Repository navigation
fix: backport 0.12.1 late batch 4 (15 fixes merged 11-08 12:05-13:24Z) - #33207
Conversation
…#33105) Agent Worlds restored the last scene by casting whatever string localStorage held to a SceneKey. A key another build saved (a scene added later, then a downgrade) named no scene, so every scene wrapper stayed hidden, the canvas was blank and no scene chip was checked. The saved key is now looked up in SCENES and falls back to office, the same default as no saved key. Original author: Rohan Khanderia (cherry picked from commit 99ebd1b)
A Hooks-page PreToolUse hook whose matcher names a kiro-cli tool (execute_bash, fs_write, web_fetch, ...) never ran on a claude call: claude's permission request carried no harness_tool_id, because claude had no HARNESS_TOOL_TABLES row, and its title is the command or a prose label, not the tool name. Add CLAUDE_TOOL_IDS_BY_KIRO_TOOL and register it under ACP_BACKEND_CLAUDE. The existing dispatch branch then names each claude permission event from the _meta.claudeCode.toolName its tool_call frame stamped (claude#Bash), so the hook matcher meets Bash, Write, Edit, WebFetch and the rest. The name is read by hook matching alone, where a hook can only refuse a call. harness_builtin_tool, the READ_ONLY proof and the deny-target name, stays Side-Chat-only, so auto_deny_tools and governance verdicts on claude are unchanged, and claude stays out of ACP_BACKENDS_CREW_FIRES_SPEC_HOOKS. A new claude-agent-acp 0.94.0 capture holds a Bash, Edit and WebFetch pair; a test replays it through the real parser. Original author: Bolin Chen (cherry picked from commit 28cb680) # Conflicts: # src/kiro_crew/docs/agent-spec-fields.md
A spec hook Crew runs in kiro-cli's place (KAS, goose, opencode, strands, claude observer events) now reads the payload kiro-cli's V2 engine sends: camelCase hook_event_name, tool_name as kiro-cli's full name (shell, read, write, subagent, @server/tool), tool_tags, and assistant_response on stop. ~/.kiro/hooks file hooks and Hooks-page hooks keep their names. Governance's events check still reads the store's PascalCase event. Original author: Joe Guo (cherry picked from commit 2ee2df5)
A webhook token can pin only an agent template, and its call runs that template on the shared default memory. The mate editor's Webhooks pane and its rail count matched a token's agent against the mate's name, so no token ever matched and the default mate's pane said nothing was bound even with a token on the kirocrew template it runs. Match on what a call actually runs as: a token belongs to a mate when the mate has no memory of its own and runs the token's template, the same line the agent picker draws for a covered mate. A mate with its own memory still lists nothing, because no token can start its turns. Closes #30412 Original author: Bolin Chen (cherry picked from commit de25ac3)
…uns (#33128) The setup panel's Clone was natively disabled while the clone ran, which took it out of the tab order and dropped the focus of the keyboard user who pressed it, though the repository step stays on screen. Enter in the URL field could also start a second clone. Mark the busy Clone with aria-disabled, route the button and Enter through one start that claims the clone synchronously so a double press in one tick starts one clone, give the busy state the muted look of a disabled primary button, and keep disabling it for an empty URL. A clone request that failed outright left nothing on screen once the spinner cleared; both that failure and the backend's own refusal now show through ErrorNotice. Original author: Rohan Khanderia (cherry picked from commit 48cac3a)
On a harness-managed backend (KAS) at an explicit threshold, a /compact whose wait gives up with no status keeps the session with the failure cooldown. The abandoned /compact can still finish, and the next turn's pre-turn drain drops its late frames only after that turn's prompt is built, so nothing restored the session-start context. That failure now arms the one-shot re-injection; a status KAS reported (a refusal is failed) arms nothing. Closes #32247 Original author: Bolin Chen (cherry picked from commit 44877d2)
… runs (#33131) The Ops Mission Control board's Poll & claim button set native `disabled` while the dispatch cycle ran. A disabled button leaves the tab order, so a keyboard user who pressed it lost focus to the page for the whole cycle. It now uses `aria-disabled` with the same dimmed look, and the click claims the cycle in a ref, so a second press is ignored even before the pending state re-renders. Original author: Rohan Khanderia (cherry picked from commit ce879f3)
…hinese (#33160) #33111 joined the chip parts with a unit list. CLDR's zh unit list has no separator, so a mixed-status request read "1 项进行中1 项已完成". A narrow conjunction list keeps English as "1 in progress, 1 done" and gives zh and ja "、". A zh-CN test pins the separator. Original author: Ray Xu (cherry picked from commit 02d4cbe)
KAS reads no cli.json, so kiro-cli's telemetry.enabled opt-out and the KIRO_DISABLE_TELEMETRY / Q_DISABLE_TELEMETRY switches did nothing. Crew now sends _meta.kiro.telemetryEnabled at initialize, failing closed on an unreadable file, and shows a session notice because KAS does not confirm the opt-out (or, for a folder-only opt-out on a shared process, that it was not applied). Original author: Joe Guo (cherry picked from commit 4a75e4c)
A single MCP tool switched off from the dashboard could still be called on the default backend, because that switch is saved outside the agent spec and only the spec's own switches were applied there. Apply every per-tool switch the other backends already read, the same way the spec's own are applied, and pin it with a test. Closes #25034 Original author: Bolin Chen (cherry picked from commit 34ba4ff)
Sensitive Change ReviewSensitive files changed:
