Complete the bounded Computer Use delta disposition

[nisavid]

Parent

Chart official-app parity and fallback sunsetting

What to build

Give both pending Linux Computer Use synchronization reminders one bounded final disposition. Preserve only improvements that are genuinely reusable in the standalone agent-sh/computer-use-linux project; do not resume finishing-fork maintenance or export ChatGPT-specific authority, identity, browser-host, or environment contracts.

Acceptance criteria

• Bind the exact current vendored and standalone revisions, then inspect both triggering ranges tracked by issues Sync Linux Computer Use → agent-sh/computer-use-linux #73 and Sync Linux Computer Use → agent-sh/computer-use-linux #132.
• Classify every meaningful delta as already present, superseded, generic and still useful, or ChatGPT-specific; record enough evidence for another maintainer to reproduce the conclusion.
• If a unique generic improvement remains, adapt it to the standalone project's vocabulary and public contract, run its required security and correctness review loop, and land or obtain an explicit upstream disposition for it.
• Do not synchronize versions, names, or code merely to make the two repositories appear numerically aligned.
• Close issues Sync Linux Computer Use → agent-sh/computer-use-linux #73 and Sync Linux Computer Use → agent-sh/computer-use-linux #132 with the audit result and any standalone pull-request or commit links.
• Make no unrelated finishing-fork, official-DMG, package, integration, or feature change.

Blocked by

• None — can start immediately.

[coderabbitai]

🔗 Related PRs

#139 - docs(domain): define official-app transition contract [merged]
#140 - fix(nix): refresh official DMG pins for 26.810.52044 [merged]
#141 - docs(readme): announce official Linux app evaluation [merged]
#142 - chore(repo/rename): finalize chatgpt-linux identity [merged]
#156 - fix(app/security): restore trusted automation contracts [open]

---

📝 Issue Planner

_{Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.}

• Create Plan

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!

[nisavid]

Bounded audit checkpoint:

• I bound the finishing-fork snapshot at 9364b98a1fc4308d9792c2d77c59cb11d49490d0 and standalone agent-sh/computer-use-linux at 663930fef8d1bee64b7b8f833b68cff3196059e3.
• The negative-origin screenshot crop correction is already present with standalone coverage, so there is nothing to export there.
• Two unique, generic correctness improvements remained: absolute uinput axis geometry and AT-SPI registry reachability diagnostics. I adapted both to the standalone vocabulary and published agent-sh/computer-use-linux#96.
• The exported diff passed the full local correctness/security review loop and all hosted checks. The PR is ready for upstream review.
• I am not exporting ChatGPT-specific authorization, plugin identity, browser-host, environment, or packaging contracts.

I will close #73, #132, and this issue after the standalone maintainers give the exported work an explicit disposition. Security-sensitive material follows the standalone project's private disclosure process and is intentionally not duplicated in this public tracker.

[nisavid]

This supersedes my earlier checkpoint, which described the first draft rather than the completed exhaustive audit.

I bound current protected main at 4762b207ce71c5fda2ba547ab6987b601990d8b7; its vendored Computer Use tree is 0f122b7cdeb26bf13faacaf8b16445ee18a7d62e. The earlier 9364b98a1fc4308d9792c2d77c59cb11d49490d0 analysis snapshot has the same tree but is not the current-revision binding. I compared both exact issue ranges against standalone base 663930fef8d1bee64b7b8f833b68cff3196059e3.

The completed export in agent-sh/computer-use-linux#96 is exact head bebbdcc41df3dac1cacb79ad3bca3c48072c06eb and contains six generic source deltas:

1. correct last-pixel absolute-axis bounds and clamp;
2. extended-button fallthrough before absolute-device setup;
3. capability advertisement gated on complete AT-SPI readiness;
4. bounded, defensive Niri listing/focus/activation with decorated visual tile bounds;
5. GNOME installed-asset change detection with conservative reload reporting; and
6. authenticated, one-shot KWin callbacks with an end-to-end timeout and ownership-safe cleanup.

The final diff is 11 files, 1,162 additions, and 97 deletions. It contains no ChatGPT authorization, identity, browser-host, cursor side-channel, package, environment, or numeric-version synchronization. I removed the first draft's AT-SPI Registry probe entirely because it was outside both trigger ranges; only the in-range capability-map correction remains. I also excluded the out-of-range all-backend unknown-button semantics change and the non-material RAII environment-guard test hygiene.

Independent spec and security reviews returned DONE_CLEAN on that exact head. Local gates covered formatting, all-target checks, clippy with warnings denied, split full tests including four private-D-Bus adversarial KWin transactions, installer regressions, Rustdoc/package, MCP safety, Zod schema, npm wrapper/package, and agnix. Every current-head hosted check is terminal and successful or expectedly skipped, and Revuto approved the exact head with no evidence-backed concerns. The publication planner verifies the remote branch equals bebbdcc41df3dac1cacb79ad3bca3c48072c06eb.

The two earlier collaborator findings targeted the now-removed Registry experiment; I answered and resolved both threads. GitHub still carries that collaborator's historical CHANGES_REQUESTED review on the superseded first head. I cannot dismiss or re-request it with my read-only permission in the upstream repository, so I asked the reviewer for a refreshed explicit disposition on the current head: agent-sh/computer-use-linux#96 (comment).

I am leaving this issue open only for that upstream disposition. The bounded source audits themselves are complete, and #73 and #132 now carry their reproducible range-specific receipts.

[nisavid]

Final disposition:

• I audited both exact source ranges against current protected chatgpt-linux main 4762b207ce71c5fda2ba547ab6987b601990d8b7 (vendored tree 0f122b7cdeb26bf13faacaf8b16445ee18a7d62e) and standalone base 663930fef8d1bee64b7b8f833b68cff3196059e3.
• I published the six reusable deltas under standalone vocabulary in agent-sh/computer-use-linux#96, exact head bebbdcc41df3dac1cacb79ad3bca3c48072c06eb. The guarded publication planner verifies that exact remote ref.
• Independent spec and security reviews returned DONE_CLEAN on the immutable head. All current-head hosted checks pass or are expectedly skipped, Revuto approved that exact head, GitHub reports the PR cleanly mergeable, and no review thread remains unresolved.
• The explicit upstream disposition is recorded on the PR: a collaborator requested removal or repair of the first draft's two Registry-probe findings. I removed that out-of-range experiment entirely, answered and resolved both threads, and asked the reviewer to refresh the historical CHANGES_REQUESTED state on the final head. My upstream repository permission is read-only, so I cannot dismiss that review, re-request it through the API, or merge the PR. The PR remains open for the upstream maintainer; I do not claim that it merged.
• I closed Sync Linux Computer Use → agent-sh/computer-use-linux #73 and Sync Linux Computer Use → agent-sh/computer-use-linux #132 with their range-specific, reproducible classifications. No version alignment or ChatGPT-specific authority, identity, browser-host, cursor, environment, package, or feature work was exported.

This completes the bounded downstream disposition and leaves the accepted standalone candidate with its proper upstream owner.