Skip to content

Release hardened Co-Engineer and Cursor control planes - #4

Closed
ajhcs wants to merge 2 commits into
mainfrom
codex/control-plane-e2e-fixes
Closed

Release hardened Co-Engineer and Cursor control planes#4
ajhcs wants to merge 2 commits into
mainfrom
codex/control-plane-e2e-fixes

Conversation

@ajhcs

@ajhcs ajhcs commented Aug 19, 2026

Copy link
Copy Markdown
Owner

Summary\n\n- release Co-Engineer 2.2.0 and Cursor Cloud Control 0.4.0\n- fix durable uncertain-submission reconciliation for Cursor create, follow-up, cancel, and lifecycle operations\n- add opt-in host-trusted local Cursor CLI control with durable receipts and PID identity fencing\n- add safe local/GitHub target staging, exact ref binding, cancellation barriers, and bounded final-response projection for Grok and DSH\n- add public marketplace metadata, licenses, onboarding, package inventories, and release validation\n- keep ACP systemd/cgroup and Grok outer-sandbox foundations packaged but unwired pending host acceptance\n\n## Exact-tree verification\n\nAuthoritative release-gate run: 20260819T015719Z-8c253e10c3e8\n\n- Co-Engineer: 247/247 tests\n- Cursor Cloud/local: 174/174 tests\n- MCP Inspector preflights\n- real Bubblewrap boundary test\n- ACPX provenance and reproducible build\n- activation/rollback fixture\n- package inventories and release validation\n- independent Luna Max P0/P1 re-reviews cleared the final fixes\n\n## Notable regression coverage\n\n- provider 404 after an uncertain Cursor create no longer poisons or duplicates the request\n- follow-up reconciliation requires repeated exact 404 confirmation before release\n- provider agent/run identity mismatches never finalize or release reservations\n- local Cursor request tombstones survive 201+ runs and fail closed at capacity\n- TERM/KILL paths revalidate durable PID start tokens immediately before signaling\n

@ajhcs

ajhcs commented Aug 19, 2026

Copy link
Copy Markdown
Owner Author

Exact-head update: 78d8f54\n\nThe first PR-triggered run exposed a real ACP event-ledger lock handoff race (lstat succeeded, lock vanished before descriptor open). The duplicate push run passed on the same old SHA, confirming timing dependence. The new head retries only that exact ENOENT observation; all other lock/identity failures remain fail-closed. A deterministic regression removes the listed fixture lock immediately before open.\n\nAuthoritative exact-tree release gate: 20260819T021738Z-54837c988763\n- Co-Engineer 248/248\n- Cursor Cloud/local 174/174\n- Inspector, real Bubblewrap, ACPX provenance/reproducibility, activation fixture, package inventories all passed

@ajhcs

ajhcs commented Aug 19, 2026

Copy link
Copy Markdown
Owner Author

Superseded by the simplified and fully gated Co-Engineer 3.0 implementation merged in PR #5.

@ajhcs ajhcs closed this Aug 19, 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.

1 participant