Skip to content

fix: 0.12.1 wave 2a (KAS projection/grant/autoApprove #33228, #33344, #33201) - #33389

Merged
buluoray merged 3 commits into
release/0.12.0from
crew/slack/0.12.1-wave2
Nov 8, 2026
Merged

buluoray merged 3 commits into
release/0.12.0from
crew/slack/0.12.1-wave2

Conversation

@buluoray

@buluoray buluoray commented Nov 8, 2026

Copy link
Copy Markdown
Contributor

Problem / Motivation

Goal: Bring 3 fixes merged on main on 2026-11-08 (between 19:28 and 19:53 UTC) onto release/0.12.0 for 0.12.1, one commit each, in a PR of their own.

Why it matters

Each fix changes behaviour a 0.12.0 user can hit today. On KAS, an agent whose spec still lists kirocrew-computer while Computer Use is off no longer fails every session with a 90 s MCP readiness timeout, and a kirocrew-* server copied from kirocrew.json still starts after an upgrade removes the payload it named. On kiro-cli, an owner-kept MCP autoApprove tool runs without a prompt, matching what SEL records. On the dashboard, the memory retry and "show more" buttons keep focus while busy, and each language places the repo name where its grammar wants it in the working-on line.

Not a goal

  • No fix that is not already merged on main; nothing is changed only here.
  • No version bump, no tag, no CHANGELOG.md change.
  • No feature, CI, test-only, docs-only, refactor, perf or internal dev-tooling change.
  • No row that needs a prerequisite feature; the hand-resolved hunks are listed under What changed.

What changed (motivation → approach → change)

Each main squash commit was cherry-picked onto release/0.12.0 @ 3697e8b0dd, one commit each in main merge order, with the (cherry picked from commit …) trailer. The repository's leak guard requires the public identity, so every commit is authored as Ray Xu and names its original author on an Original author: line.

# PR Original author main release
1 #33344 let each language place the repo name in the working-on line Bolin Chen 74809058bf f5a1b75e33
2 #33201 keep focus on memory retry and show more buttons while busy Bolin Chen d60b8c985f fdb565d539
3 #33228 three projection, grant and mcp autoapprove fixes Bolin Chen 012380ad54 ee9368adf0

Hand-resolved hunks:

The #33228 commit's per-file line counts equal main's for all 40 files.

Backwards compatibility

None: each fix keeps its main PR's compatibility notes; nothing here changes stored data or config beyond what those PRs already shipped on main.

Tests

Proof: not provable -- other: a backport of main commits whose proofs ran on their own PRs

  • CI does not run the backend suite on PRs into release/*, so the related set ran here: the 17 test files fix(kas): three projection, grant and mcp autoapprove fixes #33228 changes plus the KAS session-load, MCP env relay, app MCP session identity, KAS env launch resume and runtime bindings suites, one pytest process with no xdist and TMPDIR outside ~/.kiro/crew: 1,258 passed, 2 failed on this head (ee9368adf0). The 2 failures, in test_folder_steering_provider_dedup.py, fail the same way on the commit before the fix(kas): three projection, grant and mcp autoapprove fixes #33228 pick and on a clean release/0.12.0 @ fd33b0e1d4 (2 failed, 29 passed), so they come from the release branch, not from this batch.
  • Frontend: vitest on the 4 changed test files, 23 passed. check-i18n-strings.mjs and i18n-check.mjs pass. vitest run src/i18n/: 1,194 passed.
  • flake8 passes on the changed Python files.

Manual verification

N/A - each fix was verified on its own main PR. The 0.12.1 RC soak is the end-to-end check.

Screenshots / video

Why no screenshot: the frontend picks carry the same rendered change as their main PRs, which show it.

Related Issues

Backport to the 0.12 line for 0.12.1 of #33344, #33201, #33228.

Sibling sweep

Siblings: none -- each source PR carried its own sibling sweep; this PR only carries those fixes.

Pattern harvest

Not generalizable: this PR only carries fixes already on main. The source PRs have their own harvest.

Checklist

  • At most two commits (one is the norm), with a Conventional Commits title (feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)
  • Existing tests pass and new tests added for new functionality
  • Self-review completed; code follows project style guidelines
  • Documentation updated (if applicable)
  • No secrets, credentials, or internal references in the diff

Contribution License Agreement

Contributions are accepted under the project's Apache-2.0 license; see the Licensing section of CONTRIBUTING.md.

