Skip to content

fix: backport 0.12.1 wide batch 2 (8 user-facing fixes + 2 prerequisites) - #32808

Merged
buluoray merged 9 commits into
release/0.12.0from
crew/slack/0.12.1-wide-2
Nov 8, 2026
Merged

buluoray merged 9 commits into
release/0.12.0from
crew/slack/0.12.1-wide-2

Conversation

@buluoray

@buluoray buluoray commented Nov 8, 2026 •

Copy link
Copy Markdown
Contributor

Problem / Motivation

Goal: Bring eight fixes for problems a 0.12.0 user can hit, merged on main after v0.12.0, onto release/0.12.0, with the two main commits they need first, one commit each, in a PR of its own.

Why it matters

Each fix changes behaviour a 0.12.0 user can run into today. Every pick applied to release/0.12.0 with only the doc or test hunks listed under "What changed" resolved by hand, so the code each one replaces is on release; the right-hand column names where.

PR Reported by What a 0.12.0 user sees today Faulty code on release
#30093 bolichen97 (#30057) On the claude backend, the agents meant to run no tool (kirocrew-lite, the knowledge workers, kirocrew-guest) can run shell commands, write files and fetch URLs whenever a ~/.claude allow rule covers them. acp/client_claude_session.py:423, knowledge/llm_pool.py
#30616 buluoray On a third-party harness, an MCP server whose env uses ${VAR} gets the literal text and its tools are lost, silently. acp/session_mcp.py:91
#31147 dodgeblaster (#28446) Clicking an artifact tab's × in the chat side panel drops an open comment draft without asking. pages/chat/SidePanel.tsx:50
#30833 NicholasRBowers (#9163) Background consolidation can delete a lesson the user taught when its rewording overlaps it, and the agent falls back to a vaguer rule. vector_memory_runtime/lesson_write.py:49
#31730 bolichen97 (#31716) A sandboxed session loses its scratch window when the host empties /dev/shm, so work breaks mid-task, again every few minutes. sandbox_launcher_program.py:216
#19981 CrysisDeu (#19976) Saving a crew whose name has a space returns HTTP 502 behind a tunnel or proxy, and the change is lost. dashboard/agent_admin/crew_update.py, dashboard/routes/agents.py
#32224 RohanK6 (#27388) On a loaded Mac, one slow resolver round trip makes a Bot refuse its own spec as "source unreadable", and later starts under that prefix are refused for 30 s. hook_runtime/safe_reads.py, agent_capabilities.py:17
#32210 bolichen97 (#32197) On Amazon Linux 2 the agent has no browser: the browser tool, Settings' Test and the live view all fail, because Chromium cannot load libcups 1.6.3. browser_harness/chromium.py:14

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.
  • Sign-in, credential and sandbox tightening fixes; they go in a batch of their own.

What changed (motivation → approach → change)

Each main squash commit was cherry-picked with -x onto release/0.12.0 @ 052ce87b11, in merge order on main, one commit each. A prerequisite sits right before the fix that needs it.

# PR main commit release commit Why it is here
1 #30093 remove every built-in from a zero-tool claude session 515f616cd6 70acf3b05b fix
2 #30616 expand ${VAR} in MCP env for third-party harnesses 2fa4077886 eb11317013 fix
3 #31147 close an artifact tab's chip × through its draft guard e4f2a382cd 19667b9868 fix
4 #30833 keep a user's lesson when a consolidation rewording overlaps it f54bb352fd 3757050028 fix
5 #31730 keep the scratch window when the host empties the stand-in roots 782fcbd918 cfce6703e3 fix
6 #19981 send proxy-unsafe crew names in the request body 4d0ae46a5c e2c29c66f5 fix
7 #30909 judge each agent spec on the canonical sensitive-path gate in the snapshot scan 9f8d7d61ed af6e19edb3 prerequisite of #32224: adds the canonical_gate keyword #32224 extends
8 #32224 a resolver stall cannot refuse a member its own spec 123cc51aed 5d12bf3e61 fix
9 #32210 use the headless shell where libcups cannot load chrome 23b4795f78 2525ddb10c fix

Hand-resolved hunks, none of them in product code:

#30093, #31147, #30833, #31730, #30909 and #32224 are authored with the public identity, and each message names its original author, because the commit guard refuses the original address on a hand-resolved pick. #19981 and #32210 keep their original author.

