Skip to content

fix: backport 0.12.1 tightening batch (sign-in, credential, trust, sandbox) - #32825

Merged
buluoray merged 11 commits into
release/0.12.0from
crew/slack/0.12.1-wide-sec
Nov 8, 2026
Merged

buluoray merged 11 commits into
release/0.12.0from
crew/slack/0.12.1-wide-sec

Conversation

@buluoray

@buluoray buluoray commented Nov 8, 2026 •

Copy link
Copy Markdown
Contributor

Problem / Motivation

Goal: Bring twelve sign-in, credential, trust and sandbox fixes, merged on main after v0.12.0, onto release/0.12.0, with the one main commit one of them needs first, one commit each, in a PR of its own.

Why it matters

Each fix closes a hole a 0.12.0 user is exposed to today, or a sandbox fault that breaks their sessions. Every pick applied to release/0.12.0 with only the doc and test hunks listed under "What changed" resolved by hand, so the code each one replaces is on release.

PR Reported by What a 0.12.0 user sees today Faulty code on release
#30631 bolichen97 On Windows, the pod port scan opens a pod's port file through a link, unlike Linux and macOS. pod/runtime_ports.py
#30905 CrysisDeu (#29743) Opening the dashboard from its sign-in link can leave the token in the address bar and browser history while first-run setup shows. pages/useChatPageSlotSwitches.ts, main.tsx
#30828 chenmingwei23 (#24790) A remote crew chat can send the crew's credential over a connection that does not end at the crew. dashboard/handlers_crew_turn.py
#30784 bolichen97 (#14439) A sign-in link that was printed or posted but never opened still works as a signed-in session until its session lifetime ends. dashboard/token_auth_tokens.py, dashboard/token_auth.py
#30864 Pearcekieser (#24998) After an app's trust is revoked, later sub-agents keep auto-approving from a cached policy. dashboard/chat_utils_slots.py
#30783 iamwhatever (#22234) A sandbox read-only seal on a file can be undone by renaming or replacing a folder above it from inside the sandbox. sandbox_workspace.py, sandbox_launcher_parts/mask_stages.py
#30795 bolichen97 On macOS the sandboxed agent can connect to host credential sockets (/tmp/mcs-*.sock) and the tunnel sidecar socket. sandbox_plan.py, sandbox_seatbelt.py
#30826 chenmingwei23 (#18264) A ?workspace= selector on /api/lessons is not checked against the caller, so one workspace can read or write another's lessons. dashboard/handlers/cron_lessons.py
#30955 Pearcekieser (#25397) A sub-agent that is already running keeps auto-approving after its parent's trust is revoked. subagent trust carry, dashboard/chat_utils_slots.py
#30437 CrysisDeu (#29773) The Security page says "Interactive" while YOLO auto-approve is active. Also the prerequisite of #31602. dashboard status frame
#31602 buluoray A run as root aimed at another account's data home leaves that home broken, with no hint how to repair it. data_home_owner check (new), cli/doctor
#22916 slangerss (#22881) On a host with no per-user systemd manager, .env (and with it SLACK_BOT_TOKEN) can be truncated to 0 bytes when a sandbox bind source is torn down. sandbox_plan.py runtime-dir order
#32262 bolichen97 (#32256) On macOS the Strands harness dies at boot: realpath of its --mcp-config in the masked staging dir gets EPERM. sandbox_seatbelt.py

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 change to any client outside this repository.

What changed (motivation → approach → change)

Each main squash commit was cherry-picked with -x onto release/0.12.0 @ 3c6a160c8a, in merge order on main, one commit each.

# PR main commit release commit
1 #30631 tighten a file access check on windows 9489ca054d 199ecb2979
2 #30905 keep the sign-in token out of the address bar on every screen 6c34be0e06 830f9b89f4
3 #30828 tighten a remote crew credential check ba1ac94302 da584261e5
4 #30864 stop sub-agents auto-approving after an app revokes trust 1a40de6098 d2027a348d
5 #30783 hold sandbox read-only guards in place d0e6f36927 6f5967ac9d
6 #30795 deny host credential and sidecar sockets on macOS 91d4e26994 6adfb5dfae
7 #30826 keep lesson reads and writes inside the caller's workspace f02c887f9a 914ad70f74
8 #30955 stop a running sub-agent auto-approving once trust is revoked cfaf8aa88d 8060f4ea4a
9 #30437 carry the auto-approve grant on the dashboard status frame 41e207d472 332bbf0710
10 #22916 stage sandbox bind sources on a durable tmpfs without a user manager e21a28032b da761c289b
11 #32262 let realpath reach a file exposed from a seatbelt mask 856c11cf10 9f8115fa37

#30784 and #31602 are not carried here. #30784 was dropped by maintainer decision: the Crew Tunnel app it could refuse was not tested. #31602 was dropped because the GPT lane found a real security defect in its main code: the sudo find ... -exec chown -h {} + repair command in data_home_owner.py can be raced into changing ownership of a system path. It is fixed on main first. #30437 stays as a fix of its own (#29773). Their rows under "Why it matters" stays as written when the PR opened.

