Finish authority and documentation reconciliation

[Chris0Jeky]

Summary - route Taskdeck's active Claude, Codex, Fable, contributor, worktree, QA, and batch instructions to the canonical global review-and-ship pipeline instead of restating reviewer counts, fix loops, bot triggers, aging, or merge ownership - make the declared Taskdeck authority explicit: an authorized agent may push and merge within task scope after exact-head Required CI and the canonical review gate; workflow and dependency PRs retain designated human review without reserving the final merge click - reconcile STATUS, MASTERPLAN, REVIVAL, testing, interface-map, user-manual, threat-model, data-model, and human-action truth after merged PRs #1556, #1558, and #1559 - correct shipped Agents/Runs and Integrations routes, the Transcript/artefact persistence map, two EF delete-behavior claims, current beta/release direction, and the broken testing-guide anchor - preserve #1470 as the owner of full data-model recertification and #1305 as the remaining Transcript-product slice; no human-owned OUTSTANDING checkbox was checked ## Policy disposition The owner explicitly retired Taskdeck's blanket owner-click merge reservation. Live issue/PR instructions #1500/#1503 and #1510/#1511 now retain their T4 designated maintainer/CODEOWNER review requirements while allowing merge execution to follow declared authority after review and current exact-head gates. PR #1509 still requires recorded ADR ratification and #1514 still requires the stacked-workflow disposition/authorization; after those human decisions, an authorized agent may execute the merge. Dependency policy makes the same human-review versus unattended-auto-merge distinction. ## Verification - node scripts/check-docs-governance.mjs — passed - node scripts/check-golden-principles.mjs — passed - node scripts/check-github-ops-governance.mjs — passed - py -3 -B -m unittest discover -s scripts/agent_hooks -p test_render_failure_ledger.py — 11 passed - powershell -NoProfile -File scripts/git/Test-New-CodexIssueWorktree.ps1 — 28/28 passed on the second attempt with a 300-second tool ceiling; the first attempt hit the 124-second caller timeout without an assertion failure - one-time after-rg inventory across all active routes — no obsolete policy copies - one-time changed-Markdown target/case/fragment sweep — 51 links across 26 files, 0 failures - .agent-harness/tier.json parse, git diff --check origin/main..HEAD, and DCO — passed; 4/4 commits signed - landed-main focused regression evidence: Transcript 5, export/deletion 63, repository/migration 14, MCP tools 33, container result contracts 5, Dockerless integration 7 passed / 28 skipped - exact-main hosted Required CI run 30720236098 and CodeQL run 30720235986 — green at 6a92516c584a93ad4eea6c288b743a3151f642b5 ## Review disposition Fresh-context review #4836034180 found no CRITICAL issue and two HIGH classes; commit 7aa97160 fixed both. The automatic second and final review round at 7aa97160 found no CRITICAL issue, two HIGH documentation-safety defects, and three MEDIUM documentation mismatches. Commit edcb6257 corrects the nullable ERD cardinalities and restores ADR-0048 / #1429 as the hard PDF process-memory-containment gate. The MEDIUM findings are tracked without fix-cascade as #1563 (API-created agent runs) and #1564 (navigation catalogs). First-round MEDIUM findings remain tracked as #1561 and #1562. All eight review threads were classified, replied to, and resolved; the two-round ceiling is complete. ## Boundaries No permanent policy-parity or link meta-gate was added. No workflow, product code, repository setting, secret, protected checkout state, or pre-existing worktree was changed. Positive Docker/PostgreSQL proof is hosted evidence from #1559; it was not rerun locally. Closes #1544 Refs Chris0Jeky/agent-harness#101, #1291, and #1269.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: af9d90dd36

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[Chris0Jeky]

Fresh-context adversarial review of exact head af9d90dd36664255a1c8c86580b00721d6944d62 against base 6a92516c584a93ad4eea6c288b743a3151f642b5.

Verdict: PARK pending one bounded fix batch. I found no CRITICAL defect. Two HIGH finding groups block merge:

1. HIGH — active instructions still contradict the canonical merge gate. autodoc/AGENT_INDEX.md:42 is the mandatory first-read seam map and, despite its new 2026-08-01 verification stamp, still says ci-required is the “sole merge gate.” docs/REVIVAL_PLAN.md:117, the active planning spine's execution contract, still restates a fixed T3 recipe, omits the three-minute post-push floor, and says the former gate “with its aging rule is superseded.” Direct failure path: an agent follows these active instructions and merges after CI/one pass without the canonical aging and risk-calibrated review gate—the exact bypass class #1544 exists to remove. Replace both with pointer-only wording: Required CI is repository evidence; review count, aging, triage, and merge disposition come solely from the global laws and review-and-ship.

