Repository navigation
docs(computer-use): record the macOS check-to-launch window as a shortcut - #32811
Conversation
UX Review (Opus 5.5) — ⏭️ skippedRevision |
First Principles Review (Opus 5.5) — ✅ PASSPremise-level review of First-Principles-Verdict: PASS Nothing blocks. The change adds only a comment and a spec paragraph, and launch behaviour is the same as on main. Notes
What this change shipsInventory (2 items) — 1 justifiedIntent: Record the macOS window between the bundle check and the launch as an accepted, known risk, without changing how the launch works. This is an ADDITION (documentation only).
[FIRST-PRINCIPLES-REVIEWED] 0e85f34 Description read[DESCRIPTION-READ] cce4586d7d8e44cded1386d12288a2cb9db2b54b57ac2d35f7729c9246f2cb7a That sha256 names the PR description this verdict read. Recompute it with |
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS Nothing blocks: only a comment and a spec paragraph change, launch behaviour is the same as on Notes
[DESIGN-REVIEWED] 0e85f34 Description read[DESCRIPTION-READ] 18595b5271e83b8cbb80cf00b746b3e115694a603de94819976b48f34dcda9e1 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 |
Opus 5.5 Review — ✅ no blocking findingsReviewed Review detailsComment-only and doc-only changes. Nothing executable changed. No findings. [OPUS-REVIEWED] 0e85f34 Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
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: |
1feecf7 to
cba1804
Compare
|
span=88452e2064fb
|
|
🤖 Kiro Crew Auto-Pipeline [operator: buluoray]: macOS evidence for the Design lane's "Clears when" on head cba1804: the full macOS suite ran on this head ( |
bolichen97
left a comment
There was a problem hiding this comment.
Reviewed head cba1804. The fingerprint code reads cleanly, but I'm requesting one change of direction.
This adds new refusals to the existing macOS launch path, and this repo does not take added security strictness. On main, a bundle that passes resolve_target's checks launches. This PR adds three refusals on top of that:
resolve_targetraises_ERR_BUNDLE_CHANGEDwhen the_bundle_fingerprintread before_writable_componentdiffers from the one after it, or when it isNone(any of the fingerprinted files unreadable).spawn_detachedraisesOSErrorfor any path not in_verified_bundles. That is new for a caller that hands it a path without going throughresolve_targetin this process.spawn_detachedalso raises when the path's fingerprint changed since the resolve. That would include a legitimate app update landing between the resolve and the launch.
Please drop _verified_bundles, _bundle_fingerprint, _ERR_BUNDLE_CHANGED and the three checks, so resolve_target and spawn_detached behave as on main, and move the new tests out with them. If you want the rename-and-replace risk recorded, a shortcut: comment in spawn_detached naming the window, plus the spec note, is welcome without the refusal.
Separately, macOS backend shard 6 is red on test/test_cloud_ssm_cov80.py::TestInstallCommandHint::test_deb_one_liner_reaches_sudo_only_on_a_matching_download[True] (a sha256sum usage error). This PR does not touch that file, so I am not counting it against the change.
cba1804 to
de740f0
Compare
|
🤖 Kiro Crew Auto-Pipeline [operator: buluoray]: @bolichen97 done in de740f0, following your direction. |
|
🤖 Kiro Crew Auto-Pipeline [operator: buluoray]: The frozen Goal was amended by a human maintainer, not by the pipeline. It now reads: "Record the macOS check-to-launch bundle-swap window in |
bolichen97
left a comment
There was a problem hiding this comment.
Approving head de740f0. The refusals are gone, so resolve_target and spawn_detached behave as on main, and the window is recorded as a shortcut: comment in spawn_detached plus the spec paragraph.
The red macOS shard 6 is the BSD sha256sum test that #32915 fixed on main. Could you rebase onto main? The push will dismiss this approval, and I will re-approve once it is green.
…tcut spawn_detached launches the bundle by the path resolve_target verified, so an admin user can swap the bundle at that path between the checks and LaunchServices reading it. The maintainer accepted that window as a documented shortcut rather than a new refusal, so the launch path stays as on main: a shortcut comment in spawn_detached names the window and the upgrade (launch by a file reference), and computer-use.md records it.
de740f0 to
0e85f34
Compare
Problem / Motivation
Goal: Record the macOS check-to-launch bundle-swap window in
computer_launch_appas an accepted, documented shortcut, with launch behaviour unchanged from main.A stock
/Applicationsisdrwxrwxr-x root admin, so_writable_componentskips the parent probe there and judges the bundle by its own directories. But every admin user can still rename an entry in it. Betweenresolve_target(the verification) andspawn_detached(/usr/bin/open -a <path>),run_launchruns the policy check and the running-app lookup. An agent's shell child running at the same time can rename the verified, root-owned bundle aside and put its own bundle at the same path, andopen -athen launches agent-written code outside the agent sandbox. Found by GPT on the 0.12.1 port of #31408 (#32791); the same code is onmain.Why it matters
The launch check exists so the agent cannot use this verb to run code it wrote. This gap reopens that for every admin user (the default for a Mac's main account).
Not a goal
/Applicationsrule.shortcut:comment).What changed (motivation → approach → change)
The first heads closed the window with a bundle fingerprint and three new refusals. The maintainer asked for the opposite direction in this review: the repo does not take added security strictness on this launch path, so the launch behaves as on
mainand the window is recorded as an accepted, documented shortcut instead.The diff is now 12 added lines:
src/kiro_crew/computer_use/launch_macos.py: ashortcut:comment inspawn_detachednaming the check-to-launch window (the launch goes by path; an admin user can swap the bundle at that path before LaunchServices reads it) and its upgrade (launch by a file reference held from the check to the spawn).docs/system-specs/modules/computer-use.md: one paragraph recording the window as accepted, and that closing it means a file-reference launch, not a refusal on top of the path launch.No fingerprint, no
_verified_bundles, no new refusal, no test change.resolve_targetandspawn_detachedbehave exactly as onmain.Backwards compatibility
Compatible: no code path changes; only a comment and a spec paragraph are added.
Removes nothing.
Tests
Proof: not provable -- other: comment and spec only, no behaviour change
test/test_computer_use_launch.pyandtest/test_computer_use_launch_macos_stock.pyare unchanged frommainand pass (108 passed, 7 skipped).scripts/docs-lint.sh, the comment-history and brand gates pass.Manual verification
N/A - no behaviour change.
Screenshots / video
N/A - no frontend file changes.
Related Issues
Refs #31408, #32791.
Sibling sweep
Siblings: none -- the only check-to-launch path on macOS is
spawn_detached; Windows resolves through its own catalog.Pattern harvest
Not generalizable: a single documented shortcut.
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.