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

Every other pick applied cleanly. #30631 is authored with the public identity because the commit guard refuses the original address on a hand-resolved pick; its message names its original author. #22916's co-author trailer carried an internal host address and is rewritten as a plain "Co-author on main" line.

Backwards compatibility

Breaking: these picks refuse things that worked before. The table names each one, who could be refused, and how that was checked. "Writer sweep" means every caller in this repository on this branch.

PR Could be refused How it was checked
#30828 A remote crew turn whose connection does not end at the crew. Every crew the product launches is registered and connected the way the check expects. The crew-turn proxy tests (160) pass.
#30905 Nothing is refused. The token leaves the address bar earlier, and the signed chat link flow still receives it. urlToken, ChatPage and useChatPageSlotSwitches vitest pass.
#30826 A caller other than the operator's browser that names a workspace other than its own. On a list it gets its own workspace's lessons; a write is refused. The dashboard Memory tab, the agent's lesson tools and the CLI send no workspace selector, so they are unchanged. The lesson and workspace tests pass.
#30864, #30955 A sub-agent whose app or parent trust was revoked now asks instead of auto-approving. Only paths that already take trust away change, and each now ends where the dashboard toggle already ends.
#30783 Inside the sandbox, renaming or removing a folder above a guarded path. A move into or out of one fails the way a move across disks does, so mv and shutil.move copy instead. Sandbox launcher and plan tests pass. test_seal_parent_pin_real_sandbox.py passes here and at the main commit.
#30795 On macOS, connects from inside the sandbox to /tmp/mcs-*.sock and the tunnel sidecar socket. git over SSH keeps working, because mcs-agent.sock stays reachable whenever the ssh agent is forwarded (standard and cc tiers). Nothing in this repo connects to the denied names from inside a sandbox, and the gateway API socket matches no pattern. Linux plans are unchanged.
#30631 On Windows, a pod port file reached through a link is now refused, matching Linux and macOS. The caller already treats a refused read as nothing to read. Pod tests (1839) pass.

#30437, #22916 and #32262 refuse nothing: they add status fields, reorder two already-trusted bind-source candidates, and add file-read-metadata allows.

Removes nothing: no setting, route or stored key is removed.

Tests

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

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 pick (#30905) removes the token from the address bar before the first screen draws; no rendered pixel changes.

Related Issues

Backport to the 0.12 line for 0.12.1 of #30631, #30905, #30828, #30864, #30783, #30795, #30826, #30955, #30437, #22916 and #32262. Source issues: #29743, #24790, #24998, #22234, #18264, #25397, #29773, #22881, #32256.

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 01:18
@buluoray
buluoray requested review from tim-finnigan and removed request for a team November 8, 2026 01:18
@github-actions

github-actions Bot commented Nov 8, 2026 •

Copy link
Copy Markdown
Contributor

Sensitive Change Review

Sensitive files changed:

  • src/kiro_crew/sandbox.py

  • src/kiro_crew/sandbox_argv.py

  • src/kiro_crew/sandbox_launcher_parts/mask_stages.py

  • src/kiro_crew/sandbox_launcher_program.py

  • src/kiro_crew/sandbox_plan.py

  • src/kiro_crew/sandbox_seatbelt.py

  • src/kiro_crew/sandbox_workspace.py

  • No reasoning template for 9f8115f 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: 9f8115fa37ba352fe393a6b77279e0ec15b7e0e1
- 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

UX-level review of 9f8115fa37ba352fe393a6b77279e0ec15b7e0e1 — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS does not.

UX-Verdict: PASS

Nothing to check. The diff adds no dashboard control, label or state. The frontend change only takes the sign-in ?token= off the address bar earlier during boot, so the lack of screenshots is not an evidence gap.

