Repository navigation
Releases
v0.1.3 — Targeted review progress and configurable timeouts
Review progress and configurable execution timeouts
- Review continuations now reuse validated findings and immutable evidence, track completed files, and target remaining work. Every group gets an initial pass before the global continuation allowance is used. Changed JS/TS dependencies are ordered together within bounded file, byte and changed-line groups.
- The default isolated execution timeout is 30 minutes. Configure
review.timeout_minuteson the protected default branch or use an authorized@CBReview review --timeout 60override. The single 100-provider-request budget remains unchanged. GitHub-hosted jobs impose a six-hour ceiling; setup/refresh/cleanup reserves leave at most 330 inference minutes. - Long reviews renew credentials only on the trusted host. Encrypted refresh-token rotation is persisted before updating the proxy-only access token. Read-only GitHub token renewal repeats live OIDC, installation, permission and commit checks. The reviewer remains read-only, network-isolated and credential-free.
- Request expiry and credential leases follow the selected timeout. Merge intent retains its separate one-hour deadline. Incomplete or unverifiable analysis still fails the review check, preserves validated findings as COMMENT, and cannot approve or merge.
review:
timeout_minutes: 30@CBReview review --effort xhigh --timeout 60
Verification
- 267 automated tests, typecheck, lint, build, clean-checkout install, real workerd/D1 and synthetic browser checks passed. Full GitHub CI includes Actionlint, Docker/native pinned-Codex isolation acceptance, dependency audit and runtime-image security scanning.
- Real isolated ChatGPT acceptance completed a 93-file/two-commit/four-distinct-PNG review: all six groups and final synthesis completed in 10m45s, and seeded tenant-isolation and stale-response defects were identified. A small clean review also completed with the authorized 60-minute override. Both tests published exact-head COMMENT reviews under advisory/no-merge policy.
- The live large fixture contains 1,668 changed lines. It does not establish completion time for every production PR or the larger reported production change. A separate deterministic >12,000-line case verifies lossless bounded grouping; synthetic tests do not measure model latency or semantic quality.
- Worker trust and both maintained callers use immutable source
23018028dc656100c1b313a01bf20e7109b8258a. Credential-only production OIDC checks passed for both callers, with inference skipped. Production health reports 0.1.3 and the merge deployment gate is restored. - Temporary test Worker, D1, profile, branches and image were removed, the test session was revoked, and the original test-repository vault was restored. No database migration was required.
This remains a CLI-onboarded release. Browser OAuth and hosted-inference flags remain disabled. Existing native verdicts, repository authorization, explicit merge policy and GitHub protections are unchanged. No production AI review was triggered during rollout.
Changes: review execution and timeout fix #59, 0.1.3 maintenance #60, immutable trust pin #61, merge-gate restoration #62.
v0.1.2 — Bounded context reuse and review efficiency
CodexBot 0.1.2 improves bounded context reuse and large-review efficiency without raising the 20-minute execution deadline or 100-provider-request limit.
- Reuse mounted, hash-verified source without consuming another distinct context-path slot. Keep 24 uncached snapshot/path lookups, a separate 100-request context-socket cap, and all existing GitHub API/byte/integrity limits.
- Return allowlisted retrieval reasons and preserve host completion/lookup diagnostics alongside the primary execution failure. No source, prompts, credentials or raw provider responses in diagnostics.
- Avoid retransmitting full modified old source alongside complete patches/current source. All base/head evidence remains independently verified and read-only.
- Stop further inference after an irrecoverable missing-evidence/output-loss report, retain validated findings and publish a non-approving COMMENT while failing the review job. Complete reviews still require every group and the final synthesis.
Validation: 251 tests plus typecheck/lint/build, clean-checkout, real workerd/D1, browser, Actionlint, catalog, dependency audit and full native Docker/Codex/security CI. Real isolated ChatGPT acceptance at gpt-6.1-sol/xhigh completed a 93-file/four-distinct-PNG review across 6 groups and synthesis, detecting both seeded authorization and stale-response defects at valid changed lines. The review step took 9m08s including preparation. A two-file clean review also completed. Both native reviews were bound to exact head commits; test policy kept approval/merging disabled.
No migration, credential rebind, dependency upgrade or repository-policy change. CLI-based ChatGPT onboarding remains supported; website OAuth and hosted website inference remain disabled. Incomplete or unverifiable evidence still prevents approval and merge. Controlled fixtures do not guarantee arbitrary PRs finish within the bounded deadline.
Implementation: #54
Release metadata/maintenance: #55
Immutable Worker pin: #56
Gate restoration: #57
Trusted reusable-workflow source: 4398779
CodexBot v0.1.1
CodexBot v0.1.1 fixes the review-execution failures uncovered by real isolated acceptance and ships bounded multi-group review orchestration with strict completion checks.
Fixed
- Project group, continuation and synthesis output schemas onto the documented Responses Structured Outputs subset. Uniqueness and string-length constraints remain mandatory in trusted-host validation.
- Stop requiring a redundant shell command in every later turn when immutable, hash-verified unit evidence is already supplied. An execution without successful inspection still fails.
- Restore the original verified image inputs and their labels after Codex context compaction removes image history. Partial loss, substitution, duplicate replacement and over-budget requests remain rejected.
- Preserve validated findings from incomplete analysis as non-approving COMMENT reviews while the analysis job remains failed. Missing essential context, unfinished groups or synthesis never authorize approval or merging.
- Include the preceding context-completeness and sanitized primary/cleanup diagnostic corrections recorded in CHANGELOG.md.
Verified acceptance
Real ChatGPT reviews completed against the exact final implementation in isolated TopkAPI acceptance:
| Fixture | Actual result |
|---|---|
| 2 files / 2 commits | Complete clean review; native review published |
| 33 files | All three groups and synthesis completed; expected cross-file authorization defect published inline |
| 93 files / 89 text patches / four distinct PNGs / 2 commits | All six groups and synthesis completed; complete clean review published |
All three used the same final implementation SHA. Native events were COMMENT under the unchanged advisory-only test repository policy. Test webhook inputs were operator-signed staging fixtures derived from fenced real comments; the downstream GitHub App dispatch, Actions OIDC, protected Environment credentials, isolated Codex, official Responses inference and native publishing were real. This does not claim direct GitHub webhook delivery to the separate test backend.
Validation passed: 246 automated tests, strict typing, lint, build, clean checkout, real workerd/D1, synthetic-provider browser tests, Docker isolation, 19 native pinned-Codex inspection scenarios including actual compaction with four distinct images, Actionlint, npm audit and full GitHub CI including the image scan. A fresh production credential-only OIDC diagnostic passed without inference or publishing.
Deployment and upgrading
The reviewed versioned reusable-workflow source is bf0510ed98ad2bb7a3798ad9fd3d1ae1286954a2. Production Worker trust and the KMU/TopkAPI callers are coordinated to this exact immutable SHA. This release tag targets the reviewed final Worker/configuration commit; the later maintenance commits change only trust configuration and its test, then restore the merge gate.
Self-hosted operators must keep their own Worker trust and caller pins aligned and follow the documented maintenance procedure. No database migration, credential reconnection or repository-policy change is required by this patch. Quotas, isolation, authorization, exact-head checks and GitHub protections are preserved.
CLI-based ChatGPT onboarding remains the supported release path. Browser ChatGPT onboarding and hosted website inference remain disabled. Deployment-level merge authorization is restored after verified cutover; repository-specific opt-in, explicit-request policy where configured, required checks and GitHub protections still apply.
Limitations and cleanup
AI review quality is not guaranteed: an earlier small-fixture Unicode-escape false positive remains visible in the acceptance history. These tests establish successful bounded execution for the tested sizes and image transport/provenance, not universal PR completeness or semantic accuracy on arbitrary screenshots. Existing request, byte and time limits remain enforced. The original production HTTP400 did not retain a usable provider rejection envelope, so its sole rejected keyword is not retrospectively proven.
The temporary ChatGPT session was revoked, its Worker and D1 were deleted, and TopkAPI was restored before the reviewed production pin-only upgrade. Fixture PRs were closed without merging their code. No KMU business code, policy, feature PR, secrets or branch protections were changed; no additional KMU AI review was triggered during rollout.
CodexBot v0.1.0
CodexBot 0.1.0 is the regular CLI-onboarded release.
- Native GitHub COMMENT, APPROVE and REQUEST_CHANGES reviews, including validated inline text findings.
- Automatic reviews with explicit owner-consented unattended credentials, and manual commands with configurable handles, model and reasoning effort.
- Opt-in Fully Autonomous merging or Human-in-the-Loop, subject to GitHub protections, required checks, live authorization and exact reviewed commits.
- Verified PNG inputs through the official pinned Codex App Server; immutable base/head context, bounded large-PR collection, safe image-summary findings and fail-closed incomplete analysis.
- Configurable unlimited rolling-day quotas while preserving authorization, concurrency, cooldown and duplicate protection.
Reviews execute in isolated repository-owned GitHub Actions runners. The Worker retains App credentials; the reviewer receives no GitHub write tokens or provider credentials. No paid API fallback is enabled.
Setup uses the existing CLI ChatGPT connection or an explicit user API key. Browser ChatGPT onboarding and hosted browser inference remain disabled pending their separate client provisioning. PNG is the implemented image format; audio, video, arbitrary files, executables and archives are not silently treated as reviewed. The real pinned Codex/Docker image transport was tested against scripted Responses; this release does not claim a new live visual-model acceptance.
Validation: 208 tests, clean-checkout checks, workerd/D1, browser checks, actual pinned Codex sandbox/image transport, Actionlint, model catalog, dependency audit and Cloudflare deployment dry-run. No additional provider inference was launched for this hotfix. Downstream unrelated business CI was not awaited, as explicitly directed by the operator; this is not represented as a passing test.
CodexBot v0.1.0-beta.2 — GitHub-qualified App approvals
Fix GitHub App review qualification without weakening repository protections.
- GitHub Required approvals stays at 1 or your stronger configured count. CodexBot no longer turns that count, CODEOWNER or last-push rules into an extra human minimum. It requires GitHub's actual APPROVED decision when approvals are required and still uses the native exact-head merge with all required CI/protection checks.
- minimum_human_approvals: 0 means no additional CodexBot-required humans. Explicit Human-in-the-Loop minimums still require distinct, current-head, write-capable non-author humans.
- Manual-only reviews and explicit-only --merge behavior remain unchanged. Unknown/unmet approval evidence, incomplete AI analysis, findings, stale heads and failed checks never authorize a merge.
- Add “Copyright © 2026 Kaan Gönüldinc” to the shared website footer.
Implementation PR #36 merged as 17608a5. Pin PR #37 merged as c56e94a. All active Worker/backend/caller pins use the actual source revision 17608a5. Disabled callers were not reactivated.
Validation: 199 tests, typecheck, lint and build passed. Full source and pin CI passed, including clean checkout, Workers/D1, synthetic browser, pinned Docker/Codex sandbox, Actionlint and security checks. The production Worker version, deployment, canonical health endpoint and rendered footer were verified. No dependency, credential, repository-protection or D1 schema changes.
Production version: 27391a59-c875-4306-8828-997847169c5d. Deployment: 4f350016-e70a-46c8-832b-71d83362acd6. CLI ChatGPT and merge deployment authorization remain enabled; browser OAuth and hosted website inference remain disabled.
Acceptance limitation: the operator explicitly waived new disposable live GitHub/inference tests for this patch; none were performed. App qualification under a particular review/last-push configuration is not claimed as newly live-proven. GitHub remains authoritative and may require another qualifying reviewer, especially when the App itself made the latest reviewable push. No fallback or protection bypass was introduced.
CodexBot v0.1.0-beta.1 — CLI onboarding
CodexBot v0.1.0-beta.1 uses CLI-based initial ChatGPT provisioning. Reviews and merges then run unattended on repository-owned GitHub Actions; a permanently running CLI or VM is unnecessary.
Supported modes:
- Native COMMENT, APPROVE and REQUEST_CHANGES with exact-commit inline findings.
- Automatic reviews with explicit administrator consent for a repository-owned credential assignment; contributors never implicitly share accounts.
- Fully Autonomous merging with zero CodexBot-required human approvals, or Human-in-the-Loop with 1–6 independent human approvals.
- Repository opt-in, separate merge authorization, required CI, App-bound guards, current-head verification and all applicable GitHub protections. No protection bypass.
Live evidence:
- Complete APPROVE and protected autonomous merge without a human: disposable acceptance PR #3, workflow 43669491863.
- A substantial authorization regression: REQUEST_CHANGES on the changed security line, PR #4, workflow 43669871592; never merged.
- Pending required CI: native auto-merge queued by CodexBot while the App guard and required CI remained pending on PR #9.
- Complete advisory COMMENT and inline findings verified separately.
- Final immutable runner pin and environment-only/OIDC credential diagnostic: f008673, workflow 43679274620.
Browser ChatGPT OAuth is deferred. WEB_ENABLED=false and OPENAI_HOSTED_INFERENCE_AUTHORIZED=false. The approved CLI ChatGPT deployment remains enabled. No API-key fallback is used.
Beta limitations: AI review can miss defects; incompatible merge queues or unverifiable protection/check evidence fail closed. Trusted repository administrators/workflow writers remain inside the credential trust boundary. Credentials, local state and generated build directories are excluded from release archives.
Validation: 198 tests, typecheck, lint, build, clean-checkout, real workerd/D1 tests, synthetic browser onboarding, pinned Codex Docker isolation, Actionlint, dependency audit, runtime image security scan and Wrangler dry-run passed in CI. Synthetic provider tests are not represented as real provider acceptance. Migration 0006 is additive; existing production migrations are current.
Production verification:
- Release implementation: #35, merged commit 067be58; CI run 43679312787 passed.
- Final immutable runner source: f008673. Worker, backend expectation, TopkAPI caller and disposable caller agree.
- Worker version bd6e416f-4596-4db2-8ca8-793bc700dbb8; deployment 5248884b-5b9a-4d97-955f-e92a13743ee3 at 100%.
- Production https://codexbot.appsolves.workers.dev/health reports 0.1.0-beta.1. CLI installation landing, canonical redirects, disabled browser authentication and cross-origin rejection were verified.
- CHATGPT_DEPLOYMENT_AUTHORIZED=true; MERGE_DEPLOYMENT_AUTHORIZED=true; WEB_ENABLED=false; OPENAI_HOSTED_INFERENCE_AUTHORIZED=false. Temporary acceptance scope and operator quota overrides are cleared.
- Owner-consented automatic review for an unconnected github-actions[bot] contributor: complete native APPROVE, https://github.com/AppSolves/codexbot-acceptance-20261115/pull/10, workflow 43679528359. No provider session belongs to the contributor.
- Human-in-the-Loop: PR #10 was held while only the Bot approved. After a genuine independent human approval, fresh complete inference (workflow 43712378653), native APPROVE and both required checks, CodexBot merged it natively at 2026-11-15T12:13:07Z as 895dd8e9f52b8add5855013fa9b2aa53f3ef383c. GitHub protection required one approval, the human had write access, and the bypass actor list was empty. The merged tree exactly matches the reviewed head.
Merged implementation and source-pin PRs:
| PR | Actual merge SHA |
|---|---|
| 23 | 929ff2a |
| 24 | 86f54dd |
| 25 | b15163e |
| 26 | 29c6254 |
| 27 | 01aba17 |
| 28 | 4cc32b8 |
| 29 | 757c28e |
| 30 | 59b00dd |
| 32 | f008673 |
| 33 | e20f3bd |
| 34 | 8dd88f6 |
| 35 | 067be58 |
Caller coordination: Worker trust, backend expectations and both verified caller repositories pin f008673. Existing Environment guards and named secret references are preserved.
Acceptance failures were not hidden: the original waiting merge intent expired and was cancelled. Two fresh attempts returned incomplete analysis, failed their review jobs and produced a non-approving COMMENT. A corrected snapshot-path context then produced the fresh complete review above. No incomplete result or stale request authorized a merge. Merge intents expire after one hour; expiry requires a fresh authorized review. Browser OAuth and a real paid API-key acceptance run remain outside this CLI/ChatGPT beta acceptance.