Skip to content

feat(settings): fold Agent Harness details behind a summary so the picker comes first - #32849

Merged
bolichen97 merged 1 commit into
mainfrom
ui/agent-harness-levels
Nov 8, 2026
Merged

bolichen97 merged 1 commit into
mainfrom
ui/agent-harness-levels

Conversation

@iamwhatever

@iamwhatever iamwhatever commented Nov 8, 2026 •

Copy link
Copy Markdown
Contributor

Problem / Motivation

Goal: Settings → Agent Harness shows the picker, the status of the checked harness and its Use button first, with everything else one click away.

Today the open harness row holds about 30 lines before the next harness name: the full Kiro sign-in chooser (also when already signed in), the tool-approval line, caveats, security notes, the 7-line capability list, MCP notes and a gateway-start footer. The control the page exists for is buried.

Why it matters

The page is where a person picks which agent new sessions run. When the switch and the harness list are under a wall of reference text, people miss which harness is in use, what is broken, and how to switch.

Not a goal

  • No change to what any line says beyond the shortened intro; no fact leaves the page.
  • No change to the server's card payload, the capability projection, or which notes are security notes.
  • No change to the Subagent Settings / Pre-started Sessions cards.
  • The Bot-harness status line from feat(crew): a Crew bot's harness follows Settings -> Agent Harness when unset #29509 is not part of this PR; its slot under the picker is untouched so it rebases cleanly.

What changed (motivation → approach → change)

Goal → put a summary in front of reference text → three folds, all native <details>/<summary> (one tab stop, Enter/Space toggle, expanded state announced):

  1. About fold (AgentBackendTab.tsx facts): approval, caveats, security notes, capability list, "Other differences", MCP rule/exception, ineffective agent-file settings and the credential note move into one closed fold under the Use button. Its summary is always visible: About Kiro Agent · 7 of 7 features · 2 safety notes.
    • Safety count (safetyCount): server security notes + Claude's own-permissions caveat + the whole-server MCP deny rule + an unverified approval mechanism + each agent-file limit that will not take effect here (denied_tools, tool_allowlist, permission_mode, hooks, hooks_pre_tool_use).
    • Warning marker (runsWithoutAsking): only an unverified approval mechanism or Claude's own pre-approved tools. The summary turns warning-coloured AND says ⚠ Some tools may run without asking, never colour alone. Crew's sandbox standing down and the whole-server MCP deny are counted, not marked: the server sends no host condition for the first (it holds only on Windows and macOS-with-sandbox), and the second keeps the denied tool denied (UX review).
    • Owner decision (Option A of the layered-layout spec): this reverses the earlier "never behind a disclosure" rule; the file header now records the decision and why the summary keeps its intent.
  2. Kiro sign-in fold (KiroSignInCard.tsx compact form): open while the person must sign in (signed out, lapsed, status read failed); closed to Kiro sign-in · Signed in with Google once signed in; re-opens if the sign-in lapses later; the chat deep link (?highlight=) opens it via a one-shot openRequest. Keeps data-setting-key/data-setting-id, an aria-label and an <h3> inside the summary.
  3. "Don't see an agent?" fold for the gateway-start/policy footer.

Intro shortened in all 12 locales: "New sessions use this agent. Open sessions keep the one they started with." New keys (summary, plural safety-note count, warning marker, missing-agent summary) added in all locales with each language's plural categories; pluralKeys.json updated; en-XA regenerated.

capture/settings-agent-harness.tsx takes ?lang= and scripts/capture-agent-harness-levels.mjs produces the screenshots below from real backend_cards.card_payload() output.

Backwards compatibility

Compatible: frontend layout only; no API, config or stored-state change.

Removes nothing: every line still renders, now inside a fold whose summary is always visible; no key, function, route or record is removed.