#30755, the prerequisite of #32210, landed on release/0.12.0 in another batch (7782dbb73c) while this PR was open, so it is no longer carried here.

Left out on purpose:

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. #30093 narrows behaviour on the claude backend only, for the agents whose spec already declares "tools": []. Those agents now get no tool there, as on every other backend.

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 after each pick: each touched test file, plus the test modules that import the touched source, about 10,000 tests in total.
  • Every failure seen also fails outside this branch, on this host:
  • Frontend: tsc -p tsconfig.app.json --noEmit passes. vitest for fix(dashboard): close an artifact tab's chip × through its draft guard #31147 (side panel and artifact panel, 437 passed) and fix(agents): send proxy-unsafe crew names in the request body #19981 (agent path, API client, roster, capabilities, 912 passed).
  • flake8, isort, black gate, subprocess-encoding gate, mypy src/kiro_crew, gen-ui-index --check and check_harness_parity.py pass. check_file_line_cap fails only on dashboard/chat_api/resume.py, which is already over the cap on release and is not touched here.

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 two frontend picks (#31147, #19981) 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 #30093, #30616, #31147, #30833, #31730, #19981, #30909, #32224 and #32210. Source issues: #30057, #28446, #9163, #31716, #19976, #20083, #30908, #27388, #32197.

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
buluoray requested review from a team and KiroCrewCaptain as code owners November 8, 2026 00:55
@buluoray
buluoray requested review from tim-finnigan and removed request for a team November 8, 2026 00:55
@github-actions

github-actions Bot commented Nov 8, 2026 •

Copy link
Copy Markdown
Contributor

Sensitive Change Review

Sensitive files changed:

  • src/kiro_crew/hook_runtime/descriptor_identity.py

  • src/kiro_crew/hook_runtime/safe_reads.py

  • src/kiro_crew/hooks.py

  • src/kiro_crew/sandbox_launcher_program.py

  • No reasoning template for 2525ddb 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: 2525ddb10c15515d42b2590260a33a9de6a54ea6
- 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 2525ddb10c15515d42b2590260a33a9de6a54ea6 — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS does not.

UX-Verdict: PASS

cannot evaluate: missing a screenshot of the discard-comment prompt that now opens from an artifact tab's ×. The PR attaches no image, so the blind read never ran on the one UI behaviour this diff changes.

Evidence gaps

  • Artifact tab × / middle-click with an open comment draft. In SidePanelTabBodies.tsx, the change ref={setArtifactRef} routes the chip's close through ArtifactPanel's requestClose. That path runs guardDraft and raises the "discard unsaved comment" confirm (components.markdownPanel.discard_unsaved_comment, plus a beforePrompt hook). Before this change the tab just closed. The blind-read report says "BLIND READ NOT PERFORMED", and the screenshot list is empty. To close the gap, attach two screenshots to the PR description: one of the side panel with an artifact tab holding a typed, unsent comment and the pointer on its ×, and one of the confirm dialog that opens after the click, so a first-time reader can see which tab and which draft the prompt is about.

Notes

  • The prompt reuses the file-tab strings ("unsaved comment" and the discard button). In the screenshot, check that the title names the artifact tab being closed. Otherwise, with several artifact tabs open, the reader cannot tell whose draft is being discarded.

[UX-REVIEWED] 2525ddb

Description read

[DESCRIPTION-READ] 00fe415c5ea1945054be49eac845cbcfdba85362d092b2c83d8c91210ebc6b3b

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

First Principles Review (Opus 5.5) — ✅ PASS

Premise-level review of 2525ddb10c15515d42b2590260a33a9de6a54ea6 — 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 PR backports eight fixes already merged on main, plus two prerequisites. Each fix links its issue and adds tests. No pin is deleted and no default changes for users.

What this change ships

Inventory (10 items) — 10 justified

Intent: give 0.12.0 users eight bug fixes that already landed on main. This is a FIX.

