Skip to content

fix: backport 0.12.1 batch 44 (10 user-reported fixes) - #32647

Merged
buluoray merged 10 commits into
release/0.12.0from
crew/slack/0.12.1-batch-44
Nov 7, 2026
Merged

buluoray merged 10 commits into
release/0.12.0from
crew/slack/0.12.1-batch-44

Conversation

@buluoray

@buluoray buluoray commented Nov 7, 2026

Copy link
Copy Markdown
Contributor

Problem / Motivation

Goal: Bring ten small user-reported fixes that merged on main after v0.12.0 onto release/0.12.0, one commit each, in a PR of its own.

Why it matters

Each fix closes an issue a user filed. Each faulty line was checked on release/0.12.0 before the pick; the right-hand column names it.

PR Reported by What a 0.12.0 user sees today Faulty code on release
#30578 el-pedrito (#30510) An app that marks itself degraded is unloaded on the next reconcile. apps/hooks_integration.py:560
#30496 TauCrus (#14618) The desktop app ships without the Feishu SDK, so Feishu cannot connect. packaging/build-desktop.sh:871
#31571 leonlaiyc (#27064) Bot work-ledger tools are missing on the goose, copilot, deepseek and strands harnesses. providers/mirrors/opencode.py:340 (place_single_binary_array)
#31727 natalieflora (#23586) A trusted grep or glob that searches for curl upload flags is refused as exfiltration. hook_runtime/gate_tiers.py:618
#24742 c200chromebook (#24475) On Windows, a project at a drive root does not contain the paths below it. dashboard/file_api/pinned_io.py:430
#30690 ganeshkarewad (#29048) On Windows, a bundled embeddings model load is discarded. embeddings_model.py:294
#31791 billygerhard (#21617) A Bot member's briefing leaks into a temporary chat. context_assembly/session_start.py:409
#32391 JuntaoMei (#29400) Scrolling a channel moves the whole page, not just the message list. ChannelPage.tsx:918
#32403 leonlaiyc (#24853) A cron job's next run shifts by an hour across a DST change. cron_service/schedule.py:642
#32087 jongaffen (#27775) On Windows, a restarted gateway has no hidden console. cli_server_restart.py:386

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.

What changed (motivation → approach → change)

Each main squash commit was cherry-picked with -x onto release/0.12.0 @ b80bdd7f7d, in merge order on main, one commit each. Every pick applied cleanly; every commit has added and removed lines identical to its main commit.

# PR main commit release commit
1 #30578 keep apps that mark themselves degraded loaded across reconcile f8c4df63ee 630d0ac35f
2 #30496 bundle the feishu sdk in the desktop app b764e20635 ae455253d0
3 #31571 mount the work-ledger stub on goose, copilot, deepseek and strands a119f8acbe 78d933a027
4 #31727 let a trusted grep or glob search for curl upload flags de983908ad fae3e2c9e6
5 #24742 let a drive-root project contain the paths below it eaf92fc277 f235468c56
6 #30690 stamp a bundled model by path so a Windows load is not discarded 09b795325c 3dbd46fb4b
7 #31791 withhold the member briefing in temporary chats 9b6af7a19f ee8b60c402
8 #32391 keep channel scrolling inside the message list a65bf4c55d 8380d550b4
9 #32403 keep a job's next run at its wall time across a dst change 83ec53299b 8e610c7005
10 #32087 give the restarted windows gateway a hidden console 430c260549 344cc309c8

#30578 carried a co-author trailer with an internal address and a non-public author email, and #30690 a non-public author email, which the push guard refuses. Those two commits are authored with the public identity and name their original author in the message; the internal trailer line is removed. The other eight keep their original author.

builder-security checked: #24742 changes the file API's project-containment check, the same change already merged on main; no bind, listener, route, auth mechanism or storage grant changes.

Backwards compatibility

Compatible: each change is the one already merged on main, and its compatibility note applies unchanged.

Removes nothing: no setting, key or caller is removed.