Tests

  • AgentBackendTab.detailCard.test.tsx: card lines live in the closed About fold and the summary carries the count; security note counted without the marker; credential note counted without the marker; Claude and unverified marked 'Some tools may run without asking'; an ineffective hooks setting counted (plural "2 safety notes"); deny rule/exception inside the fold, counted, not marked; footer under "Don't see an agent?".
  • KiroSignInCard.test.tsx: compact form folds once signed in, user toggle survives re-render, no duplicate provider line when open, heading kept; open when signed out; re-opens on lapse and on a second lapse after the person closed it; open on a failed status read; deep-link request opens it and is reported handled.
  • SettingsPage.agentHarnessSignIn.test.tsx: the deep link lands on the fold and opens it.

Proof: not provable -- frontend test (prove.py runs Python only)

Local: vitest harness + sign-in files 189/189, src/i18n 1212/1212, eslint clean, tsc clean, i18n:check pass, i18n:render with I18N_BASE_REF=origin/main: PASS [vs-base] — no surface got worse. Local GPT and Opus reviews run; all findings fixed in this commit.

Manual verification

Rendered through the real SettingsPage capture entry at 1280px and 375px, en and zh-CN, with real card payloads. Keyboard and screen-reader behaviour come from native <details>/<summary>; not hand-tested with a screen reader.

Screenshots / video

Before (left) / after (right), Kiro Agent in use, signed out:

default 1280 en

Claude Code checked, 375px:

claude 375 en

Signed in (sign-in folds to one line):

signed in 1280 en

All 8 widths/locales and the opened About fold

default 375 en

default 1280 zh-CN

default 375 zh-CN

claude 1280 en

claude 1280 zh-CN

claude 375 zh-CN

about open 375 zh-CN

signed in 375 en

Related Issues

Fixes #32847

Checklist

  • At most two commits (one is the norm), with a Conventional Commits title
  • Existing tests pass and new tests added for new functionality
  • Self-review completed; code follows project style guidelines
  • Documentation updated (if applicable) — file header records the owner decision
  • No secrets, credentials, or internal references in the diff

Contribution License Agreement

Contributions are accepted under the project's Apache-2.0 license.

@iamwhatever
iamwhatever requested review from a team and KiroCrewCaptain as code owners November 8, 2026 02:40
@github-actions github-actions Bot added the readiness: checking Automated validation is still running label Nov 8, 2026
@github-actions

github-actions Bot commented Nov 8, 2026 •

Copy link
Copy Markdown
Contributor

Design Review (Opus 5.5) — ✅ PASS

Design-level review of 928b563aaff6cc4438146fa75fcd588ec45ce3ea — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS does not.

Design-Verdict: PASS

Nothing blocks. Folding the security lines reverses a rule from the spec and the file header, not from a docs/decisions/ entry. The PR names it as an owner decision, and the always-visible count and written warning keep its intent.

Notes

  • The owner reversal ("Option A of the layered-layout spec") is recorded only in the AgentBackendTab.tsx header and harness-onboarding.md. A short docs/decisions/ entry from the maintainer would stop a later review from treating the folded security notes as a regression and re-arguing it.

[DESIGN-REVIEWED] 928b563

Description read

[DESCRIPTION-READ] b070319d65f3f62c90dbb0475fd0cacebf28050ff2aef56f8e12371648801d8f

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 8, 2026 •

Copy link
Copy Markdown
Contributor

First Principles Review (Opus 5.5) — ✅ PASS

Premise-level review of 928b563aaff6cc4438146fa75fcd588ec45ce3ea — 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

Nothing blocks. The reversed "never behind a disclosure" rule is named in the description and attributed to an owner decision tied to #32847. Two riders and three stale comments are below as notes.

Notes

  • Item 8 — rides along: the new script website/scripts/capture-agent-harness-levels.mjs (123 lines) is a second capture script for the same page as website/scripts/capture-13888-agent-harness.mjs (127 lines). Its only consumer is this PR's screenshots (0 references found in *.json, *.mjs, *.ts, *.yml, *.md). Subtraction: drop the script from the diff and keep the screenshots in the PR body.
  • Item 1 — symptom-level: 3 of the comments that still state the rule this PR reverses are left as they were (grep outside (every|any) disclosure): src/kiro_crew/agent_sdk/backend_cards.py:87, src/kiro_crew/agent_sdk/backend_cards.py:344 and website/src/api/client/config.ts:130. They now contradict the rendering. Subtraction: delete the "rendered OUTSIDE any disclosure" clauses in those three places.