  1. On the claude backend, zero-tool agents (lite, knowledge workers, guest) can no longer run shell, write or fetch tools. — justified
  2. On third-party harnesses, ${VAR} in an MCP server's env now expands, so its tools load. — justified
  3. Closing an artifact tab with its × now goes through the unsaved comment-draft guard. — justified
  4. Consolidation keeps a user's lesson when a rewording overlaps it. — justified
  5. A sandboxed session keeps its scratch window after the host empties /dev/shm. — justified
  6. Saving a crew whose name has a space no longer returns 502 behind a proxy. — justified
  7. The snapshot scan judges each agent spec on the canonical sensitive-path gate (prerequisite of fix(capabilities): a resolver stall cannot refuse a member its own spec #32224). — justified
  8. A slow resolver no longer makes a member refuse its own spec, or block that prefix for 30 s. — justified
  9. On hosts where libcups cannot load Chrome, the browser falls back to the headless shell. — justified
  10. Hand-resolved doc, CI-list and test-digest hunks. — justified

[FIRST-PRINCIPLES-REVIEWED] 2525ddb

Description read

[DESCRIPTION-READ] dd67239d77fb59bde717f62ad40ad3668db6cf2ee002884e9ef33d79f10b9211

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

Design Review (Opus 5.5) — ✅ PASS

Design-level review of 2525ddb10c15515d42b2590260a33a9de6a54ea6 — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS does not.

Design-Verdict: PASS

Nothing blocks: all nine commits are cherry-picks of fixes already merged on main, and the hand-resolved hunks touch only docs and tests.

Notes

[DESIGN-REVIEWED] 2525ddb

Description read

[DESCRIPTION-READ] 78c11592b7f7d4c7033d9e89f74c5ac648dae37ed2f2b9660bf47033eac3bfa7

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

GPT 6.1 Review — ✅ no blocking findings

GPT 6.1 completed its review of 2525ddb10c15515d42b2590260a33a9de6a54ea6 and found no blocking issues.

This comment is updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] 2525ddb

False positive or not applicable? A repository writer can comment:
/ai-review override gpt 2525ddb10c15515d42b2590260a33a9de6a54ea6 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 2525ddb10c15515d42b2590260a33a9de6a54ea6 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): 12 classified, 0 newly refused, 12 still allowed.
  • ubuntu-latest (platform posix): 12 classified, 0 newly refused, 12 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: No script-confirmed regression. Nothing in this diff changes the shell deny composite: src/kiro_crew/hooks.py only adds a re-export of canonical_path_refusal and is_sensitive_canonical_path, and the sensitive-bash, exfil and deny-rules tiers are untouched. The real tightening is in the claude zero-tool session options and a new refusal of session mounts. Neither is a shell command line, so neither could be classified; that gap is named under UNADJUDICATED.

TIGHTENING: Claude zero-tool session composition (client_claude_session.py / client_start.py) -- it stops a "tools": [] claude session from running a tool that an inherited ~/.claude allow rule pre-approves: + options["tools"] = [] / + options["allowDangerouslySkipPermissions"] = False, plus + if spec_zero_tools and mcp_servers: / + raise AcpError( in refuse_claude_zero_tool_mounts, and the adapter-floor stop now also applies when _spec_zero_tools is set.

ADJUDICATED: 17 candidates, 0 confirmed newly-refused

PLATFORM COVERAGE: Candidates were written for posix, windows and any. The posix and windows rows are paired for the spec read (~/.kiro/agents against %USERPROFILE%\.kiro\agents), the agents-directory listing, the new test runs (.venv/bin/python against .venv\Scripts\python.exe) and the base64url route probe (curl against curl.exe). They cover the one path-shaped assumption the diff makes, the realpath and canonical-gate reads of agent specs, in both spellings. The sandbox stand-in staging (tmpfs mount, /proc/self/fd, logind) and the libcups / headless-shell choice are Linux-only by construction, so they have posix rows only. No Windows spelling of them exists. The macOS resolver-pool stall the canonical gate addresses is not a shell spelling, and no candidate probes it.

