decisions/points/skills_select.py:651-661) publishes turn_id, baseline, jev, agree, p, tokens_saved, candidates, history_chars, truncated — and nothing else. So three of the expanded card's rows read a key no producer writes: batches renders 0 because candidate batching was deleted before the backend half merged and no code will ever stamp it, and dropped and error are absent on the publish path, which the reader turns into [] and null (website/src/pages/chat/decisionRecord.ts:160-161,172).
truncated is the sharper half, and it is worse than a missing row: the producer increments it per clipped HISTORY entry (skills_select.py:425,442), the strip labels it "Skills cut from the offer", and both numbers are plausible, so a reader cannot tell the label is wrong. A count that is present and mislabelled is the one failure this surface's own rule forbids — the module header says a record it cannot check renders nothing, because a half-drawn receipt is worse than no receipt.
Two things are true about scope and neither shortens the fix. The spec's fenced record is a floor, not a promise that every field arrives, so the reader's tolerance of absence is correct and stays. But a floor does not license a ROW that prints a default as if it were measured, and it does not license a label that names something the number is not.
Nothing here reaches readiness this round in any case: the base this PR now sits on fails an import-time assertion in src/kiro_crew/sandbox.py:1053 (_DELEGATED_OVERLAP_LEAF_REASONS is missing decisions, member-memory-bindings and subagents), so test/conftest.py cannot import and every backend, E2E and desktop-build lane is red on every open PR, with the Coverage Gate failing closed behind them. The repair is already open as #12193. The field and label correction lands after that, so the round that fixes it is a round whose checks can actually be read.

@iamwhatever

Copy link
Copy Markdown
Contributor Author

span=aa54965f8969
span=b1e5121d3942
self-added: yes
mechanism: none this round — the rebase changed no rendered pixel, and the three unshot states need frames the capture script does not take yet

  • Cannot evaluate: missing screenshots of a pressed thumb and both error states

Accepted as a real evidence gap, and NOT closed in this round. This round is the rebase onto the base that carries the backend half; it changed no component, no label and no style, so the six existing frames still describe the surface exactly as this head renders it.
The three named states are genuinely unshot, and each is reachable from the existing capture: a strip after a vote (variant='active'), a record carrying error, and the failed-POST notice beside the thumbs. Producing them means three more frames from website/scripts/capture-decision-strip.mjs, which is a code change this round deliberately does not make — a rebase whose own diff is provable file-for-file against the base is worth more than a rebase that also grew a script.

  • The bare grey rating labels don't say they label the thumbs

Same mechanism as the item answered at 44a71339a3 under span 4057ffdffd77, recorded again here because the text moved and with it the span. The labels are this lane's own earlier wording, adopted after four rounds on one span; the structural relabel is tracked in #11997 and is not a thing the rebase round decides.

  • Verdicts live in an in-page Map; a reload unlights a cast vote

Same mechanism as the item answered at 44a71339a3 under span 3bde8afe02b6, re-recorded under its new span. Still accepted, still not fixable on this side: the remedy is a server that reports a recorded verdict back on the record, and no route returns one — GET /api/decisions/summary folds a window rather than answering per turn. A client-side store cannot become durable truth by trying harder.

@iamwhatever

Copy link
Copy Markdown
Contributor Author

span=2f6082500a24
self-added: yes
mechanism: correction to my own earlier record — the published row is build_row(point=..., extra=build_outcome(...)), so point and error ARE stamped and only batches and dropped have no producer

  • The receipt promises diagnostics the producer does not emit

Still accepted, and the earlier record of it is corrected here: I counted build_outcome's return as the published record, and it is not. _record_outcome publishes _log.build_row(point=POINT, session_key=..., latency_ms=..., extra=outcome) (src/kiro_crew/decisions/points/skills_select.py:689-710), and build_row writes seven core fields first — ts, point, session, latency_ms, scrubbed, answers, error — then merges the extras flat (src/kiro_crew/decisions/log.py:241-253). So the published record carries sixteen keys, not nine.
That narrows the finding to what is actually true, which matters because the remedy differs per field:
point IS stamped, by build_row — so no strip row is unfed by it and nothing about it needs changing.
error IS stamped, and is always null on a published record, because the publish is conditional on a real answer (skills_select.py:704-710) and a failed decision's error category lives on the gate's own row, which is never published. The key is fed; the row it drives is unreachable. That is a different defect from a missing key and a different fix.
batches and dropped have no producer at all: batching was deleted before the backend half merged, and nothing in src/kiro_crew/decisions/ ever computes refused answers. These two are the real zero-producer rows.
agree IS stamped although the § 8 fixture drops it, which costs nothing: the reader derives agreement from the two lists and ignores keys it does not know, and a test pins exactly that.
truncated is the one genuine mislabel, unchanged from the finding's own words: the producer counts clipped HISTORY entries (skills_select.py:425,442) and the strip calls them skills cut from the offer.
Nothing is fixed in this round, which is the rebase; and the base cannot be read green in any case — test/conftest.py fails to import behind the module-level assert at src/kiro_crew/sandbox.py:1053, so every backend, E2E and build lane is red on every open PR and the Coverage Gate fails closed behind them. A hotfix for that is being opened separately. The shape correction lands in a round whose checks mean something.