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 was run here: every touched test file plus every test file that imports a touched module, 52 files, -n 4 --dist loadgroup. 1893 passed, 26 skipped, 0 failed.
  • Frontend: tsc -p tsconfig.app.json --noEmit passes. vitest run ChannelPage channelNarrow: 8 files, 154 passed.
  • Electron: node --test electron/test/packaging.test.js: 48 passed, 3 failed. The same 3 (Linux desktop naming, first-download installer design contract) fail on a clean release/0.12.0 @ b80bdd7f7d.
  • 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 one frontend pick (#32391) carries the same rendered change as its main PR, whose screenshots show it.

Related Issues

Backport to the 0.12 line for 0.12.1 of #30578, #30496, #31571, #31727, #24742, #30690, #31791, #32391, #32403 and #32087. Source issues: #30510, #14618, #27064, #23586, #24475, #29048, #21617, #29400, #24853, #27775.

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.

buluoray and others added 10 commits November 7, 2026 21:41
…30578)

on_app_enable and boot recorded the loaded hook signature only when
ctx.health was healthy, so an app whose successful on_startup called
ctx.health.mark_degraded was torn down and re-enabled every reconcile
tick. Gate the record on the wiring itself: routes registered without a
failure mark and on_startup returned normally. Health is still
published.

(cherry picked from commit f8c4df6)

Original author: Zezhen Xu (#30578).
…ds (#31571)

The shared place_single_binary_array admitted the gateway's
session-bound kirocrew-work broker stub only behind an opt-in keyword
that opencode alone passed. goose, copilot, deepseek and strands place
their mcpServers array through the same function, so a worker on any
of them got no work_brief, work_report or work_ledger_record, and its
conductor could not tell it from a dead one.

The stub is safe to mount for a reason that does not depend on the
harness: gatewayd checks its stub token on every call, and only the
exact gateway shape bound to this session's token is admitted. Drop
the keyword so the admission applies on every caller. A narrowed
server is still withheld whole on the harnesses without a per-tool
deny channel, the tools allowlist and disabled still apply, and every
other identity-bound server stays withheld.

Builds on #27066 by leonlaiyc.

Closes #27064

(cherry picked from commit a119f8a)
#31727)

A file search for the text `-d @` or `-F x=@` was refused by the exfil
gate: with no command, the only text a grep or glob call is judged on
is its title, which is the text it searches for, and the curl rows read
it as a command line. A call whose host-trusted identity names the grep
or glob built-in, with no MCP server and no command of any kind, runs no
shell, so the exfil rule no longer judges it. Every other tier still
does; an untrusted or MCP-served name, the code built-in and any call
carrying a command keep the rule.

Builds on #23845.

(cherry picked from commit de98390)
…24742)

_resolve_project_relative and the path-completion witness check tested
containment with startswith(root + os.sep). realpath keeps the
separator of a root, so for a project at H:\ or / the prefix became a
doubled separator and every relative path or subdirectory under the
project was refused as outside it. Both now ask
sandbox_plan.path_within, the helper the validate_cwd and terminal
sites use.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
(cherry picked from commit eaf92fc)
…ot discarded (#30690)

(cherry picked from commit 09b7953)

Original author: Javen Cinocca (#30690).
A temporary chat promises no memory reads, but a legacy member's
session-start, slim-resume and post-compaction sections still read the
agent-writable briefing into [CURRENT ASSIGNMENT]. Pass
include_briefing=not blocks_reads at those three sites, as the
private-member envelope already does for temporary turns, so the section
carries the withheld placeholder instead.

Builds on #21630.

(cherry picked from commit 9b6af7a)
The Channels page followed new messages with scrollIntoView on an empty
end-of-list sentinel, and scrollIntoView scrolls every scrollable
ancestor, not only the list. The channel pane was sized h-full inside
the page's overflow-y-auto content container, so anything shown above
it (an error notice) made the container overflow: an agent reply then
scrolled the page and the notice out of view, and the notice pushed
the composer below the bottom edge.

Scroll the message list element itself, and make the channel pane a
shrinking flex child of the standard content container, so a new
message moves only the list and the composer stays in view. On a phone
the pane's existing -mb-8 now cancels the container's bottom padding,
as it was written to, so the composer sits on the bottom edge.