What this change ships

Inventory (9 items) — 7 justified

Intent: let a person choose and switch the agent harness without first reading 30 lines of reference text. This is an ADDITION (a layout change).

  1. Approval, security notes, capabilities and MCP notes fold into a closed "About <name>" fold — justified
  2. The fold's summary always shows "N of M features · K safety notes" — justified
  3. A written warning "Some tools may run without asking" for Claude or an unverified approval — justified
  4. The Kiro sign-in folds to one line once signed in, and opens while signed out, lapsed or failed — justified
  5. The chat's sign-in deep link now also opens the sign-in fold — justified
  6. The gateway-start footer folds under "Don't see an agent?" — justified
  7. The intro is shortened in all 12 locales — justified
  8. The capture page takes ?lang=, plus a new screenshot script — rides along
  9. The spec and feature-map rewritten for the reversed visibility rule — symptom-level (3 sibling comments still stale)

[FIRST-PRINCIPLES-REVIEWED] 928b563

Description read

[DESCRIPTION-READ] 64a80442b308574deb894f816a9580627fdf35181f75d9fe843462aaade7fc5d

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 8, 2026 •

Copy link
Copy Markdown
Contributor

UX Review (Opus 5.5) — ✅ PASS

UX-level review of 928b563aaff6cc4438146fa75fcd588ec45ce3ea — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS does not.

UX-Verdict: PASS

Nothing blocks. The blind reader correctly identified all three new folds (About, Kiro sign-in, "Don't see an agent?") and was willing to open each one. Screenshots show both states of the sign-in and About folds.

Notes

  • No screenshot shows "Don't see an agent?" opened. The BEFORE side shows its text, so this is not a gap. Still, add one opened capture so this revision shows the fold's open state.
  • The Kiro Agent summary "2 safety notes" is neutral-coloured, while Claude Code's is orange with ⚠. The reader "can't tell how serious either one is". Name the most serious note in the neutral summary as well, e.g. · 2 safety notes: sandbox off on Windows.
  • In zh-CN, "Claude Code is a separate app you sign into yourself…" still shows in English (shots 08–10). It is on the BEFORE side too, so this PR did not cause it. It now sits inside the About fold, and the fold's summary is translated.

[UX-REVIEWED] 928b563

Description read

[DESCRIPTION-READ] e3b5ae46d24418d31b5f96d987fc89c6acdfc81b9a618dbd7b31b2e7f813b4c7

That sha256 names everything this verdict read: the PR description, plus the 14 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 8, 2026 •

Copy link
Copy Markdown
Contributor

Opus 5.5 Review — ✅ no blocking findings

Reviewed 928b563aaff6cc4438146fa75fcd588ec45ce3ea — this comment is updated in place on each push.

Review details

No findings.

[OPUS-REVIEWED] 928b563

Verdict parsed from the review's SHA-scoped output markers for commit 928b563aaff6cc4438146fa75fcd588ec45ce3ea.

False positive or not applicable? A repository writer can comment:
/ai-review override fable 928b563aaff6cc4438146fa75fcd588ec45ce3ea false-positive: <reason, 20 to 500 characters>

@github-actions

github-actions Bot commented Nov 8, 2026 •

Copy link
Copy Markdown
Contributor

GPT 6.1 Review — ✅ no blocking findings

GPT 6.1 completed its review of 928b563aaff6cc4438146fa75fcd588ec45ce3ea and found no blocking issues.

This comment is updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] 928b563

False positive or not applicable? A repository writer can comment:
/ai-review override gpt 928b563aaff6cc4438146fa75fcd588ec45ce3ea false-positive: <reason, 20 to 500 characters>

@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 8, 2026
@iamwhatever
iamwhatever force-pushed the ui/agent-harness-levels branch from b4b9b49 to 45a399d Compare November 8, 2026 03:36
@iamwhatever

Copy link
Copy Markdown
Contributor Author

UX Review on b4b9b49, addressed in 45a399d:

  • Blocker (Weaker safety on a cost, not a safety gap): fixed. weakerSafety no longer includes the MCP whole-server deny rule; it stays in the safety-note count and inside the fold. Test: states the deny rule in the About fold and counts it, without the weaker-safety marker asserts data-weaker="false". Note: Kiro Agent still shows the marker, from crew_sandbox_stands_down (its own sandbox makes Crew's stand down on Windows/macOS), which is a safety fact rather than a cost.
  • "Weaker safety" wording: kept as is in this PR, by decision, to avoid a 12-locale change; the marker always sits next to the folded lines that explain it. Open to a follow-up rename.
  • Stray "·" at 375px: fixed; each separator is glued with a no-break space to the item before it.
  • "Don't see an agent?" open state: its content is the unchanged gateway-start sentence shown in the before shots.

@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 8, 2026
@iamwhatever
iamwhatever force-pushed the ui/agent-harness-levels branch from 45a399d to e13e0a2 Compare November 8, 2026 06:19
@iamwhatever

Copy link
Copy Markdown
Contributor Author

Opus 5.5 Review BLOCK on 45a399d (feature-map-correctness), fixed in e13e0a2 — docs only, verified against the code:

  • docs/feature-map/README.md agent row: the capability card is now described as one closed About <name> fold with an always-visible summary (N of M features · K safety notes, each part omitted when empty), what K counts, and the Weaker safety marker (unverified approval, Claude, Crew's sandbox standing down; the whole-server tool-off rule is counted, not marked). The "OUTSIDE that disclosure" clause is gone. Also documents the Kiro sign-in fold and the Don't see an agent? fold.
  • docs/system-specs/modules/harness-onboarding.md: the security bucket row (was "rendered OUTSIDE every disclosure"), the routing paragraph (was "never hidden behind a disclosure"), and the "one card, nothing to bury" paragraph a local re-review found, now match the fold.

check_feature_map.py with FEATURE_MAP_BASE_REF=origin/main: pass. A docs/ grep finds no other claim that these lines are always visible.

@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 8, 2026
@iamwhatever

Copy link
Copy Markdown
Contributor Author

UX Review on e13e0a2:

  • Blocker, "Weaker safety" wording / no comparison point: same point as the earlier review's note, answered above. The wording was kept by decision to avoid a 12-locale change. Escalated to the maintainer together with the new part below; no code change on this head.
  • Blocker, new part: crew_sandbox_stands_down marks every host, while its text says Windows, or macOS with the internal sandbox on: acknowledged as a real gap. The server sends this note without a host condition, so gating it per host is a server-side change to backend_cards.py, not a layout change. It is escalated as a product decision, not patched here.
  • Note, lone "·" on phones: the attached shots predate 45a399d, which glues each separator to the item before it with a no-break space. No change.

@iamwhatever
iamwhatever force-pushed the ui/agent-harness-levels branch from e13e0a2 to 980a32e Compare November 8, 2026 07:07
@iamwhatever

Copy link
Copy Markdown
Contributor Author

UX Review BLOCK on e13e0a2, addressed in this push:

  • Marker on the default harness for a host-conditional note: fixed. crew_sandbox_stands_down no longer sets the marker; it stays in the safety-note count and inside the fold. The server sends no host condition for it, and its own text limits it to Windows and macOS-with-sandbox, so a warning would fire where it is false. Kiro Agent and Kiro CLI now show no marker.
  • "Weaker than what?": fixed. The marker now fires only for an unverified approval mechanism or Claude's own pre-approved tools, and it states that fact: ⚠ Some tools may run without asking (all locales, warning colour plus text). Feature map, harness-onboarding spec and the file header updated to match.
  • Lone "·" on phones: the screenshots in the description are recaptured on this head.