[UX-REVIEWED] 9f8115f

Description read

[DESCRIPTION-READ] ae6d7ec16fa67168d24cee275e1977bdb48b10a1021b5ead0d12eaeda3ee49bf

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 (a BLOCK with no Trigger and no Outcome does not block)

Design-level review of 9f8115fa37ba352fe393a6b77279e0ec15b7e0e1 — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS does not.

Design-Verdict: PASS

#31602 isn't in this diff, but the description says it was picked. A 0.12.1 built from this branch would still break a person's data home on a root run and give no repair hint.

Blockers

Phantom pick: #31602 (refuse root runs on a person's data home)
Trigger: The description lists "| 10 | #31602 refuse root runs on a person's data home and name the repair | 64beae3ae3 | 7561166cbb |" and says its crew-mode.md paragraph was hand-resolved with "main's text about the 409 data_home_not_owned refusal". The patch has no data_home_owner check, no cli/doctor change and no crew-mode.md hunk, and data_home_not_owned appears nowhere in the tree at HEAD.
Outcome: The pick is missing → the root-run path stays as it is on 0.12.0 → anyone who runs as root against another account's home on 0.12.1 still ends up with a broken home and no repair hint, while this PR (and any release notes built from it) records the fix as shipped. It also leaves #30437's "prerequisite of #31602" framing with nothing that depends on it.
Fix: add the #31602 cherry-pick, with its crew-mode.md resolution, as its own commit, or take #31602 out of the description the way #30784 was taken out.
Clears when: the diff contains #31602's product code and its data_home_not_owned refusal, or the description no longer claims #31602.

Notes

  • The description is cut off inside ## Backwards compatibility, so the take-away check was not judged against readers listed past the cut.

[DESIGN-REVIEWED] 9f8115f

Description read

[DESCRIPTION-READ] 4fd05a419bf46950c91345484989639cd675125d770cc9ff3a69575f42e00d71

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 9f8115fa37ba352fe393a6b77279e0ec15b7e0e1 — 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 backports fixes that main already merged. Nothing in the diff goes beyond those picks. One gap between the description and the diff is noted below.

Notes

What this change ships

Inventory (9 items) — 8 justified

Intent: give 0.12.0 users sign-in, credential, trust and sandbox fixes that already merged on main. This is a FIX (backport).

  1. The Windows pod port scan no longer follows a link to a port file — justified
  2. The sign-in token leaves the address bar on every screen — justified
  3. A remote crew turn refuses to send the credential over a connection that does not end at the crew — justified
  4. Sub-agents stop auto-approving after an app's trust or the parent's trust is revoked, including ones already running — justified
  5. A sandbox read-only seal holds when a folder above it is renamed or replaced — justified
  6. On macOS, sandboxed code can no longer connect to mcs-* credential sockets or the tunnel sidecar socket — justified
  7. /api/lessons?workspace= is checked against the caller; a write to another workspace gets 403 workspace_not_owned — justified
  8. The dashboard status frame carries the YOLO grant to dashboard users only; the sandbox stages bind sources on /dev/shm when no user manager is running; the Strands harness can realpath its exposed config on macOS — justified
  9. The fix: refuse root runs on a person's data home and name the repair #31602 root data-home refusal is claimed but missing from the diff — undeclared

[FIRST-PRINCIPLES-REVIEWED] 9f8115f

Description read

[DESCRIPTION-READ] c7b34cba7654a64a8e3b4d5561bc222d53023d212141c8d4c7c401b32d32d312

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 9f8115fa37ba352fe393a6b77279e0ec15b7e0e1 and found no blocking issues.

This comment is updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] 9f8115f

False positive or not applicable? A repository writer can comment:
/ai-review override gpt 9f8115fa37ba352fe393a6b77279e0ec15b7e0e1 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 d8a4bfd08dd353f2e02f603d668c1e0691f13ad1 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): 14 classified, 0 newly refused, 12 still allowed.
  • ubuntu-latest (platform posix): 14 classified, 0 newly refused, 12 still allowed.
  • windows-latest (platform windows): 6 classified, 0 newly refused, 5 still allowed.

No candidate flipped from allowed to refused.

Scope-Verdict: PASS

Reason: Nothing this diff tightens is in the shell deny composite: no hunk touches hooks.py, SHELL_DENY_TIERS, sensitive-bash, exfil or deny-rules. So no candidate can be confirmed newly refused, and none was handed back as confirmed. The real tightenings are a sandbox mount and Seatbelt change, a session-cookie check, a root-run CLI refusal and a revocation check on subagent approval. The shell classifier sees none of them, so they are named under UNADJUDICATED. Two of them have plausible collateral I could argue but not confirm: a root container that bind-mounts a host user's data home, and link-as-cookie callers outside the paths this diff updated.