UNADJUDICATED:

  • Guard: refuse_claude_zero_tool_mounts(spec_zero_tools, mcp_servers).
    • Call path: AcpClient._new_session_following_substitution (both the first build and the re-seed after substitution) and _initialize_session's session/load branch → refuse_claude_zero_tool_mounts → AcpError, and the session is not started.
    • Inputs I would have submitted:
      • a claude-backed kirocrew-lite helper whose spec declares "tools": [] and also declares an mcpServers entry it references, so the mirror does not withhold it;
      • a zero-tool spec whose wire array keeps Crew's own control-plane server;
      • a session/load resume of a zero-tool helper created before this change.
    • Each of these is now refused at session start. Whether any such spec ships is not settled by a shell verdict.
  • Guard: the adapter-floor stop in _pin_claude_starting_mode, which now also covers _spec_zero_tools.
    • Call path: session/new response → _pin_claude_starting_mode → _claude_adapter_below_setting_sources_floor() → _kill_process + AcpError.
    • Inputs I would have submitted: a background lite helper (knowledge llm_pool and similar) on a host whose claude-agent-acp is below CLAUDE_ACP_SETTING_SOURCES_MIN_VERSION, or whose handshake reports no version. Those helpers were not stopped before and are now killed before their first prompt.
  • Guard: decode_agent_name, the canonical base64url check behind the new /api/agents/b64/{name_b64}/... routes.
    • Call path: aiohttp route → agent_route_name → invalid_agent_name_response(), a 400.
    • Inputs I would have submitted: a padded segment, or one from a client using standard base64 (+/) rather than base64url.
    • These routes are new, so no existing operation is lost, and the path-form routes still go through request.match_info["name"].
  • Guard: safe_read_file / safe_read_file_bytes_nolink / validate_file_path with canonical_gate=True, plus _spec_path_shape_refusal and default_project_dir.
    • Call path: loader_snapshots._scan_materialized_index, agent_capabilities._read_spec, and the member start under prepare_runtime.
    • Inputs I would have submitted: an agent spec reached through a symlinked ~/.kiro (dotfile-managed home), and a workspace under a symlinked project root.
    • Off the event loop this swaps the bounded resolver-pool gate for the inline is_sensitive_resolved_path. The docs present it as the same gate list, so it is expected to widen, not narrow. I could not get a verdict to confirm that.

[SCOPE-REVIEWED] 2525ddb

@github-actions

github-actions Bot commented Nov 8, 2026 •

Copy link
Copy Markdown
Contributor

Opus 5.5 Review — ✅ no blocking findings

Reviewed 2525ddb10c15515d42b2590260a33a9de6a54ea6 — this comment is updated in place on each push.

Review details

No findings.

[OPUS-REVIEWED] 2525ddb

Verdict parsed from the review's SHA-scoped output markers for commit 2525ddb10c15515d42b2590260a33a9de6a54ea6.

False positive or not applicable? A repository writer can comment:
/ai-review override fable 2525ddb10c15515d42b2590260a33a9de6a54ea6 false-positive: <reason, 20 to 500 characters>

@github-actions github-actions Bot added merge conflict Branch has merge conflicts with its base — author must resolve before merge readiness: action required A blocking check or review needs attention and removed readiness: checking Automated validation is still running labels Nov 8, 2026
buluoray and others added 9 commits November 8, 2026 01:19
A spec declaring "tools": [] was not a ban on claude: the session got
only strictMcpConfig, so claude-agent-acp kept Claude Code's built-in
preset, and an allow rule in the user's or the project's settings ran
Bash, Write or WebFetch unasked. The kirocrew-lite helpers ran with
those tools, and the knowledge pool fell back to kiro-cli on claude.

A zero-tool claude session now sends options.tools empty beside
strictMcpConfig with bypass off, refuses a non-empty wire array, and
is stopped before its first prompt on an adapter below the
session-options floor. A live capture on claude-agent-acp 0.93.0 shows
the adapter reads the empty list as "no tool": under allow rules for
all three, no tool_call, no permission request, nothing on disk. On
that evidence claude joins ACP_BACKENDS_HONOR_ZERO_TOOL_BAN, which also
binds _deny_zero_tools there, and the knowledge pool keeps claude.

