fix(chat): label purpose-less shell pills from the command, not the title

[jingchaodev]

Problem / Motivation

In simplified-tool-names mode the pill label is the agent-authored __tool_use_purpose. kiro-cli's injection of that argument is inconsistent for built-in shell calls (allocated per spawned process — two sessions one minute apart observed at 16/16 vs 0/22 execute calls carrying it). When it is absent, kiro-cli's auto-title for a bash call is a digest of argument fragments, and the pill renders it verbatim:

⊙ --title, --text, Three, 1. ...

#6435's deriveShellSummary does not catch this: it only fires above 200 chars / on multi-line labels, and it parses the TITLE — which in this failure mode is not the command.

Why it matters

Every session that lands on an untreated kiro-cli process renders unreadable step rows for its whole transcript — the reader cannot tell what any shell step did without expanding it. Restored/old sessions are additionally affected (see the isShell fix below).

What changed (motivation → approach → change)

Goal: the pill should be readable from ground truth, not from a title that is unreliable exactly when it is needed, and it should read in the same verb-led register as agent-authored purposes ("Read the full ticket …" beside "Run ledger.py ticket-log").

Approach considered and rejected: LLM back-fill of missing purposes (per-call cost/latency on the hot streaming path for chrome text). Chosen: deterministic re-label from the call's raw input.

• ToolCallLine: when a shell pill would show its raw label (no purpose, or purpose suppressed by the language guard) and the raw input carries a command, label from the command — verbatim when short/single-line, deriveShellSummary digest when flood-length — framed by a new catalog key pages.chat.toolCallLine.run_command ("Run {{cmd}}", translated in all 12 locales + pseudolocale regen).
• deriveShellSummary: steps over wrappers (sudo), variable references ($PY), and interpreters (python3 script.py → script.py), and attaches one subcommand word to the first meaningful name — the observed oncall command derives to ledger.py ticket-log instead of cd. Two existing test pins changed to strictly richer labels (make → make build, docker-compose, tee → docker-compose up, tee).
• commandFromToolInput (new, utils/toolLabel.ts): reads command out of the raw-input JSON.
• Historical rows: the restored-session path hardcoded isShell: false, so no shell derivation ever ran on old transcripts; it now reads the persisted meta.kind === 'execute', mirroring the live branch.

Tests

• toolLabel.test.ts (16): interpreter/varref/wrapper stepping, subcommand attaches to the first meaningful name only, commandFromToolInput parsing and rejection shapes.
• ToolCallLine.test.tsx (41): soup-title pill re-labels from the command (short verbatim, flood-length derived); an agent-authored purpose still wins over the re-label; short verbatim-command titles unchanged.
• Catalog parity suite (717 total across i18n): new key present in all locales incl. regenerated en-XA.

Manual verification

Screenshots below were produced by the committed fixture harness (website/scripts/capture-purposeless-shell-pill.mjs) driving the real built SPA with the exact transcript shape from the reporting session (P503465049 triage). tsc 0 errors, eslint 0 warnings at the CI cap, i18n added-lines gate clean.

Screenshots / video

Dark — an agent-authored purpose (unchanged), a short command re-labeled verbatim, a flood-length command derived to its digest:

Light theme

Related Issues

None open upstream for this; the root cause (inconsistent __tool_use_purpose injection) is filed with the kiro-cli team internally, and this change makes the dashboard robust to it either way.

Pattern harvest

Rule candidate: review-prompt
Pattern: "UI fallback renders an upstream-generated summary verbatim — when the upstream summary is generated from the same missing data that triggered the fallback, derive from ground truth (raw input) instead."

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) — N/A, behavior documented in code comments per repo comment rules
• No secrets, credentials, or internal references in the diff

[dwu96]

👋 Hi! This PR's description is missing some required sections from our PR template. Workflow runs won't be auto-approved until the description is updated.

Missing sections:

• ## Problem / Motivation
• ## Why it matters
• ## What changed
• ## Tests

Please update your PR description to include these sections, then push or re-save the description. The workflows will be approved on the next cycle.

[iamwhatever]

👋 Hi! This PR's description is missing some required sections from our PR template. Workflow runs won't be auto-approved until the description is updated.

Missing sections:

• ## Problem / Motivation
• ## Why it matters
• ## What changed
• ## Tests

Please update your PR description to include these sections, then push or re-save the description. The workflows will be approved on the next cycle.