TIGHTENING: Read-only seal parents can no longer be renamed. sandbox_argv._live_plan_host now gathers every entry that the run dir, the kiro agents tree, the crew-home ceilings and the install root resolve through into readonly_ancestor_guards. The namespace launcher binds each one over itself (pin_seal_parents: _mount_or_die(launch, _pin_target, _pin_target, _MS_BIND | _MS_REC, "pinning seal parent %s" % d)), and Seatbelt write-denies each as a literal. -- The ONE threat it refuses is the one the security.md hunk names: "a renamed parent carries the namespace launcher's bind away with it and moves the object out from under a Seatbelt path rule, and either way the sealed name is free for the agent to fill with what the unsandboxed gateway reads next".

ADJUDICATED: 20 candidates, 0 confirmed newly-refused

PLATFORM COVERAGE: I wrote posix and windows rows, and no any rows, because every row names an executable or a path.

  • posix (14 rows): These cover what the diff assumes on POSIX:
    • Linux namespace pins: git mv inside a worktree, a mv out of ~/.kiro/agents, and a venv pip install -e ..
    • macOS Seatbelt: the socket denies (git clone and ssh -T over SSH, ssh-add -l) and the exposed_ancestors stat allow (realpath ~/.aws/config).
    • The POSIX-only data-home check: kirocrew doctor, sudo kirocrew stop, stat on config.json, and the exact sudo find ... -exec chown -h ... && chmod -R u+rwX repair line doctor prints.
    • The token change: kirocrew token and the curl .../api/status?token= exchange.
  • windows (6 rows): git mv with \ separators, SSH git clone, kirocrew.exe doctor, kirocrew.exe token, curl.exe with %VAR% and NUL, and .venv\Scripts\pip.exe. They cover the diff's Windows assumptions: the data-home check returns "nothing wrong" there, and the token exchange is platform-neutral.
  • Not separated: The diff assumes Linux-namespace behaviour and macOS-Seatbelt behaviour separately, and the candidate schema has only posix. So no row targets one of those two backends on its own; whether each POSIX leg ran namespace or Seatbelt is for the per-leg fold to report.

UNADJUDICATED:

  • Seal-parent pins (namespace launcher). Call path: wrap_argv → _live_plan_host (_resolution_entries + _renameable_ancestor_guards) → _namespace_plan.runtime_ancestor_guards → namespace_payload["parent_pins"] → place_masks → pin_seal_parents. Each pinned directory is its own mount, so a rename(2) or link(2) that crosses one fails with EXDEV. The diff says so itself: "a bare os.replace or git mv across one fails".
    • Pins plausibly include $HOME/.kiro, $HOME/.kiro/crew, and the user-owned ancestors of a venv's site-packages.
    • Inputs I would have submitted to a sandbox-level classifier: a bare os.replace("~/.kiro/crew/x.tmp", "~/x") from a tool that does not fall back to copying; git mv of a file across a pin; a hard-linking installer (uv with link-mode=hardlink) writing into a pinned site-packages; pip uninstall stashing a package from site-packages into /tmp.
    • A spawn fails outright ("changed while it was being pinned") if a pinned name is retargeted between the bind and the re-check.
  • Seatbelt ancestor write-denies. Call path: _seatbelt_plan → ancestor_guards (now every readonly_ancestor_guards entry, not just the install root's). The literal file-write* denies now plausibly cover $HOME, ~/.kiro and ~/.kiro/crew themselves. Inputs: touch ~/.kiro (utimes), chmod 700 ~/.kiro/crew, mv ~/.kiro ~/.kiro.bak.
  • Seatbelt socket denies. Call path: _live_plan_host → denied_socket_patterns / credential_agent_socket_patterns → render_seatbelt_profile (network-outbound and file-write*/file-link regex denies). Inputs:
    • SSH through IdentityAgent /tmp/mcs-agent.sock on a strict spawn without ssh_auth_sock_consent.json, which the diff accepts as a cost.
    • A legitimate developer tool that creates its own /tmp/mcs-<name>.sock: it can neither bind nor rename that name.
  • Session-cookie kind=session gate. Call path: token_auth_middleware → validate_token_with_app(use_session_exp=True) → LINK_NOT_A_SESSION. Inputs: any client that mints at /api/token/local or runs kirocrew token and sends the result straight as mc_token_<port>. That includes an App Kit client given token=<link>, which the api-reference hunk now says is refused. The base tree has many minters (website/electron/local-token.js, memory_forget_client.py, app_lifecycle_client.py, cron_script_context.py, cli_server_requests.py); only the paths this diff changes were moved to the exchange.
  • Root-run refusal. Call path: cli.main → data_home_owner.root_run_refusal (exit 1 unless the command is service or stop). It refuses whenever euid is 0 and the nearest existing ancestor of the data home belongs to a non-root uid. Inputs: sudo kirocrew doctor, sudo kirocrew update, and especially a container running as root that bind-mounts a host user's data home (docker run -v "$HOME/.kiro:/root/.kiro" ... kirocrew gateway), where the mount is owned by the host uid.
  • Lessons workspace ownership (doc hunk). Call path: /api/lessons POST and GET → _lesson_caller_slot_key. A non-operator caller that names another slot's workspace is now refused with 403 workspace_not_owned, or no longer sees that workspace's rows. Input: an internal caller listing or adding a lesson for a workspace it does not own.
  • Revoked parent trust. Call path: resolve_parent_policy → _approval_lineage → parent_trust_revoked → ParentPolicyAuto.offer. Running subagents lose parent_policy=auto when a stamp changes. Inputs: a long-running child whose parent's trust is suspended and restored by the yolo ceiling around a governance read, and a continuation run on a subagent: key whose lineage is None.

