Repository navigation
Conversation
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A desktop prompt's content-free push also reaches devices marked for computer use. A store that cannot name owners sends no approval push. Three tests that used ordinary devices or fakes without the new store method updated on purpose. Co-Authored-By: Claude Opus 5.5 <[email protected]>
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Co-Authored-By: Claude Opus 5.5 <[email protected]>
…ush offers no Approve Built in: desktop prompts, hold lifts, arguments cut or masked, long descriptions. Owner lists push.critical_tools/commands/paths, defaulting to deletes, git push, sends, spends and config/device/permission changes. One value from serialize_pending for REST, the frame and the push; the fanout key gate gains exactly 'critical', on purpose. server_capabilities: approval_critical. Co-Authored-By: Claude Opus 5.5 <[email protected]>
…fig, downloads (review round 1) Co-Authored-By: Claude Opus 5.5 <[email protected]>
…mmands and bash writes into config Wrappers with options (sudo, env -i, nice -n, xargs, find -exec), inline interpreters (bash -c, python -c, eval), find -delete, subshells, groups and if/then are matched; a shell line naming a config path is critical; downloads into config are, the downloads folder is not; a relative path is judged by its components; a call the matcher cannot read is critical. One test fake fixed (an empty dict was never read). Review round 1 on #917. Co-Authored-By: Claude Opus 5.5 <[email protected]>
…case, dynamic command words, cd writes (review round 2) Co-Authored-By: Claude Opus 5.5 <[email protected]>
Shells are always critical; interpreters with an inline program (grouped flags too), '-', or fed on stdin; python -m is scanned; after a wrapper every later word is tried and a command-line argument is matched whole; a command word built from $, a glob or a brace is critical; names compare case-folded; a write relative to a cd into home or config, a glob that could reach config and XDG_CONFIG_HOME count; more senders. Review round 2 on #917. Co-Authored-By: Claude Opus 5.5 <[email protected]>
…, spends, versioned interpreters (review round 3) Co-Authored-By: Claude Opus 5.5 <[email protected]>
…nd knows more deletes, sends and spends A wrapper met inside a wrapper's scan is not scanned again (was exponential: 'env '*22 hung). Pattern words after the verb match anywhere (git reset -q --hard); versioned interpreters (python3.12), pwsh; cd -- handled; deletes (git checkout ., worktree remove, tag -d, docker rm/prune, npm unpublish, dd, :), sends and spends (publish/push commands, rclone, gsutil, httpie, kubectl and terraform changes, wrangler, vercel, fly, heroku, shortcuts), chgrp. Review round 3 on #917. Co-Authored-By: Claude Opus 5.5 <[email protected]>
…cation and data deletes (review round 4) Co-Authored-By: Claude Opus 5.5 <[email protected]>
…es are critical A shell line over 4 KB or 400 words is critical unread (bounds every quadratic shape); find -exec recursion goes through the scan's guard (nested find chains were exponential); '> f' at the head of a statement is critical; docker volume prune / compose down, git switch --discard-changes, checkout -f and package uninstalls added. One test corrected: a short harmless find chain is not critical, only fast. Review round 4 on #917. Co-Authored-By: Claude Opus 5.5 <[email protected]>
…eview round 5) Co-Authored-By: Claude Opus 5.5 <[email protected]>
'> f', '1> f', '>| f' to anything but /dev/null, /dev/std*, /dev/tty or a descriptor; appends are not. The path check is skipped above the length cap (the line is critical already). One earlier expectation updated on purpose: 'cd project && echo hi > out.txt' is now critical (an overwrite), and the ordinary case uses '>>'. Review round 5 on #917. Co-Authored-By: Claude Opus 5.5 <[email protected]>
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Checked on the whole line, since the statement split cuts at '&'. Review round 6 on #917. Co-Authored-By: Claude Opus 5.5 <[email protected]>
…not a security guard Co-Authored-By: Claude Opus 5.5 <[email protected]>
…iew LOW) Co-Authored-By: Claude Opus 5.5 <[email protected]>
…read count as cut Co-Authored-By: Claude Opus 5.5 <[email protected]>
…revoked marked device and a raising store From the second review of this PR. Co-Authored-By: Claude Opus 5.5 <[email protected]>
…o-owner-devices # Conflicts: # CHANGELOG.md
…o-owner-devices # Conflicts: # CHANGELOG.md
…ices' into feat/approval-critical-flag # Conflicts: # CHANGELOG.md # docs/reference/config-keys.md # src/prometheus/web/ws_server.py
…o-owner-devices # Conflicts: # CHANGELOG.md
…ices' into feat/approval-critical-flag # Conflicts: # CHANGELOG.md
…l-flag # Conflicts: # src/prometheus/web/ws_server.py
OAraLabs
changed the base branch from
fix/approval-pushes-to-owner-devices
to
main
November 15, 2026 05:14
OAraLabs
marked this pull request as ready for review
November 15, 2026 05:14
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Beacon iOS Piece 15. Stacked on #915; retarget to
mainonce it merges.What this is, and what it is not
criticaldecides where an approval request is answered (in the app, or from a lock-screen notification), not whether. It is a UX nudge, not a security boundary, and nobody should treat the lists as a guard. Every request it touches has already been stopped by the permission checker and needs a person either way. A request wrongly judged ordinary is still read by the owner and approved on the lock screen behind Face ID, which the owner could equally do in the app.The lists are best effort, and they are not a second permission checker. Two things ARE hard rules:
CriticalRules.matchesanswersunreadable, and a shell line over 4 KB or 400 words is critical unread.The same statement is in
permissions/critical.py, the config template,docs/guide/api.mdand the CHANGELOG.What it does
ApprovalQueue.serialize_pendingcarriescritical. It is computed once, soGET /api/approvals, theapproval_pendingframe and the push cannot disagree.Built in (always critical):
Owner lists:
push.critical_tools,push.critical_commandsandpush.critical_paths. Leave a key out for the default;[]means none.git push, commands and tools that send or spend, and changes to devices, permissions or the agent's configuration.cd, failing toward critical on anything it can't read. Overwriting redirects count too.The push: a critical request is pushed with no
category, so the notification offers no Approve, and with"critical": true. It is never sent asfalse. Every other approval push keepsAPPROVAL.test_approval_push_fanoutgains exactly one key,critical, on purpose.Capability:
server_capabilitieslistsapproval_critical.Review
Independent reviews in rounds, each round's red tests committed before its fix. They found two runtime blowups, both fixed:
find -execchains were exponential.There is now a size cap. A 4 KB line runs in about 0.1 s or less, and
matchescannot raise. The remaining findings were matcher false negatives; per the decision above they are not blocking, and the rounds were stopped there.Tests
tests/test_approval_critical.py(190). Local Mac suite: 16,102 passed, 0 failed (test_generated_referenceandtest_gate_manifestdeselected: they fail only in a worktree). Ruff, the mypy gate and pip-audit are clean.No released client acts on these pushes yet.
🤖 Generated with Claude Code