[jingchaodev]

CI triage note (head 537e2bc97): the one red CI check, Backend Tests (Windows) (1), is base-branch breakage: the same five test_autonudge_stop_auth.py tests fail on the same Windows shard on #8307 (an unrelated backend diff), while this PR touches only website/ and the Linux shards pass the file. No commits touch that test or its subject between this PR's base and main's tip. Will wait it out / rebase once main is green there.

[jingchaodev]

CI triage note (head 16fc50100): backend tests and PR Hygiene are green after rebasing past #8317. The remaining red, Dependency Audit / Audit Production Dependencies, is npm audit timed out after 120s (fail-closed infra guard, no vulnerability named) — currently failing on all 8 open PRs in the repo, so it is a registry/runner outage, not this diff. Will re-trigger once the audit lane recovers.

[github-actions]

Design Review (Fable 5, fork) — ✅ PASS

Design-level review of 947a344965bd08db81a652c829f2fd9528d972e3 via the fork AI-review pipeline — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

All checks complete. The design holds up: real observed harm, deterministic ground-truth re-label at the display layer (the right layer for a display-data problem), the LLM alternative considered and reasonably rejected, root cause filed upstream, no test pinning the old isShell: false as a decision, and both changed test pins preserve their tests' actual subjects (2>&1 handling, bookkeeping-line skipping) with richer labels. The committed screenshots and capture script follow an established repo convention. Two small design-level improvements surfaced.

Design-Verdict: PASS

Ground-truth re-label at the display layer is the right shape; deterministic, opt-scoped, and the restored-session isShell fix mirrors the live branch's existing contract.

Suggestions

• When the pill is re-labeled from the command, set the hover title to the raw command instead of displayLabel: pillLabelTitle still surfaces the argument-fragment soup — the exact text this PR exists to hide — and the adjacent comment ("Hover reveals the verbatim command") becomes false on the new path.
• The cmd !== label guard leaves only bare-command titles untouched; a Running: <cmd>-prefixed title with input present also re-labels to "Run {{cmd}}", wider than the described soup-title case — and the existing Running: git status pin survives only because its fixture omits input, which real sessions carry. Either skip when label === \Running: ${cmd}`or add a test fixture withinput` present so the pinned shape matches production.

[DESIGN-REVIEWED] 947a344

[github-actions]

UX Review (Fable 5, fork) — 🟡 CONCERNS

UX-level review of 947a344965bd08db81a652c829f2fd9528d972e3 via the fork AI-review pipeline — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

UX-Verdict: CONCERNS

Good fix, but the hover tooltip on a re-labeled pill now shows the exact argument-soup this PR exists to remove.

Watch

• Tooltip regression on the new path: pillLabelTitle = pillLabelText === displayLabel ? undefined : displayLabel — the re-label makes pillLabelText ("Run ledger.py ticket-log") differ from displayLabel, but displayLabel in this failure mode is the soup title ("--title, --text, Three, 1. ..."), not the command. A user hovering the derived pill to see the full command gets the unreadable fragment digest instead and must expand the row. Hits every purpose-less flood-length shell pill (the motivating case), every hover — friction, not failure. Fix: when the command re-label fires, set the pill title to cmd, not displayLabel.

Evidence gaps

• The three pill states the PR renders (purpose kept, short command verbatim, flood-length digest) exist only in added screenshots pills-dark.png / pills-light.png, which are binary markers in a fork patch — unopenable here, and no blind read ran on them. Push the branch to this repository to close both.

[UX-REVIEWED] 947a344

[github-actions]

First Principles Review (Fable 5, fork) — 🟡 CONCERNS

Premise-level review of 947a344965bd08db81a652c829f2fd9528d972e3 via the fork AI-review pipeline — 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.

All evidence is gathered and verified against the base tree. Producing the review.

First-Principles-Verdict: CONCERNS

The soup title this PR fixes still renders unfixed at two sibling pickToolLabel fallback sites — the fix covers only the transcript pill.

Not justified as shipped

All inventory items are justified; the concern is coverage, not existence — see Watch.

What this change ships

Intent: make shell tool-step rows readable when kiro-cli omits the agent-authored purpose and its auto-title is argument-fragment soup — a FIX (repro in description; new tests fail on base).