[SCOPE-REVIEWED] d8a4bfd

@github-actions

github-actions Bot commented Nov 8, 2026 •

Copy link
Copy Markdown
Contributor

Opus 5.5 Review — ✅ no blocking findings

Reviewed 9f8115fa37ba352fe393a6b77279e0ec15b7e0e1 — this comment is updated in place on each push.

Review details

No findings.

[OPUS-REVIEWED] 9f8115f

Verdict parsed from the review's SHA-scoped output markers for commit 9f8115fa37ba352fe393a6b77279e0ec15b7e0e1.

False positive or not applicable? A repository writer can comment:
/ai-review override fable 9f8115fa37ba352fe393a6b77279e0ec15b7e0e1 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
buluoray force-pushed the crew/slack/0.12.1-wide-sec branch from d8a4bfd to 442b720 Compare November 8, 2026 04:12
@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 commented Nov 8, 2026

Copy link
Copy Markdown
Contributor Author

Dropped #30784 (dashboard sign-in link tightening) from this batch by maintainer decision: the Crew Tunnel app it could refuse was not tested. Rebased onto release/0.12.0 @ 3c6a160; new head 442b720 carries the other 12 picks. The Goal line and the #30784 row under Why it matters are left as written when the PR opened (frozen); What changed and Backwards compatibility describe the current diff.

@github-actions github-actions Bot added readiness: action required A blocking check or review needs attention and removed 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
buluoray and others added 8 commits November 8, 2026 04:45
One read the CLI makes on its own behalf documents that it refuses a
link planted where it reads, and it does on Linux and macOS. On Windows
it did not, so the documented refusal held on only two of the three
platforms. It now holds on all three, through the helper the rest of
the codebase already uses for this, and its test runs on Windows hosts
that can create links instead of being skipped there.

Builds on #26701 by @leonlaiyc.

Part of #9731

Co-authored-by: Leon <276034861+leonlaiyc@users.noreply.github.com>
(cherry picked from commit 9489ca0)

Originally authored on main by Bolin Chen (bolichen97); authored here with the public identity because the commit guard refuses the original address.
…30905)

A dashboard opened from its sign-in link could keep the token in the
address bar, and in history, while a screen shown before the chat
page was up, first-run setup among them. The token now leaves the
address bar as soon as the page loads, whichever screen comes first,
and a signed chat link still opens its prompt.

Closes #29743

(cherry picked from commit 6c34be0)
Co-authored-by: Pearce Kieser <5055971+Pearcekieser@users.noreply.github.com>
(cherry picked from commit 1a40de6)
A read-only sandbox guard could be sidestepped from inside the
sandbox. It now holds on both sandbox backends, and nothing else the
agent may write changes.

Part of #22234

(cherry picked from commit d0e6f36)
The macOS Seatbelt profile is (allow default) plus file denies, so a
sandboxed agent could connect() to any UNIX socket its uid can reach.
Some host listeners trust the caller's uid alone: the local credential
agent's sockets in /tmp and the tunnel sidecar's socket in the data
home's run dir. Reaching either lends the agent an unsandboxed
process's authority.