Builds on #29402 by JuntaoMei.

Co-authored-by: Juntao Mei <14120031+JuntaoMei@users.noreply.github.com>
(cherry picked from commit a65bf4c)
…#32403)

A cron job's advertised next run and its timer boundary were computed
by handing croniter an aware start in the job's zone. croniter 2 then
shifts any run that lies past a DST change by the offset change a
second time: a 03:00 job read 04:00 PST the day before the fall-back,
a 22:00 job 21:00 PDT the day before the spring-forward, and is_due
refused the advertised instant.

Step the zone's naive wall clock instead and attach the zone to the
result, as the schedule label already does. A repeated wall time whose
first instant is not after now takes its second; a spring-forward gap
resolves as fold=0 does, as before.

Builds on #24864.

Closes #24853

(cherry picked from commit 83ec532)
@buluoray
buluoray requested review from a team and KiroCrewCaptain as code owners November 7, 2026 21:48
@buluoray
buluoray requested review from tim-finnigan and removed request for a team November 7, 2026 21:48
@github-actions

github-actions Bot commented Nov 7, 2026

Copy link
Copy Markdown
Contributor

Sensitive Change Review

Sensitive files changed:

  • src/kiro_crew/hook_runtime/gate_tiers.py

  • No reasoning template for 344cc30 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: 344cc309c85f8c3f4a72e3dc0a4923d6b2d6c826
- 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 7, 2026
@github-actions

github-actions Bot commented Nov 7, 2026

Copy link
Copy Markdown
Contributor

UX Review (Opus 5.5) — ✅ PASS

UX-level review of 344cc309c85f8c3f4a72e3dc0a4923d6b2d6c826 — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS does not.

UX-Verdict: PASS

Nothing to check. The only website change is a scroll fix in ChannelPage.tsx: the channel message list now scrolls inside its own box. It adds no control, string or state, so the missing screenshots leave nothing unreviewed.

[UX-REVIEWED] 344cc30

Description read

[DESCRIPTION-READ] 78050fab4081367e08905862bc3a7b31f4712b2734a496b99617d7edcd1725b6

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 7, 2026

Copy link
Copy Markdown
Contributor

First Principles Review (Opus 5.5) — ✅ PASS

Premise-level review of 344cc309c85f8c3f4a72e3dc0a4923d6b2d6c826 — 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: ten cherry-picks from main, each tied to a user-filed issue. The pins it removes (opencode-only stub admission, the desktop note saying not to install Feishu, the DETACHED_PROCESS assertion) were each reversed on main with a linked issue.

What this change ships

Inventory (10 items) — 10 justified

Intent: give 0.12.x users ten bug fixes already merged on main. FIX.

  1. An app that marks itself degraded stays loaded and is no longer torn down on each reconcile (apps: an app that marks itself degraded in on_startup is torn down and re-enabled on every reconcile tick #30510) — justified
  2. The desktop app bundles the Feishu SDK, and the Windows build precompiles it (Feishu channel: desktop-app upgrade wipes optional lark-oapi dependency, silently breaking the channel (needs reinstall every upgrade) #14618) — justified
  3. Goose, copilot, deepseek and strands now mount the work-ledger tools (goose, copilot, deepseek and strands workers get no work-ledger tools: the session-bound stub admission is opencode-only #27064) — justified
  4. A trusted grep or glob that searches for curl upload flags is no longer refused as exfiltration (Exfil guard false positive: '-d @' substring match blocks GNU date -d @<epoch> and Grep searches for the literal #23586) — justified
  5. A project at a drive root now contains the paths below it, in file reads and path completion (subagent cwd allow-list never matches anything under a drive-root entry (e.g. H:\) on Windows #24475) — justified
  6. On Windows, a bundled embeddings model load is no longer discarded as changed (Embedding model load fails permanently on Windows with false "changed on disk" warning (0.12.0-rc.4) #29048) — justified
  7. A temporary chat withholds the member briefing (Temporary session still injects V1 member briefing ([CURRENT ASSIGNMENT]) #21617) — justified
  8. Scrolling a channel moves only the message list, and the pane shrinks below error notices (Channels: a new agent message scrolls the page past the composer #29400) — justified
  9. A cron job's next run keeps its wall time across a DST change (Cron next-run timestamps shift off the configured wall time across DST #24853) — justified
  10. On Windows, a restarted gateway gets a hidden console, so its children open no windows (kirocrew restart (Windows): the respawned gateway has no console, so every console child it starts with only CREATE_NEW_PROCESS_GROUP opens a visible window #27775) — justified