Closes #30057
(cherry picked from commit 515f616)

Originally authored on main by Bolin Chen (bolichen97); authored here with the public identity because the commit guard refuses the original address.
kiro-cli expands ${VAR} and ${env:VAR} in an MCP server's env, and
the agent materializer leaves those references for it. The harnesses
that take Crew's session mcpServers array start the server with the
element as written, so on codex, opencode, DeepSeek, Copilot CLI and
Strands (and ${env:VAR} on Claude Code) the server got the literal
text and a server that reads a token or URL from it listed no tools.

Resolve both forms in the shared session projection with the KAS
launcher's expander (same filtered view, same miss rule). goose writes
each server's env into its own session database, so its mirror keeps
the reference unresolved rather than put the value on disk.

(cherry picked from commit 2fa4077)

# Conflicts:
#	docs/system-specs/modules/agent-host-contract.md
#31147)

The side panel's tab-strip × and middle-click on an artifact tab closed
it directly, so an open inline comment draft was dropped without the
"Discard your unsaved comment?" prompt the panel's own close asks.

ArtifactPanel now exposes the same requestClose handle MarkdownPanel
does, running its comment-draft guard (and the host's beforePrompt only
when that guard asks), and the side panel registers it for artifact
tabs, so the chip close asks before discarding a draft.

Closes #28446
(cherry picked from commit e4f2a38)

Originally authored on main by Bolin Chen (bolichen97); authored here with the public identity because the commit guard refuses the original address.
…30833)

The substring and topic-overlap dedup branches of write_lesson retired
a stored lesson on text alone, so a consolidation rewording that shared
half its keywords with a user_explicit lesson replaced it and dropped
the user's specifics. Only the semantic branch weighed authority.

Every dedup delete site now applies the same authority rule the
semantic branch uses (user_explicit over another source, or strictly
higher stored confidence): an outranking row is kept and the
submission is stored beside it. A user_explicit write is unaffected.

Closes #9163
(cherry picked from commit f54bb35)

Originally authored on main by Bolin Chen (bolichen97); authored here with the public identity because the commit guard refuses the original address.
…-in roots (#31730)

The namespace launcher created every mask stand-in directly under a
host-shared root (/run/user/$UID or /dev/shm). On a host where the user
has no login session, systemd-logind's RemoveIPC=yes deletes the uid's
/dev/shm entries at the next session close, directories included.
Deleting a stand-in removed the private-window mount points inside it,
which detached the scratch window in the running sandbox, so $TMPDIR
vanished and every create under it failed with ENOENT.

The launcher now mounts a tmpfs on one stage directory in its own mount
namespace and creates every stand-in inside it through a descriptor on
the tmpfs root. Once the masks are placed it detaches the stage and
removes the empty host directory. A refused tmpfs mount keeps the old
host-root behaviour with a warning.

Co-authored-by: w <w@k.local>
(cherry picked from commit 782fcbd)

Originally authored on main by Bolin Chen (bolichen97); authored here with the public identity because the commit guard refuses the original address.
* fix(agents): send proxy-unsafe crew names in the request body

Fixes #19976

When a crew's config.json key contains a space (e.g. 'My Agent'), saving
from the dashboard editor calls PUT /api/agents/My%20Agent. A reverse
proxy that normalises percent-encoded paths decodes %20 to a literal
space, which is the HTTP request-line delimiter and causes a 502.

Fix: add PUT /api/agents (no path variable) that reads the crew name
from a 'name' field in the JSON body. The frontend uses this body-form
route automatically when encodeURIComponent(name) !== name, so a
space-containing key is never placed in a URL path segment a proxy can
misinterpret. The existing PUT /api/agents/{name} is unchanged for
slug-keyed agents.

The api_kirocrew_agent_update handler now serves both routes: it reads
the name from the URL path when present, or from the request body when
the path-variable route is not matched. The body name is used verbatim
(never stripped) so a stored key that differs only by edge whitespace is
not silently overwritten. Both forms share all update logic including
_require_owner, so auth and owner checks are identical.

change_card_catalog: add PUT /api/agents to HOOKED_ROUTES and
describe_manual_change so memory events are recorded for body-form
saves the same way they are for path-form saves.