The plan now carries denied_socket_patterns (Seatbelt only) and the
profile renders one network-outbound unix-socket deny per pattern.
The gateway's API socket is deliberately not denied: in-sandbox MCP
servers reach the gateway through it, and a Seatbelt deny is EPERM,
which loopback_urlopen does not fall back from.

(cherry picked from commit 91d4e26)
…30955)

A sub-agent that started under its parent's trust kept auto-approving
its tool calls after a person revoked that trust, and so did any
sub-agent it had started. Its next tool request now takes the normal
approval path instead, and a re-trust gives the grant back.

Builds on #25398 by @Pearcekieser.

(cherry picked from commit cfaf8aa)
bolichen97 and others added 3 commits November 8, 2026 04:45
The Security page read "Interactive" while auto-approve was on: each
WebSocket status frame replaces the stored status, and the frame had
no grant fields, so status.yolo went undefined seconds after load.

The dashboard user's frame now carries the grant fields /api/status
reports, from one shared reader, with the expiry in whole seconds so
an unchanged grant leaves the frame unchanged. An app socket's Tier-0
frame stays without them.

Closes #29773

(cherry picked from commit 41e207d)
…ager (#22916)

* fix(sandbox): stage bind sources on a durable tmpfs without a user manager

The namespace launcher stages the empty sources it binds over .env and the
other HIDDEN crew-home leaves on /run/user/$UID first. That runtime dir is
kept coherent only by a per-user systemd manager; on a host without one,
logind tears it down and remakes it across session churn, and a bind source
pinned inside it when that happens leaves the host leaf a bind of a now-empty
inode -- .env truncated to 0 bytes, silently dropping SLACK_BOT_TOKEN.

Order the launcher's stand-in roots so the host-wide, manager-independent
/dev/shm leads the volatile /run/user/$UID whenever no per-user manager keeps
the runtime dir coherent, detected host-side by the systemd subdirectory the
manager creates in it. Both roots stay in the list, so the launcher's cross-fs
check and its mid-create move to the next candidate are unchanged, and every
HIDDEN leaf stays bind-masked on whichever root. Where a manager IS running,
/run/user/$UID still leads (login-scoped, auto-reclaimed at logout).

* fix(sandbox): judge the staging root by its own manager socket, not the session bus

Review (bolichen97): a listener at XDG_RUNTIME_DIR/bus proves nothing about a
per-user systemd manager -- a plain dbus-daemon serves it -- and XDG_RUNTIME_DIR
can name a different directory from the /run/user/$UID the bind sources stage on.

_runtime_dir_is_durable now takes the staging root itself
(sandbox_plan.user_runtime_root, the one spelling namespace_payload also lists)
and connects only to <root>/systemd/private, which only the manager creates.
Tests cover a live bus with no manager and a live manager under a different
XDG_RUNTIME_DIR.

---------

Co-author on main: Kyle Seaman.
(cherry picked from commit e21a280)
…32262)

On macOS the Strands harness exits at boot with EPERM on realpath of
its --mcp-config, which Crew stages in the masked auth-store-staging
directory and hands back with a require-not literal carve-out. The
subpath read deny still covers the masked directory itself, and
realpath lstats every component, so a Node CLI cannot canonicalize a
file it is allowed to read.

Each Seatbelt mask now records exposed_ancestors, the masked
directories between the mask root and each exposed file, and the
profile allows file-read-metadata on exactly those literals after the
mask's last deny, as it already does for a private window. The
directories stay unlistable and their siblings unreadable.

Closes #32256

(cherry picked from commit 856c11c)
@buluoray
buluoray force-pushed the crew/slack/0.12.1-wide-sec branch from 442b720 to 9f8115f Compare November 8, 2026 04:52
@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 commented Nov 8, 2026

Copy link
Copy Markdown
Contributor Author

Dropped #31602 (data-home owner check) from this batch. GPT 6.1 flagged a security-class finding on 442b720 that holds against the main code too: the repair command sudo find <home> -xdev ... -exec chown -h <user> {} + (src/kiro_crew/data_home_owner.py:81-84, unchanged on main) is pathname-based, so an agent writing to the user-owned data home can swap a parent directory for a symlink between the scan and the chown and get a system path re-owned. It gets fixed on main first. #30437 stays as a fix of its own (#29773). New head 9f8115f carries 11 picks.

@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 38c3f6d into release/0.12.0 Nov 8, 2026
22 of 35 checks passed
@buluoray
buluoray deleted the crew/slack/0.12.1-wide-sec branch November 8, 2026 05:31
@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.

3 participants