2. HIGH — the changed docs record a false transcript-retention boundary. docs/STATUS.md:6, autodoc/AGENT_INDEX.md:37, and docs/architecture/DATA_MODEL.md:560,568 call Transcript.Text the sole durable transcript-text home. Current capture still serializes transcript text into LlmRequest.Payload, and the explicitly remaining #1305 linkage work does not create/link a Transcript for that path. This is not just wording: STATUS/Data Model are operational inputs to privacy, export, deletion, and retention audits, so the false “sole home” boundary can make durable user content in LlmRequest invisible to those audits. Describe Transcript as an independent durable persistence foundation until #1305 removes or reconciles the duplicate capture payload. This confirms and broadens the existing thread: #1560 (comment)

The automatic exact-head review also posted two confirmed, non-blocking MEDIUM false-shipped-flow findings; fix or explicitly track/decline them under the canonical triage rule:

• Connector management is registry/CRUD only; no shipped connector execution path ingests content into capture: #1560 (comment)
• View linked proposal writes ?proposalId=, while both review surfaces consume only #proposal-...; deferred/terminal/non-page proposals therefore do not open as documented: #1560 (comment)

No other CRITICAL/HIGH finding remains from this pass.

Read-only verification completed: worktree guard passed at the exact head; merge base equaled the pinned base; docs governance, Golden Principles governance, GitHub-ops governance, git diff --check, tier JSON parsing, and DCO for both commits passed. NOT verified here: runtime/UI tests, a complete independent link/fragment sweep, or final Required CI completion (checks were still running when inspected).

[Chris0Jeky]

Review fix round complete at exact head 7aa9716012174c548da64eb57bbc13116c6dd925.

• HIGH active-gate drift: fixed in autodoc/AGENT_INDEX.md and docs/REVIVAL_PLAN.md; both now route disposition to live authority + canonical review-and-ship and retain only Taskdeck's Required CI evidence/lenses.
• HIGH retention boundary: fixed in docs/STATUS.md, autodoc/AGENT_INDEX.md, and docs/architecture/DATA_MODEL.md; Transcript is an independent durable store, while LlmRequest.Payload remains a second durable text store until REVIVAL-09: Durable Transcript entity + evidence spans - every proposed card deep-links to its transcript span #1305 linkage.
• MEDIUM findings: tracked as [Docs] Describe connector registry without claiming ingestion #1561 (connector registry-only docs) and [Frontend] Make agent-run proposal links reveal their target #1562 (agent-run proposal deep link), replied to, and resolved without a fix cascade.

Scoped post-fix verification: docs governance PASS; Golden Principles PASS; GitHub-ops governance PASS; failure-ledger projection 11/11; worktree-helper headless contract 1/1; git diff --check origin/main..HEAD PASS; obsolete-claim rg PASS; DCO 3/3. No other CRITICAL/HIGH finding remains from review #4836034180.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 7aa9716012

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[Chris0Jeky]

Final bounded review-fix batch pushed at edcb6257fcde7c5cb0818fbaa468321c987d943c.

• HIGH: corrected zero-or-one ERD parent cardinalities for nullable SetNull artefact/transcript relationships.
• HIGH: restored ADR-0048 / Memory-capped extraction worker process (decompression-bomb containment boundary, ADR-0048) #1429 as the hard PDF process-memory-containment gate and stated the current host-OOM residual risk.
• MEDIUM: tracked the unwired agent-run trigger claim as [Docs] Describe agent runs as API-created until triggers are wired #1563.
• MEDIUM: combined the command-palette and Workbench catalog mismatches in [Docs] Align navigation manual with palette and Workbench catalogs #1564.

All five second-round threads were replied to and resolved. This completes the canonical two-round review ceiling; no manual Codex re-review was requested. Local docs governance, Golden Principles, GitHub-ops governance, failure-ledger tests (11/11), full worktree-helper contract (28/28 on the longer second attempt), diff check, and DCO 4/4 pass. Exact-head hosted CI is now the remaining merge gate.