Inventory (6 items)

1. Purpose-less shell pills now read "Run <command>" (verbatim short, digest long) instead of the fragment soup — justified
2. Long shell digests everywhere now name the script behind wrappers/varrefs/interpreters and attach one subcommand (make → make build) — justified
3. Restored/old sessions' shell rows now get shell labeling (base hardcoded isShell: false; persisted kind verified at chat_runner.py:2331) — justified
4. New run_command catalog key in all 13 locale files — justified
5. New exported commandFromToolInput helper, one consumer — justified
6. Committed screenshot harness + two PNGs — justified

Watch

Point patch: grepped pickToolLabel — 2 other fallback sites render the same soup title for a purpose-less shell call: toolStatusLabel.ts:36 (session-row status; WS payload carries no input, so genuinely larger) and ChatInput.tsx:1080 (approval bar, where extractPreview in CollapsibleToolGroup.tsx:71 already reads meta.tool_input.command, so the fix is reachable there). The description ("Every session … renders unreadable step rows") never names this pill-only scope.
Clears when: the approval-bar label re-labels from tool_input.command, or the description states pill-only scope and what is left.

Subtractions

Shrink isBrowseCommand (browseCommand.ts:24-32): its inline JSON-parse-for-command (counted: 2 pre-existing inline spellings, there and CollapsibleToolGroup.tsx:71) is now commandFromToolInput; replace the inline parse with the helper so one spelling remains.

[FIRST-PRINCIPLES-REVIEWED] 947a344

[github-actions]

GPT 5.6 Review (fork) — ✅ no blocking findings

Reviewed 947a344965bd08db81a652c829f2fd9528d972e3 via the fork AI-review pipeline; updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] 947a344

[github-actions]

Opus 4.8 Review (fork) — ✅ no blocking findings

Reviewed 947a344965bd08db81a652c829f2fd9528d972e3 via the fork AI-review pipeline; updated in place on each push.

Review details

No blocking issues; one advisory finding.

FINDING — website/src/pages/chat/ToolCallLine.tsx:742 — for a purpose-less execute row with a flood-length/multiline command, the new re-label branch sets pillLabelText to Run <digest> while the unchanged pillLabelTitle = pillLabelText === displayLabel ? undefined : displayLabel resolves displayLabel to the kiro-cli argument-fragment soup title, so hovering the substitute pill shows --title, --text, … instead of the verbatim command the adjacent invariant promises (command reachable only via expanded details) → Fix: in the re-label branch, carry cmd as the pill title (return it alongside the label) so the flood-length re-labeled pill reveals the verbatim command on hover.

[OPUS-REVIEWED] 947a344

[jingchaodev]

Disposition — GPT validation finding on e5d58a6fc (fixed on bffb3266f): correct — pillLabelText resolves the run_command frame through the catalog at call time, so uiLang belongs in the memo deps; without it a language switch left the pill in the previous language. Added with an exhaustive-deps disable (the linter cannot see through i18nT; same precedent as ActivityViewer/AssistantMessage). Tests 57/57, tsc/eslint/i18n clean.

[bolichen97]

Open PR relationship audit

This is a consolidated, point-in-time code-level audit note. It compares complete merge-base diffs and current/merged code; it does not treat a shared topic as duplication or partial coverage as completion.

Relationship findings

• This PR is PARTIALLY_COVERED with PR #6435. Coverage is explicitly incomplete; this finding is not a completion or closure claim. Recommended action for PR #8294: KEEP. PR #6435 is the merged foundation this PR builds on, not a substitute for it: the merged code covers the flood-length-title case only, and none of the primary's three behaviors (command-sourced label, interpreter/wrapper/varref stepping with subcommand attachment, historical shell-ness) exist in current origin/main. Files: website/src/pages/chat/ToolCallLine.tsx, website/src/utils/toolLabel.ts.

No PR, Issue, label, branch, or review state was changed by the relationship-note portion of this audit.

[bolichen97]

Rebased onto main 41dcadf2c by a maintainer as part of the 2026-09-08 open-PR audit. Old head bffb3266f, new head 947a34496.

Conflicts: none, clean rebase (no manual resolution, no behaviour change).

Gates run locally in an isolated worktree: npx tsc --noEmit -p website/tsconfig.json passed; vitest run src/test/ToolCallLine.test.tsx src/test/toolLabel.test.ts passed (57 tests); the src/i18n suite passed 664/665, with the single failure being navLabels.test.tsx unable to resolve yaml from a stale local node_modules (unrelated to this diff, and CI installs deps fresh).

Please review the rebased branch. Note that a maintainer push makes the maintainer the last pusher, so under this repo's last-push rule a second approver is needed. Reply here if anything looks wrong.