…ne (#33344)

The auto improvement setup panel built its current-repository line by
gluing a translated "Working on" label in front of the repository name,
so every language had to read in English word order. Korean, Japanese,
Hindi and other languages that put the object first could not say it
naturally.

Render the line from one message with a repo slot instead, so each
catalog decides where the name goes, and translate it in every shipped
language. The repository name stays a component child, never a
translated value.

Part of #33103

Original author: Bolin Chen
(cherry picked from commit 7480905)
#33201)

A memory read's Retry, Show more or Load more button lost keyboard focus
the moment it was pressed. Each was natively disabled while its read
ran, which takes it out of the tab order, and each Retry rendered only
while its query held an error, which React Query clears when a query
with nothing cached refetches. So the pressed Retry was swapped for a
loading card or unmounted with its card, and focus fell to the page.

Keep each failure surface up through its retry with queryReadFailed,
and mark the busy buttons with aria-disabled plus a click guard, as the
memory cards fixed in #32186 do. Covers the memory graph Retry, the
vector memory card's Retry and Load more buttons, and the Show more
buttons on record history, replaced experiences and seed memory.

Original author: Bolin Chen
(cherry picked from commit d60b8c9)
- kiro-cli route honours a kept MCP autoApprove list. kiro-cli has no
  autoApprove field, so a verb governance keeps still prompted while
  the SEL said it skips the gate. The native skill view now grants each
  kept verb as an @server/tool ref in allowedTools, through the same
  strip and ceiling the KAS projection applies; no spec is written.
- KAS withholds a gate-closed managed server. A spec Kiro Crew did not
  write could declare kirocrew-computer with Computer Use off, and
  every session failed on the 90 s managed MCP readiness timeout. The
  projection now skips a managed server whose spec gate is closed.
- A kirocrew-* entry a user spec keeps in the block launches the
  managed invocation. A spec copied from kirocrew.json named a payload
  an upgrade removed, so a key such as timeout made the session start
  a missing launcher (ENOENT). The KAS hoist and the kiro-cli skill
  view now swap in the managed command and args and keep other keys.

Fixes #33198

Original author: Bolin Chen
(cherry picked from commit 012380a)
@buluoray
buluoray requested review from a team and KiroCrewCaptain as code owners November 8, 2026 20:12
@buluoray
buluoray requested review from Zedmor and removed request for a team November 8, 2026 20:12
@github-actions

github-actions Bot commented Nov 8, 2026

Copy link
Copy Markdown
Contributor

Sensitive Change Review

Sensitive files changed:

  • src/kiro_crew/security/paths_lists.py

  • No reasoning template for ee9368a yet.

How to clear this check:

  1. The PR author posts the template below as a PR comment, with
    every field filled (N/A needs a reason). Keep the
    Head commit line: it must name the current head SHA.
  2. The check turns green within a minute, with no push.
  3. A new push resets this check: post it again for the new head.

Only the author's comment counts this way. A reviewer can instead
paste the filled template into an "Approve" review on this head.
Merging still needs a reviewer's approval.

## Sensitive change review
- Head commit: ee9368adf047350f8261ff157b3f019a531dd28d
- What rule changed:
- Before -> after (what was allowed/blocked, now):
- Worst case if wrong (breaks commands / leaks secret / locks user out):
- Evidence checked (which sandbox result proves it):
- No regression:
  - Tools still work (which tools/commands were run, result):
  - Backward compatible (old config / old data / old callers still work, how checked):
  - Existing tests (which suites passed on this head):
- How to undo:

@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

UX Review (Opus 5.5) — ✅ PASS (a BLOCK with no Trigger and no Outcome does not block)

UX-level review of ee9368adf047350f8261ff157b3f019a531dd28d — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS does not.

UX-Verdict: PASS

cannot evaluate: missing screenshots. The PR changes the working-on line in every locale and the busy state of the memory Retry and "Show more" buttons, and no image shows either change.