Security: test_agents_endpoints_owner_auth.py already walks all routes
and verifies owner-gating; PUT /api/agents is added to its floor set.
test_agent_update_body_form.py pins auth parity between both routes.

* docs: add PUT /api/agents body-form route to crew-mode, learn-cron-dashboard, and feature map; update Pattern harvest and #20083 action

* fix(agents): keep proxy-unsafe crew names out of every agent route path (#28628)

Every crew route that names the crew in its URL path also answers under
/api/agents/b64/{name_b64}..., an unpadded base64url segment, so a
path-normalising proxy has no %20 to decode into a bare space (502).
The twins reach the same handlers behind the same owner gate. The
dashboard (delete, avatar upload and image, capabilities, worker kit)
and change-card Apply/Undo steps use the twin or the body-form PUT for
such names; the card hook reads a twin as its path-form route.

Fixes #20083

---------

(cherry picked from commit 4d0ae46)

Originally authored on main by Zezhen Xu (CrysisDeu); authored here with the public identity because the commit guard accepts only that identity.
…ate in the snapshot scan (#30909)

(cherry picked from commit 9f8d7d6)

Originally authored on main by Pearcekieser; authored here with the public identity because the commit guard accepts only that identity on a hand-resolved pick.
…ec (#32224)

A member start (session_capabilities.prepare_runtime, run under
asyncio.to_thread) resolves its workspace, its agent spec's name and the
spec's descriptor itself, then asks whether each path is sensitive.
Three of those checks handed the canonical path to is_sensitive_path,
which resolves it again on the bounded mc-pathres pool and fails closed.
On macOS the pool has no syscall sample, so one resolver child that
misses its deadline on a loaded runner is charged as a mount stall, and
every path under that prefix then reads as sensitive for 30 s. The
member start refused its own healthy spec as source_unreadable.

The three checks now use the canonical-path gate, which answers inline
off the event loop and stays bounded on it, with the same gate list:

- agent._spec_path_shape_refusal judges the spec's Path.resolve() with
  is_sensitive_canonical_path.
- config.loader_pipeline.default_project_dir judges the workspace's
  realpath with is_sensitive_canonical_path.
- hooks.safe_read_file_bytes_nolink takes canonical_gate, as
  safe_read_file already does: the realpath from validate_file_path and
  the descriptor's kernel path (identity and within-root checks) use the
  canonical gate. agent_capabilities._read_spec passes it. The
  keyword is passed on only when set, and the identity comparison
  itself is unchanged.

A sensitive path is refused exactly as before.

Closes #27388
(cherry picked from commit 123cc51)

Originally authored on main by Bolin Chen (bolichen97); authored here with the public identity because the commit guard refuses the original address.
…#32210)

On a Linux host whose libcups lacks the symbols the full Chrome for
Testing build imports (Amazon Linux 2: cups-libs 1.6.3), Kiro's
Chromium exits 127 at load and Browser Harness reports a 60 s daemon
timeout instead. Pick the same version's chrome-headless-shell there,
which links no libcups. Download, state and launch all follow
chromium.wanted_build, so a full build already on disk reads as not
downloaded and the startup auto-install fetches the shell. The probe
reads libcups' ELF dynamic symbol table from the linker's system
directories, so the status path still launches nothing.

Closes #32197

(cherry picked from commit 23b4795)
@buluoray
buluoray force-pushed the crew/slack/0.12.1-wide-2 branch from b23d90f to 2525ddb Compare November 8, 2026 01:22
@github-actions github-actions Bot added readiness: checking Automated validation is still running and removed readiness: action required A blocking check or review needs attention labels Nov 8, 2026
@buluoray
buluoray merged commit c318c7d into release/0.12.0 Nov 8, 2026
26 of 39 checks passed
@buluoray
buluoray deleted the crew/slack/0.12.1-wide-2 branch November 8, 2026 01:38
@github-actions github-actions Bot removed the readiness: checking Automated validation is still running label Nov 8, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

merge conflict Branch has merge conflicts with its base — author must resolve before merge

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants