Skip to content

Latest commit

 

History

History
49 lines (34 loc) · 6.62 KB

release-readiness.md

File metadata and controls

49 lines (34 loc) · 6.62 KB

Release review

This document records the historical 2026-10-16 review. See the verification record and acceptance matrix for the original scope, corrected failures and untested external-service boundaries. Historical counts and results below do not certify later commits; check CI on the exact public revision before merging or publishing.

Conclusion

Suitable for submission as an initial public, self-hosted source release for trusted teams. The repository owner requested the final review and authorized opening a public PR on 2026-10-16. This is a source-release readiness conclusion, not certification of every provider, operating system or production topology; merging and publishing release artifacts remain separate steps.

The supported access model is explicit: signed-in users can read project data across the instance. Project writes require an administrator, owner or permitted member; viewer membership does not grant agent execution rights. The default installer binds to localhost. Disable self-registration (AGENTFLOW_ALLOW_SIGNUP=false) before allowing access from outside the trusted team. The default container setup controls the host Docker daemon and should only be operated by trusted administrators. See deployment boundaries.

Final source and deployment review

  • Independently checked all 1,718 recorded source blob hashes against the recorded master. After the hosted-CI compatibility fixes, the candidate has 1,464 unchanged upstream files, 180 reviewed adaptations and 74 recorded exclusions, plus 60 audited public additions in the platform export scope. Private deployment files outside the export scope remain excluded. No private Git history is imported.
  • Latest upstream features include durable group-to-SOP handoff and isolated script views. Their source and tests are included. MySQL concurrent claims/lease recovery and Docker script views (including non-root identity) were explicitly exercised.
  • The preview had still been running an older executable after the source sync. Rebuilt the service and runner/gate helpers, restarted the preview, and compared its key JavaScript assets byte-for-byte with the current source. The handoff acceptance case now exercises the current endpoint behavior.
  • Fixed a remaining viewer permission gap in group creation, role dispatch, SOP control and handoff retry, and blocked run-bound assistant execution for read-only members. Tests cover denial for viewers/non-members and continued access for owners; viewer reads remain available. This generic permission fix should be ported back to internal upstream.
  • Updated the source audit record and the checker: --source now verifies the actual source blobs as well as the remote commit label. No unrelated source hashes were reset.

Publication checks

Area Evidence
Source scope Source manifest/override checks and public-content checks
Credentials Gitleaks and exact checks for the supplied credentials in indexed blobs and inspected reports; ignored local secrets are not exported
License packaging Root MIT license, editor notices, complete Go dependency notice bundle and generated Local distribution notices
Application Full Go/vet checks, real HTTP/API suite, browser regressions and clean container acceptance; detailed counts in the verification record
New user Browser administrator creation, teammate registration, onboarding and missing-key guidance; no-key script/approval demo and repository connection
GitHub Real issue import, preservation of local clarification, PR discovery/merge between disposable branches and closed-issue synchronization in the authorized test repository
Documentation Bilingual README, deployment/access boundaries, Local positioning, contribution/security policies and upstream sync procedure
Repository presentation Description, topics and issue labels prepared in .github/repository-metadata.json; apply only after publication approval

Limits of the evidence

  • External knowledge retrieval/write-back, enterprise notifications and paid Codex/Claude services lack live credentials. Their protocol/adapter tests do not count as live service verification.
  • Real model results use the supplied CodeBuddy environment. The four-role Python delivery case checks actual tests, artifacts, approval and code hashes; it does not guarantee success for arbitrary tasks.
  • The earlier 47-minute run used the frontend SOP template on a Python test repository and required explicit human clarification/acceptance. It demonstrates orchestration and recovery, not verification of every frontend framework or fully autonomous delivery. No PR was created by that run.
  • Linux container/native acceptance is the primary deployed environment checked here. Cross-compilation of macOS/arm64 helpers is not runtime testing on every supported host.
  • GitHub-hosted CI must validate the submitted PR. A package, image, tag or GitHub Release is not published by this review.

Final submission review — 2026-10-16

The final review fixed the runner cancellation issue also identified in the review of PR #46: cancellation now disables the completed context channel, a single goroutine drains the output streams and owns cmd.Wait, and early control failures reap the command before returning. Focused process/managed-CLI regressions passed five repetitions; cancellation and early-failure cases also passed three repetitions with the race detector.

An additional full-suite run exposed a preview-test race: the preview becomes ready before its asynchronous workspace cleanup finishes. The test now waits for both the persisted cleanup marker and directory removal before closing its store; it passed 40 repetitions. The first full suite and the corrected final suite are recorded separately in validation.md.

Existing pull requests

  • PR #46 was closed as superseded by PR #48. The replacement uses the audited master export, public adapters, MIT packaging and the existing workflow under local/; closure does not imply its commits were merged.
  • PR #47 was closed after the owner explicitly retired its separate workspace redesign and room-crew proposal. Only its language/theme mechanism is reused. Closure records that product decision; it does not claim parity with its room mentions, quoted replies, crew planning or crew-to-draft-SOP features, and those features need no follow-up port.