@iamwhatever

Copy link
Copy Markdown
Contributor Author

GPT 6.1 BLOCK on 980a32e (narrow-viewport-required, warning clipped at 320px): real, fixed in 82ba98f8b8d5.

The warning span no longer has whitespace-nowrap; the icon is inline and the text wraps like prose. Checked in a 320px render of Claude Code's detail: the warning's right edge stays inside the row in es (239 of 308px), it (274 of 316px) and de (325 of 371px). Test: warns that tools may run without asking only for Claude or an unverified approval now asserts the warning span carries no nowrap.

The other reds on 980a32e were the installation token's primary rate limit (HTTP 403 at 07:08Z) and passed on one rerun at 08:05Z: UX PASS, Opus no blocking findings.

@iamwhatever
iamwhatever force-pushed the ui/agent-harness-levels branch from 980a32e to dcace4a Compare November 8, 2026 08:34
@github-actions github-actions Bot removed the readiness: action required A blocking check or review needs attention label Nov 8, 2026
@github-actions github-actions Bot added the readiness: checking Automated validation is still running label Nov 8, 2026
@iamwhatever

Copy link
Copy Markdown
Contributor Author

Correction to the GPT reply above: the fix commit is dcace4a (current head), not 82ba98f8b8d5.

@github-actions github-actions Bot added readiness: action required A blocking check or review needs attention 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: checking Automated validation is still running readiness: action required A blocking check or review needs attention labels Nov 8, 2026
@iamwhatever
iamwhatever force-pushed the ui/agent-harness-levels branch from dcace4a to 215eab8 Compare November 8, 2026 20:07
@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 merge conflict Branch has merge conflicts with its base — author must resolve before merge readiness: checking Automated validation is still running labels Nov 8, 2026
@iamwhatever
iamwhatever force-pushed the ui/agent-harness-levels branch from 215eab8 to 928b563 Compare November 8, 2026 22:12
@github-actions github-actions Bot added readiness: checking Automated validation is still running readiness: passed Eligible automated validation passed for the current revision and removed readiness: action required A blocking check or review needs attention readiness: checking Automated validation is still running labels Nov 8, 2026

@bolichen97 bolichen97 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 928b563. In Settings, Agent Harness, the open harness row now shows the state block and the Use button first, and the capability card sits in one closed "About " fold below them. The Kiro sign-in section folds too (open while the person must sign in or the status read failed, closed to "Signed in with " once signed in, and re-opened if the sign-in lapses), and the "listed set is fixed at gateway start" note moves under "Don't see an agent?". No fact is removed from the page.

I read the whole diff. The folded safety lines stay findable: the always-visible summary counts them (security notes, Claude's own-permissions caveat, an unverified approval, the whole-server tool-off rule, and each tool-limiting agent-file setting that will not take effect), and it turns warning-coloured with the written words "Some tools may run without asking" for an unverified approval or Claude's own pre-approved tools. The chat's sign-in deep link opens the row and the fold through openRequest, reported handled so a remount does not force it open again. The new strings are in all 13 locale files, with the plural forms each language needs, the plural key is registered in pluralKeys.json, and the tests cover the summary text, the warning marker, the plural count, and the sign-in fold's open, close and re-open paths. The capture script and page are dev helpers. No regex in product code and no new restriction, and all 97 checks pass on this head, including Publish readiness signal after its rerun on the installation API rate limit.

@bolichen97
bolichen97 merged commit 961604f into main Nov 8, 2026
97 of 100 checks passed
@bolichen97
bolichen97 deleted the ui/agent-harness-levels branch November 8, 2026 23:16
@github-actions github-actions Bot removed the readiness: passed Eligible automated validation passed for the current revision label Nov 8, 2026
@LuisBrel LuisBrel mentioned this pull request Nov 8, 2026
5 tasks done
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.

Settings → Agent Harness: too much text before the picker; fold details behind a summary

2 participants