Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
76 commits
Select commit Hold shift + click to select a range
08c68d5
fix(ux): close 3.4.0 final-art with approved identity
ajhcs Aug 28, 2026
bfe97f6
feat: prepare codex-co-engineer 3.4.1 candidate
ajhcs Sep 1, 2026
3f62984
Simplify Co-Engineer launches and prepare Astra routing guidance
ajhcs Sep 4, 2026
6412a73
Prepare Codex-Co-Engineer 3.4.2 release
ajhcs Sep 4, 2026
94d1b82
Align release-note assertion with version 3.4.2
ajhcs Sep 4, 2026
fe9dbed
Keep model selection evidence-based and coordination lightweight
ajhcs Sep 4, 2026
3b15aee
Document native consent and stable run continuation
ajhcs Sep 4, 2026
19cb8f9
test: exercise native consent through stdio
ajhcs Sep 4, 2026
67c836e
fix: keep experience presentation truthful
ajhcs Sep 4, 2026
45513e2
fix: exclude undispatched outcomes and simplify coordination projection
ajhcs Sep 4, 2026
8ac7bd2
docs: explain native consent in advanced run controls
ajhcs Sep 4, 2026
1ee2245
test: cover explicit native consent retry on the same run
ajhcs Sep 4, 2026
eabd124
test: use protocol-shaped dismissal responses
ajhcs Sep 4, 2026
e7ba8e3
fix: connect native consent and preserve durable run receipts
ajhcs Sep 4, 2026
580436c
test: define native lifecycle ownership and reproduce bridge failures
ajhcs Sep 5, 2026
f6815da
docs: qualify the complete native run workflow
ajhcs Sep 5, 2026
7e63442
test: prevent unsuccessful tasks from becoming successful runs
ajhcs Sep 5, 2026
4bd07a4
fix: follow native provider tasks through durable completion
ajhcs Sep 5, 2026
b37b739
fix: wait for a populated descendant PID in release preflight
ajhcs Sep 5, 2026
ca576c3
Document provider boundary acceptance and honor configured cache storage
ajhcs Sep 5, 2026
c96d5f2
Synchronize packaged provider-fix release notes
ajhcs Sep 5, 2026
669d909
Preserve safe provider failure categories in native run receipts
ajhcs Sep 5, 2026
41efd2a
fix: adapt Cursor Cloud SDK terminal results
ajhcs Sep 5, 2026
de41f23
Route Muse 1.3 through OpenRouter
ajhcs Sep 5, 2026
e057e60
Describe one-shot ACPX transport qualification
ajhcs Sep 5, 2026
6b5177a
fix: run Muse through bounded ACPX exec
ajhcs Sep 5, 2026
b5bf1b7
Synchronize cursorless wait regression with actual wait state
ajhcs Sep 6, 2026
a7f5dbf
Simplify routine Co-Engineer launch instructions
ajhcs Sep 6, 2026
39a8a66
fix: preserve pending dispatch lifecycle evidence
ajhcs Sep 6, 2026
5b75448
fix: stabilize Cursor SDK discovery
ajhcs Sep 6, 2026
f535420
feat: compact semantic run receipts
ajhcs Sep 6, 2026
ffc7067
Document compact runs and native efficiency acceptance
ajhcs Sep 6, 2026
301fcea
fix: align aggregate root contention bound
ajhcs Sep 6, 2026
c69379a
Reduce skill discovery overhead and clarify response encoding
ajhcs Sep 6, 2026
7b04df0
test: align native consumers with compact receipts
ajhcs Sep 6, 2026
1f0e5eb
fix: preserve compact run UI metadata
ajhcs Sep 6, 2026
b86fb1c
fix: preserve compact semantic UX
ajhcs Sep 6, 2026
c056904
fix: bound semantic overflow receipts
ajhcs Sep 6, 2026
3df7b1a
fix: bound overflow lifecycle state
ajhcs Sep 6, 2026
127a3ea
fix: validate semantic response modes
ajhcs Sep 6, 2026
bcc89a5
fix: preserve Cloud answers that overlap prompts
ajhcs Sep 7, 2026
5d1816d
docs: clarify Desktop cache and Cloud outcomes
ajhcs Sep 7, 2026
061640e
docs: keep provider launch prompts task-focused
ajhcs Sep 7, 2026
2437c77
fix: distinguish Grok auth from readiness noise
ajhcs Sep 7, 2026
6a01986
Fix structured ACP attention handling
ajhcs Sep 7, 2026
9ee1515
test: isolate consent SDK discovery
ajhcs Sep 7, 2026
f08be4a
docs: clarify complete attention continuation behavior
ajhcs Sep 7, 2026
563c4a8
Preserve explicit ACP question prompts
ajhcs Sep 7, 2026
c7c6490
fix: persist structured attention options
ajhcs Sep 7, 2026
0dcc3c3
docs: describe Linux descendant cleanup repair
ajhcs Sep 7, 2026
4da40a5
fix: retain ACP descendants without ps
ajhcs Sep 7, 2026
c5eab02
docs: qualify local release installs against project marketplace refresh
ajhcs Sep 8, 2026
8dc565d
fix provider execution guidance
ajhcs Sep 8, 2026
16ca286
Fail early when worker runtime files are missing
ajhcs Sep 8, 2026
58360f3
Surface runtime preflight through run admission
ajhcs Sep 8, 2026
00a898c
Document reusable consent and launch reliability fixes
ajhcs Sep 8, 2026
2552ebb
Clarify consent command invocation for plugin installs
ajhcs Sep 8, 2026
67549ba
Add reusable native consent grants
ajhcs Sep 8, 2026
3332d9a
Sync installed documentation with consent and runtime changes
ajhcs Sep 8, 2026
d39ce8a
Tolerate vanished proc entries during verification cleanup
ajhcs Sep 8, 2026
1070ff7
Record verification cleanup scan correction
ajhcs Sep 8, 2026
239383d
Document concise provider response contract
ajhcs Sep 8, 2026
6ceecc9
fix provider response guidance precedence
ajhcs Sep 8, 2026
59f9003
Document Grok session response rules
ajhcs Sep 8, 2026
90fb31b
Clarify bounded provider output handling
ajhcs Sep 8, 2026
75787cd
fix Grok response rules at session creation
ajhcs Sep 8, 2026
0e2335e
clarify Grok tool-call response rules
ajhcs Sep 8, 2026
c69d70a
Document Grok final response framing and conservative fallback
ajhcs Sep 8, 2026
af4fb25
select Grok final response by tool framing
ajhcs Sep 8, 2026
f8b5b89
Shorten upstream citation to avoid credential scanner false positive
ajhcs Sep 8, 2026
9e0b847
Prepare 3.4.2 release on the complete published baseline
ajhcs Sep 8, 2026
7e03eb5
Document self-contained setup and optional model coordination
ajhcs Sep 8, 2026
3bd2ea6
Record bundled setup and public release date
ajhcs Sep 8, 2026
374af37
Bundle worktree bootstrap runtime
ajhcs Sep 8, 2026
60e9064
Clarify shared setup Python prerequisite
ajhcs Sep 8, 2026
f57ae01
docs: keep legacy host relay optional
ajhcs Sep 8, 2026
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion .agents/plugins/marketplace.json
Original file line number Diff line number Diff line change
Expand Up @@ -11,7 +11,7 @@
"plugins": [
{
"name": "codex-co-engineer",
"version": "3.4.0",
"version": "3.4.2",
"description": "Give Codex a team of external co-engineers without giving up control. Delegating to Co-Engineer starts one bounded run. The honest shape is up to eight isolated external co-engineers, one bounded run, one coordinated wait, one verified decision.",
"keywords": [
"codex",
Expand Down
3 changes: 0 additions & 3 deletions .codex/release-gate.toml
Original file line number Diff line number Diff line change
Expand Up @@ -21,7 +21,6 @@ kind = "bootstrap"
command = ["npm", "--prefix", "tools/acpx-vendor", "ci", "--ignore-scripts", "--no-audit", "--no-fund"]
failure_class = "environment_blocked"
timeout_seconds = 180
env = { ACPX_NPM_CACHE = "/tmp/codex-acpx-release-npm-cache", NPM_CONFIG_CACHE = "/tmp/codex-acpx-release-npm-cache" }

[[stages]]
name = "co-engineer-unit"
Expand Down Expand Up @@ -74,15 +73,13 @@ kind = "build"
command = ["npm", "--prefix", "tools/acpx-vendor", "run", "test:reproducible"]
failure_class = "product_test_failed"
timeout_seconds = 180
env = { ACPX_NPM_CACHE = "/tmp/codex-acpx-release-npm-cache", NPM_CONFIG_CACHE = "/tmp/codex-acpx-release-npm-cache" }

[[stages]]
name = "acpx-publish-provenance"
kind = "security_checks"
command = ["npm", "--prefix", "tools/acpx-vendor", "run", "verify:publish-provenance"]
failure_class = "product_test_failed"
timeout_seconds = 180
env = { ACPX_NPM_CACHE = "/tmp/codex-acpx-release-npm-cache", NPM_CONFIG_CACHE = "/tmp/codex-acpx-release-npm-cache" }

[[stages]]
name = "co-engineer-package-inventory"
Expand Down
94 changes: 94 additions & 0 deletions CHANGELOG.md
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,100 @@

## [Unreleased]

## [3.4.2] - 2026-09-08

### Fixed

- Instruct providers to honor the requested answer format and omit routine
narration and unrequested evidence headings. Controller evidence requirements
remain intact and do not act as an answer template.
- Return Grok's final response after a settled tool round on successful tasks.
Preserve progress in task events and retain aggregate output when framing is
ambiguous, incomplete, over limit, or includes inline web search.
- Handle processes disappearing during verification cleanup scans and parse
parenthesized process names correctly, while preserving fail-closed errors.
- Remember explicit native approval for the same repository and selected
providers, with a one-run option and local revocation. Remove the redundant
approval checkbox and allow more time to answer the native form.
- Detect missing installed worker entrypoints before preparing workspaces or
dispatching providers, with actionable reinstall and restart guidance.
- Direct provider workers to their assigned working directory and clarify
controller-owned receipts, reducing unnecessary setup and evidence searches.

- Adapt Cursor Cloud SDK results before internal validation so documented
timing, model, and usage metadata does not prevent terminal result delivery.
- Use ACPX's one-shot command for DSH, preserving bounded provider errors and
results without a nested flow or replaying a submitted prompt.

- Deliver the compiled assignment instructions and pin local workspaces to
the recorded commit; reject unsupported provider model overrides instead
of silently launching another model.

- Align semantic runs with the authoritative provider task lifecycle, including
result delivery, recoverable observation failures, cancellation, and restart.
Reuse event-driven task waits and keep unchanged run receipts stable.

- Wire semantic launch consent to native MCP forms, with explicit unsupported
host handling and same-run consent retry. Preserve pending/terminal receipts
and cursors; distinguish planned review work from completed evidence.

- Reject future-dated repository-exposure approvals and make consent-window
fixtures independent of today's date. Remove ambient Git/network and
fixed-sleep dependencies from affected regression tests.

- Semantic `run_request` launches now satisfy the advertised MCP JSON Schema
without legacy single-task fields; ambiguous dual envelopes are rejected.
- Valid disjoint writer globs use the authoritative overlap validator instead
of a duplicate check that rejected separate directories.
- Repeated status checks no longer repeat terminal cleanup grace periods for
previously reconciled receipts; ownership inspection remains active.
- Preserve structured Cursor permission options through grouped questions and
stop treating shell command punctuation as user input.
- Distinguish explicit Grok login status from ancillary command errors during
readiness checks.
- All-Cursor Cloud runs skip the irrelevant local systemd/cgroup prerequisite;
mixed and local runs still enforce it.
- Assignment access can be derived from its explicit role, eliminating a
redundant launch field while rejecting explicit role/access conflicts.

### Changed

- Bundle MIT-licensed Worktree Bootstrap 1.1.0 with recorded source provenance.
Local setup no longer depends on a separately installed private tool; Python
3.11+ is required. Validate runtime prerequisites before installation.
- Reorganize installation and first-use documentation, include detailed upgrade
notes, and keep Luna/Sol coordination explicitly optional.

- Default Muse to `meta/muse-spark-1.3-contributor` through OpenRouter with
XHigh reasoning, using only the OpenRouter credential for that route.
- Release checks honor the operator's configured npm cache and temporary
storage instead of forcing a machine-specific cache path.

- Added evidence-qualified Luna / Terra / Sol / Astra task-selection guidance.
Prefer one owner with bounded workers; an extra coordinator and reasoning
effort changes require task-specific justification. Model choice stays host-owned.
- Made the installed skill path use existing provider choices and semantic
admission directly, with setup and manager tasks outside routine launch.
Clarified continuation, typed approvals, dependent reviews, and verification.
Provider prompts stay focused on requested output, tests and brief evidence;
the controller creates managed worktrees while the worker wrapper owns
verification, hidden writer tokens, lifecycle and machine receipts.

## [3.4.1] - 2026-09-01

Reliability candidate for bounded Co-Engineer implementation lanes. Preserves
the five public tools and 3.4.0 compatibility while adding a server-compiled
`run_request`, atomic pre-prompt admission, truthful dispatch evidence,
restart-safe session recovery, typed repository-exposure consent,
capability-aware attention batching, idempotent cancellation, silence-aware
handoffs, readiness caching, and bounded simple-run responses. See
[`docs/releases/v3.4.1.md`](docs/releases/v3.4.1.md) and the exact baseline
ledger in [`docs/releases/v3.4.1-baseline.md`](docs/releases/v3.4.1-baseline.md).

The candidate still requires host consent integration, an official exact-SHA
local-only `worktree-bootstrap` capability, the Codex Desktop history-bound
regression, and live provider acceptance before release publication.

## [3.4.0] - 2026-08-28

3.4.0 is the Co-Engineer experience and efficiency release. It keeps the
Expand Down
Loading
Loading