[FIRST-PRINCIPLES-REVIEWED] 344cc30

Description read

[DESCRIPTION-READ] f0314807ee34948cf084ed1ee11dde10e8e5357d72b995946518098a39a076c2

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 7, 2026

Copy link
Copy Markdown
Contributor

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

Design-level review of 344cc309c85f8c3f4a72e3dc0a4923d6b2d6c826 — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS does not.

Design-Verdict: PASS

cannot evaluate: missing screenshot or recording of the Channels page at this revision. The #32391 pick changes the page's layout, and the workflow's evidence list is empty: 0 attachments and 0 committed images.

Blockers

The ChannelPage layout change has no rendered evidence of this revision
Trigger: the pick changes the page container to "overflow-y-auto flex-1 min-h-0 flex flex-col" and the channel body from "flex h-full relative" to "flex flex-1 min-h-0 relative" in website/src/pages/ChannelPage.tsx. The description says "the one frontend pick (#32391) carries the same rendered change as its main PR, whose screenshots show it".
Outcome: the main PR's screenshots show the page with main's surrounding layout, and nothing confirms release/0.12.0 has the same ancestor flex chain → the height chain could resolve differently on the release branch, for example leaving the composer off-screen or the list collapsed → a broken Channels page would ship in 0.12.1 with no artifact able to catch it. Images from another PR are not evidence of this revision.
Fix: attach a screenshot or short recording of the Channels page built from this branch, showing a long channel where only the message list scrolls and the composer stays visible, on desktop and mobile widths.
Clears when: the PR description carries a github.com/user-attachments image or recording of this revision's Channels page, and the workflow lists it in design-evidence.txt.

Notes

[DESIGN-REVIEWED] 344cc30

Description read

[DESCRIPTION-READ] d124c1d7a571017a77daf157786684d26e14f42fb3f42c47cf27c76d48bee78c

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 7, 2026

Copy link
Copy Markdown
Contributor

Security Scope Review -- ✅ PASS

Which legitimate operations does 344cc309c85f8c3f4a72e3dc0a4923d6b2d6c826 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): 10 classified, 0 newly refused, 10 still allowed.
  • ubuntu-latest (platform posix): 10 classified, 0 newly refused, 10 still allowed.
  • windows-latest (platform windows): 8 classified, 0 newly refused, 7 still allowed.

No candidate flipped from allowed to refused.

Scope-Verdict: PASS

Reason: The only change in the shell deny composite loosens it. _rule_exfil gets an early return None for one case, a host-trusted grep/glob built-in call that has no command line, and adds no new refusal. Every other security-relevant hunk also widens what is admitted: the drive-root containment fix, and the work-ledger stub now admitted for goose, copilot, deepseek and strands. Two new stricter checks sit outside the shell classifier (the embedding-model stamp mismatch refusal, and the briefing withheld on temporary turns), so they are named under UNADJUDICATED below and do not block.

TIGHTENING: No shell-tier tightening. The exfil tier gains an exemption, not a pattern: if (call.identity_trusted and not call.mcp_server and call.mcp_tool in _HOST_SEARCH_BUILTIN_TOOLS and not call.command and not facts.raw_shell_commands): return None. The stricter mechanisms this diff does add are outside the composite: the embedding-model load guard if (info.st_ino, info.st_size, info.st_mtime_ns) != (stamp[1], stamp[2], stamp[3]): ... return None, and member-section assembly passing include_briefing=not blocks_reads.