Evidence gaps

  • Auto-improvement setup panel, working-on line (autoImprovement.currentRepoLine). The repo name now sits where each language's grammar puts it, for example de Arbeitet an <repo/> and ja 「<repo/>」で作業中. No screenshot shows it. To close: attach stills of the line in en plus one reordered locale (ja or de).
  • Memory graph and vector memory Retry while a read is running. With nothing cached, Retry now stays on screen at opacity-30 next to a new role="status" loading line, where the loading card used to replace it. No still shows this. To close: attach one still of the failed state and one of Retry pressed and busy, from MemoryGraphTab and VectorMemoryCard.
  • "Show more" / "Load more" busy state. These buttons switch from disabled to aria-disabled styling in MemberMemoryPanel, MemoryRecordsEditor, MemoryRetiredCard, and the episodic and audit cards. No screenshot shows them. To close: attach one still of a busy "Show more" button.
  • Blind read not performed. The PR description has no image attachment and no image is committed under temp-screenshots/ or .github/screenshots/. Attaching the stills above to the PR description closes this gap.

[UX-REVIEWED] ee9368a

Description read

[DESCRIPTION-READ] f048a6fe2658d62739f566d6be1971df904d2da3e45701a3535b05a4ec37a14a

That sha256 names everything this verdict read: the PR description, plus the 3 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

Design Review (Opus 5.5) — ✅ PASS

Design-level review of ee9368adf047350f8261ff157b3f019a531dd28d — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS does not.

Design-Verdict: PASS

Nothing blocks: this backports three fixes already merged on main, and nothing reads the renamed i18n key or the reshaped hoist test outside the patch.

Notes

  • The kiro-cli change widens auto-approval on a stable line: an owner-kept autoApprove verb now becomes an allowedTools grant in the skill view. Because the backend suite does not run on release/* PRs, the 0.12.1 RC soak is the only end-to-end check of that grant path. Name it explicitly in the soak plan.

[DESIGN-REVIEWED] ee9368a

Description read

[DESCRIPTION-READ] 741cecb6104a3d9fb433399d48c3a02664a50078a93ba13d6f368a1a0360bfd8

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 ee9368adf047350f8261ff157b3f019a531dd28d — 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

This is a backport of three fixes already merged on main, one commit each. Both hand-resolved hunks keep only what 0.12 needs, and nothing new is added only on this branch.

What this change ships

Inventory (7 items) — 7 justified

Intent: give 0.12.0 users, in 0.12.1, three bug fixes that already landed on main. FIX.

  1. A KAS agent whose spec lists kirocrew-computer while Computer Use is off no longer fails every session with a 90 s MCP timeout — justified
  2. A kirocrew-* server copied from kirocrew.json still starts after an upgrade removes the payload it named — justified
  3. On kiro-cli, an MCP autoApprove tool the owner kept runs without a prompt, through a skill-view allowedTools grant — justified
  4. The memory retry and "show more" buttons keep focus while busy — justified
  5. Each language places the repo name where its own grammar wants it in the working-on line — justified
  6. Specs and comments now say the grant goes through allowedTools, because kiro-cli has no autoApprove field — justified
  7. The hoist_servers_for_session docstring keeps 0.12's reserved_names paragraph, and preparation.py brings in only _relaunch_reserved_servers — justified

[FIRST-PRINCIPLES-REVIEWED] ee9368a

Description read

[DESCRIPTION-READ] 4b954f658ff7ac42dce458fbaaa1f758b5a54789d1fa1704436803e68b2c34ef

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

GPT 6.1 Review — ✅ no blocking findings

GPT 6.1 completed its review of ee9368adf047350f8261ff157b3f019a531dd28d and found no blocking issues.

This comment is updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] ee9368a

False positive or not applicable? A repository writer can comment:
/ai-review override gpt ee9368adf047350f8261ff157b3f019a531dd28d false-positive: <reason, 20 to 500 characters>

@github-actions

github-actions Bot commented Nov 8, 2026

Copy link
Copy Markdown
Contributor

Security Scope Review -- ✅ PASS

Which legitimate operations does ee9368adf047350f8261ff157b3f019a531dd28d newly refuse? Updated in place on each push. A script-confirmed regression blocks; the model proposes candidates and never decides a refusal.

Verdict basis: adjudicated, zero confirmed regressions.

Security scope review -- adjudicated candidates

  • macos-latest (platform posix): 9 classified, 0 newly refused, 9 still allowed.
  • ubuntu-latest (platform posix): 9 classified, 0 newly refused, 9 still allowed.
  • windows-latest (platform windows): 5 classified, 0 newly refused, 5 still allowed.

No candidate flipped from allowed to refused.