@iamwhatever

Copy link
Copy Markdown
Contributor Author

span=c37ae281ac99
span=bd0cd2d078ab
self-added: yes
mechanism: description corrected in place; the count-row change is code and this round posts none

  • The "strip is inert until the backend lands" premise is false

Accepted, and the half of the "Clears when" clause that is prose is done in this round.
The description was written when the producer was a sibling branch; the rebase put that producer in the base, so three passages in it became false and are now corrected: the backwards-compatibility claim ("every row on main, since the backend does not stamp this field yet"), the contract paragraph's "the producer side is not bound yet", and the related-issues line, which now names #11997 and says the backend half is in the base. docs/feature-map/README.md carried the same claim in its row and is corrected in the diff.
One part of the item is narrower than stated, and the correction is worth having on the record because the remedy differs per field. The published record is not build_outcome's return: _record_outcome publishes build_row(point=POINT, ..., extra=build_outcome(...)), and build_row writes ts, point, session, latency_ms, scrubbed, answers, error before merging the extras (src/kiro_crew/decisions/log.py:241-253). So point IS stamped, and error IS stamped and always null, because publishing is conditional on a real answer. batches and dropped are the two keys with no producer anywhere, and agree is stamped while the fixture drops it — harmless, since the reader derives agreement and ignores unknown keys.
truncated is the real defect and it is not a missing row: the producer counts clipped history entries and the strip labels them skills cut from the offer, so the number is present, plausible and wrong.

  • Distinguish "key absent" from a real 0 and omit the row

Accepted as the right shape, NOT made here. This round is the rebase and posts no code; the base also cannot be read green, since test/conftest.py fails to import behind the module-level assert at src/kiro_crew/sandbox.py:1053 and every backend, E2E and build lane is red on every open PR. The count rows and the label land in the round after that, where a check result means something.

@iamwhatever

Copy link
Copy Markdown
Contributor Author

span=3e40ecd07e9f
span=65842f4aa8c7
self-added: yes
mechanism: counts confirmed for batches and dropped and for the top-level spelling; point and error are stamped by build_row, and no subtraction is made in this round

  • Item 3 — zero consumers: the expanded panel's batches, dropped and error rows have zero producers

Accepted for two of the three, corrected for the third, and nothing is subtracted in this round.
batches and dropped: confirmed, and confirmed as permanent rather than pending. Candidate batching and its runoff were deleted before the backend half merged, so batches names a mechanism that no longer exists; dropped has no computation anywhere under src/kiro_crew/decisions/. Two rows, no producer, no producer coming.
error: stamped, and the reason the grep missed it is worth recording. The published record is not build_outcome's return — _record_outcome publishes build_row(point=POINT, ..., extra=build_outcome(...)), and build_row writes ts, point, session, latency_ms, scrubbed, answers, error before merging the extras (src/kiro_crew/decisions/log.py:241-253). So point is stamped too. error is nonetheless always null on a published record, because the publish is conditional on a real answer and a failed decision's category stays on the gate's own unpublished row. The key is fed and the row it drives is unreachable, which is a different defect from an unstamped key and takes a different fix.
The subtraction you name is the right one for batches and dropped. It is code, and this round is the rebase.

  • Item 5 — zero consumers: the top-level decisions_strip spelling

Accepted as an accurate count, NOT subtracted here. grep decisions_strip src/ finds only the meta path (chat_runner.py:4087, and the two finalizers that pass it), so nothing stamps the top level and the branch has no producer today.
Kept for this round on one narrow ground rather than on symmetry: the branch is two ?? operands in one exported reader, pinned by a test of its own, and deleting it also deletes the ChatMessage.decisions_strip field and a spec sentence — a three-file subtraction inside a round whose whole claim is that its diff is provable file-for-file against the base. It goes in the same round as the batches/dropped rows, which is the round after the base can be read green: test/conftest.py fails to import behind the module-level assert at src/kiro_crew/sandbox.py:1053, so every backend, E2E and build lane is currently red on every open PR.