ADJUDICATED: 18 candidates, 0 confirmed newly-refused

PLATFORM COVERAGE: posix and windows. The posix rows cover these neighbours of the exfil tier: a grep/rg search for the -d @ / --data-binary text, curl -d with an inline body and no @, and the /api/ready probe. They also cover the build_backend Feishu pip install. The windows rows use findstr/rg.exe with \ paths, curl.exe, python.exe with %OUT%/%ROOT_URI%, and the precompile_windows.py --module lark_oapi step, which match the Windows spellings this diff adds in build_backend_windows and in the CREATE_NO_WINDOW respawn path. No row is any: every candidate names an executable or a path. The built-in grep/glob tool call itself (no command line) cannot be written as a shell row; see UNADJUDICATED.

UNADJUDICATED:

  • Exfil carve-out for host-search built-ins. Call path: HookManager.on_tool_call(title, mcp_tool_name="grep"|"glob", raw_params={pattern, path}, mcp_identity_trusted=True) → gate_tiers._rule_exfil. This is a structured tool call, not a command line, so the classifier cannot settle it. It only loosens, so there is no regression to find. The inputs I would have submitted are trusted grep/glob searches for -d @, --data-binary=@ and -F file=@, with the title "Searching for '…'"; the diff's own tests cover these.

  • Embedding model stamp check. Call path: embeddings_model._model_file_stamp_unaliased → require_unaliased_model_file → os.fstat(fd) compared with _model_file_stamp(path) on inode, size and mtime. It newly refuses to load the model when the descriptor and the path disagree. Legitimate inputs I could not get a verdict for:

    • an unchanged models/model.gguf on a Windows NTFS volume;
    • the same file on a network or SMB share, or a ReFS volume, where fstat and stat may report st_ino from different APIs (the diff already drops st_dev for exactly this reason);
    • a model reached through a junction or symlink under the models directory.

    This is plausible but not demonstrated, so it does not block.

  • Briefing withheld on temporary turns. Call path: build_session_context / build_message → _build_member_section(include_briefing=not blocks_reads). This is context assembly, with no shell surface. The inputs I would have submitted are an incognito turn (it blocks writes only, so the briefing should still be delivered) and a temporary turn on each of the fresh, slim-resume and post-compaction paths.

  • Dashboard drive-root containment. Call path: path_complete._complete_path_listing and pinned_io._resolve_project_relative → sandbox_plan.path_within. This only loosens; path_within normalizes the parent and strips trailing separators, so it is never stricter than the old root + os.sep check. The input I would have submitted is a project at H:\ or / with a relative path below it.

[SCOPE-REVIEWED] 344cc30

@github-actions

github-actions Bot commented Nov 7, 2026

Copy link
Copy Markdown
Contributor

GPT 6.1 Review — ✅ no blocking findings

GPT 6.1 completed its review of 344cc309c85f8c3f4a72e3dc0a4923d6b2d6c826 and found no blocking issues.

This comment is updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] 344cc30

False positive or not applicable? A repository writer can comment:
/ai-review override gpt 344cc309c85f8c3f4a72e3dc0a4923d6b2d6c826 false-positive: <reason, 20 to 500 characters>

@github-actions

github-actions Bot commented Nov 7, 2026

Copy link
Copy Markdown
Contributor

Opus 5.5 Review — ✅ no blocking findings

Reviewed 344cc309c85f8c3f4a72e3dc0a4923d6b2d6c826 — this comment is updated in place on each push.

Review details

No findings.

[OPUS-REVIEWED] 344cc30

Verdict parsed from the review's SHA-scoped output markers for commit 344cc309c85f8c3f4a72e3dc0a4923d6b2d6c826.

False positive or not applicable? A repository writer can comment:
/ai-review override fable 344cc309c85f8c3f4a72e3dc0a4923d6b2d6c826 false-positive: <reason, 20 to 500 characters>

@buluoray
buluoray merged commit 2f55268 into release/0.12.0 Nov 7, 2026
24 of 28 checks passed
@buluoray
buluoray deleted the crew/slack/0.12.1-batch-44 branch November 7, 2026 21:57
@github-actions github-actions Bot removed the readiness: checking Automated validation is still running label Nov 7, 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.

3 participants