How to clear this check:
Only the author's comment counts this way. A reviewer can instead |
UX Review (Opus 5.5) — ✅ PASS (a BLOCK with no Trigger and no Outcome does not block)UX-level review of UX-Verdict: PASS cannot evaluate: missing screenshots. This backport changes several dashboard surfaces, but the description attaches no image, so the blind read was not performed and nobody has looked at any of them. Evidence gaps
[UX-REVIEWED] b6ac0e2 Description read[DESCRIPTION-READ] 837782f25c0285ada4a29934e0847016f8e2518298b91c8d7a79a1d58d4a116d 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 |
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS This backports fixes already merged on Notes
[DESIGN-REVIEWED] b6ac0e2 Description read[DESCRIPTION-READ] b5ac5b0af71b5d74a0a86a27cfa30f098f3ca49d21a705368a9165dd4942fd37 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 |
First Principles Review (Opus 5.5) — ✅ PASSPremise-level review of First-Principles-Verdict: PASS Every item is a fix already merged on Notes
What this change shipsInventory (10 items) — 9 justifiedIntent: ship 16 user-visible fixes already merged on
[FIRST-PRINCIPLES-REVIEWED] b6ac0e2 Description read[DESCRIPTION-READ] 6ac5a1a9010a3797b648e0c1038aa8692e549e9723a0bf1fc4a24caad15eda7d That sha256 names the PR description this verdict read. Recompute it with |
GPT 6.1 Review — ✅ no blocking findingsGPT 6.1 completed its review of This comment is updated in place on each push. Review detailsNo findings. False positive or not applicable? A repository writer can comment: |
Security Scope Review -- ✅ PASSWhich legitimate operations does Verdict basis: adjudicated, zero confirmed regressions. Security scope review -- adjudicated candidates
No candidate flipped from allowed to refused. Scope-Verdict: PASS Reason: The patch does not touch the shell deny composite ( TIGHTENING: hook tool-name matching on the claude backend -- a Hooks-page PreToolUse deny hook written in kiro-cli names now meets claude's built-ins, so a claude call the user's hook means to refuse is no longer let through for lack of a name match: ADJUDICATED: 15 candidates, 0 confirmed newly-refused PLATFORM COVERAGE: posix and windows. The diff makes path assumptions in UNADJUDICATED:
[SCOPE-REVIEWED] b6ac0e2 |
Opus 5.5 Review — ✅ no blocking findingsReviewed Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
Problem / Motivation
Goal: Bring 16 fixes merged on
main(13 on 2026-11-08 between 12:05 and 13:24 UTC, plus #30086, #30874 and #33221) ontorelease/0.12.0for 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: agent spec hooks get kiro-cli's stdin field names, the telemetry opt-out reaches KAS, a dashboard MCP tool switch stays off, the agent role comes back after a timed-out compaction, and KAS status reads during a gateway restart show offline. A claude PreToolUse hook now meets claude's built-in tools. The rest fix lost keyboard focus, Chinese chip separators, the Worlds scene fallback, the wake judge retry, page embed reset and the Bot webhooks token list.
Not a goal
main; nothing is changed only here.CHANGELOG.mdchange.What changed (motivation → approach → change)
Each main squash commit was cherry-picked onto
release/0.12.0@80d6211963, in merge order onmain, one commit each, 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 anOriginal author:line.d86392c54528307e799199ebd1ba0eeed6f5e54294a6b06192fc7e84708c1cd288151fa4c46b4a7528cb6801de3bddd5a7622ee2df52375308db5b8edf553b06438b7db8a04fde25ac3b9cb9ca9ee1f29b1933a6512a32d0875148cac3a9f39b398db11b44877d2df9cbf5125225ce879f309d8ebbeaa0e102d4cbe1c42f9a41ed804a75e4cfad67425d10d934ba4ffd3c7c210b0f6e0826d055aeb6ac0e2092#33221 is the
mainfix for #33026'skiro_homeimport (kas_cli_settings.pybound its ownkiro_home, so the skill projection facade tests saw two objects). It ships with #33026 so the two stay together on 0.12.Hand-resolved hunks:
src/kiro_crew/session_compaction.py: hand port. Onmainthereportedstatus tracking in the compaction wait came with feat(session): opt-in idle compaction before prompt-cache expiry #24384 (idle compaction, a feature). The port adds only that tracking (reporteddeclared, set from the stream's terminal status and fromwait_for_compaction), then fix: restore the agent role after a timed-out kas compaction #33149's ownneeds_context_reinjectionhunk.src/kiro_crew/acp/runtime_session_load.py: import line isfrom kiro_crew.acp import kas_cli_settings, kas_env_launch(0.12 has nobatch_guardimport there).src/kiro_crew/acp/harness/kas.py: import union with release's imports. Its sentence is added to release's wording indocs/system-specs/modules/agent-host-contract.mdanddocs/system-specs/modules/harness-parity.md.website/src/test/OpsMissionControlPageCoverage.test.tsx, where that file still lives on 0.12.Left out of this batch:
acp/batch_guard.py; needs #31274's creator-token rework (_shell_creates). Too long a chain.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
release/*, so the related set ran here, one pytest process at a time with no xdist, withTMPDIRoutside~/.kiro/crew: the 15 Python test files the batch changes, 866 passed; 22 related compaction, KAS agent/env/settings, hook and decisions files, 1,052 passed.test_kas_cli_settings.pyplus the skill projection family (13 files:test_acp_runtime_skill_projection_opt_out.py,test_native_skill_projection*.py,test_skill_projection_facade.py,test_skill_projection_sandbox_seal.py), 415 passed.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 #33111, #33105, #33092, #32465, #30086, #30937, #33059, #32597, #33108, #33128, #33149, #33131, #33160, #33026, #30874, #33221.
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
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)Contribution License Agreement
Contributions are accepted under the project's Apache-2.0 license; see the Licensing section of CONTRIBUTING.md.