@iamwhatever
iamwhatever force-pushed the feat/decisions-strip-ui branch 2 times, most recently from 49e6f99 to 62a19cf Compare September 20, 2026 06:07
@github-actions

Copy link
Copy Markdown
Contributor

GPT 5.6 Review — 🔴 changes requested (blocking)

GPT 5.6 found at least one blocking issue that must be resolved before merging 62a19cf467def7c6762ec3d38d98481eeee08338. Adjudication upheld 1 of 1 blocking finding(s); those must be resolved before merging.

This comment is updated in place on each push.

BLOCKING -- docs/feature-map/README.md:87 -- The feature map contradicts itself

Rendered by every transcript host, the main chat and the panes alike.
Decision-strip lookup -> this row says it ships -> line 285 says no shipped surface renders it -> mutually exclusive states.
Anchor: feature-map-correctness
Fix: Remove the stale “NO shipped surface” / #11997 text from the developer row.
[BLOCK-MERGE] 62a19cf
[GPT-REVIEWED] 62a19cf

Adjudication (Opus 4.8) — is blocking on each finding proportionate?

Line 87 (Decision strip row) lists the strip as shipped — "Rendered by every transcript host, the main chat and the panes alike" — with real component hosts (pages/chat/DecisionStrip.tsx, added by this PR's feat(chat): show which skills Jev picked commit). Line 285 still asserts "NO shipped surface renders the record or calls the route yet: the strip and its verdict control are tracked in #11997." Both confirmed on disk; they are a direct, mutually exclusive contradiction in the same doc the PR modifies.

The remedy is a one-clause deletion of the stale #11997 / "NO shipped surface" text — trivial cost, no new state, no maintenance load. Because the fix is cheap, it cannot "clearly exceed" the harm, so disproportionate-remedy does not apply; a self-contradicting feature-map entry survives the weighing.

Harm rung: LOW (developer-doc contradiction, visible, self-correcting on code read). Conditions: contradiction confirmed at docs/feature-map/README.md:87 (shipped claim) and :285 ("NO shipped surface"). Real fix: remove the stale clause at :285 — near-zero cost, so no disproportion exists to justify downgrade.

[ADJUDICATION] 62a19cf467def7c6762ec3d38d98481eeee08338 total=1 uphold=1 downgrade=0
UPHOLD F1 docs/feature-map/README.md:87 reason=harm-warrants-remedy
[GPT-ADJUDICATED] 62a19cf467def7c6762ec3d38d98481eeee08338

False positive or not applicable? A repository writer can comment:
/ai-review override gpt 62a19cf467def7c6762ec3d38d98481eeee08338: <one-sentence reason>

A one-line strip under an assistant reply says how that turn's skills were
chosen: which the shipped word-matching rule would have loaded, which Jev
answered, how sure it was, and the prompt tokens the narrower set saved.
Agreement reads as one muted check over one list; a divergence names both.

Opening it adds the question's own shape -- candidates offered, batches,
characters of history carried, what was left out, which answers the gate
refused and with what score -- and a second thumbs pair, so a reader can
rate the word-matching rule and not only Jev.

Every field is validated at the read, and `agree` is derived from the two
name lists rather than trusted, so the strip cannot print a check over a
real divergence. The strip appears only where the record exists AND the
Decisions (Jev) switch is on, read through the Settings card's own consent
query.
@iamwhatever
iamwhatever force-pushed the feat/decisions-strip-ui branch from 62a19cf to 54cb701 Compare September 20, 2026 06:18
@github-actions

Copy link
Copy Markdown
Contributor

GPT 5.6 Review — ⚠️ review incomplete

GPT 5.6 did not produce a complete verdict for 54cb7014e17c05ee0e01a43ed7e573d09c616ccd; inspect the workflow logs and re-run it.

This comment is updated in place on each push.

See the GPT 5.6 Review job logs; this commit has no completed GPT verdict.

False positive or not applicable? A repository writer can comment:
/ai-review override gpt 54cb7014e17c05ee0e01a43ed7e573d09c616ccd: <one-sentence reason>

@iamwhatever
iamwhatever merged commit 7a2a27d into main Sep 20, 2026
119 of 133 checks passed
@iamwhatever
iamwhatever deleted the feat/decisions-strip-ui branch September 20, 2026 07:29
@github-actions github-actions Bot removed the readiness: action required A blocking check or review needs attention label Sep 20, 2026
@github-actions

Copy link
Copy Markdown
Contributor

GPT 5.6 Review — ⚠️ review incomplete

GPT 5.6 did not produce a complete verdict for 54cb7014e17c05ee0e01a43ed7e573d09c616ccd; inspect the workflow logs and re-run it.

This comment is updated in place on each push.

See the GPT 5.6 Review job logs; this commit has no completed GPT verdict.

False positive or not applicable? A repository writer can comment:
/ai-review override gpt 54cb7014e17c05ee0e01a43ed7e573d09c616ccd: <one-sentence reason>

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.

4 participants