Scope-Verdict: PASS

Reason: The diff changes no shell deny tier. security/paths_lists.py only changes a comment, and nothing in hooks.py, the sensitive-command tier, the exfil shapes or the rule catalog is touched. So no shell command can newly be refused. The real tightenings are a KAS server-projection filter and a per-tool ceiling re-check on skill-view grants. The classifier cannot reach either one, so they are named below as unadjudicated.

TIGHTENING: KAS MCP server projection (kas_agents._project_mcp_servers) -- it stops declaring a managed server whose spec_gate is closed (kirocrew-computer while Computer Use is off or unsupported), so a legacy or hand-written spec cannot launch a backend that advertises no tools and stalls session readiness: + if name in gated_off: ... continue, with gated_off = _gated_off_managed_servers(), which counts a gate that raises as closed. A secondary filter: grant_in_skill_view asks may_skip_gate_now(ref) per tool before adding a kept autoApprove verb to the view's allowedTools.

ADJUDICATED: 14 candidates, 0 confirmed newly-refused

PLATFORM COVERAGE: posix and windows rows for the operations around this area: kiro-cli mcp list, kirocrew doctor, reading ~/.kiro/settings/mcp.json, reading the agent spec's allowedTools, and running the new tests. The Windows rows use .exe names, %USERPROFILE%, \ separators and cmd double quotes. The diff's own platform assumption is that spec_gate is closed on Linux (noted in test_kas_engine_features.py). That lives in the projection, not in shell text, so no shell row on either platform can probe it. The relaunch rewrite (with_managed_launch) swaps command/args for the managed launcher no matter what path spelling the entry used. It refuses nothing, so there is no platform spelling for a row to probe.

UNADJUDICATED:

  • Guard: kas_agents._project_mcp_servers gated-off withhold. Call path: KAS session start → to_client_custom_agent → _project_mcp_servers → _gated_off_managed_servers() → agent._gated_off_servers() → continue (the entry is dropped from mcpServers; the @kirocrew-computer ref stays in tools). Inputs I would have submitted:
    • a hand-written or legacy agent spec declaring kirocrew-computer on macOS with Computer Use turned on, where the gate should be open and the entry kept;
    • the same spec when the keystone is briefly unreadable. A raising gate counts as closed, so a user who has Computer Use enabled would lose the server for that session. That costs one session's capability and does not refuse a shell command; it is a narrowing worth confirming against how often the keystone read fails.
  • Guard: kas_auto_approve.grant_in_skill_view per-ref may_skip_gate_now check. Call path: kiro-cli spawn → prepare_native_skill_projection → grant_in_skill_view(agent.name, view) → auto_approve_refs (server-level strip_ungoverned_auto_approve) → per-tool may_skip_gate_now(ref). This adds grants that did not exist before on kiro-cli, so the change mostly widens. Its refusal side only withholds a grant the old code never gave in the view. Inputs I would have submitted:
    • an owner-written autoApprove: ["read_page"] on an ungoverned server where the ceiling governs only some other tool on that server;
    • an allowedTools that is not a list (left alone, so no grant);
    • a verb holding a glob or / (_NON_LITERAL, so it keeps prompting, same as before).
  • Guard: hoist_servers_for_session / _relaunch_reserved_servers command/args replacement. It rewrites the launcher of a reserved kirocrew-* entry rather than refusing it. One input would be a user who deliberately pointed kirocrew-core at a dev checkout's launcher with a timeout key; their custom command is now replaced by the installed one. That is a lost customization, not a refusal, and the classifier cannot see it.

[SCOPE-REVIEWED] ee9368a

@github-actions

github-actions Bot commented Nov 8, 2026

Copy link
Copy Markdown
Contributor

Opus 5.5 Review — ✅ no blocking findings

Reviewed ee9368adf047350f8261ff157b3f019a531dd28d — this comment is updated in place on each push.

Review details

No findings.

[OPUS-REVIEWED] ee9368a

Verdict parsed from the review's SHA-scoped output markers for commit ee9368adf047350f8261ff157b3f019a531dd28d.

False positive or not applicable? A repository writer can comment:
/ai-review override fable ee9368adf047350f8261ff157b3f019a531dd28d 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
@buluoray
buluoray merged commit 109d953 into release/0.12.0 Nov 8, 2026
24 of 27 checks passed
@buluoray
buluoray deleted the crew/slack/0.12.1-wave2 branch November 8, 2026 20:31
@github-actions github-actions Bot removed the readiness: action required A blocking check or review needs attention label Nov 8, 2026
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.

1 participant