You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
docs(truth): reconcile roadmap docs that lagged the shipped code
Verified every roadmap item against code at HEAD (grep/reads plus running
prove_update_platform.py, audit_test_quality.py, validate_wiring.py, and the
flagship approval tests) before editing. The decomposed backlog and the
harness-first task set are finished in code; the docs trailed reality in three
places, now corrected:
- work-log/task-004-blocked-2026-06-13.md: added a RESOLVED banner. The
scoped-approval CLI shipped as `--approve-scoped TOOL:SHA256`
(preapproved_payload_digests) and call-id preapproval was removed (G-P2-2;
policy.py:46 kept inert), so the flagship surfaces already run the
non-deprecated path (test_first_hour_e2e_flow.py:205,
test_five_minute_proof_flow.py:222/332, five-minute-proof-demo.sh:240) and 18
approval/flagship tests pass. The pre-run scoped-approval CLI that doc waited
on is therefore unnecessary.
- roadmap-status.md packaging row: replaced the stale "no owner-platform proof
yet" with the reproducible single-platform update proof
(scripts/prove_update_platform.py emits artifact_sha256/delta_sha256/rollback_ok;
ran green 2026-07-01). Status left "Partially fixed -- unwired" because
update/* still has no CLI surface. The proof's JSON output is a regenerated,
machine-local artifact and is not committed.
- backlog-priority.md: TASK-002/003/004/006 provenance rows moved off "Promote"
to Done with evidence.
Added work-log/roadmap-verification-2026-07-01.md recording the full enumeration
with an evidence-vs-attribution boundary and the held/external remainder (cloud
GTM, RBAC enforce flip, live-model eval gate, non-maintainer sessions). Also
regenerated the docs inventories and the OKF catalog bundle for roadmap-status.
No code or behavior change.
Action: G-P2-8
Constraint: docs-only truth corrections; each status verified against code before editing; held/external items left untouched; constitution-tier positioning (TASK-001) and the 650->663 acceptance count left for Human Review
Tested: validate_docs_consistency.py passes (29/29 risk, 21/21 ticket, consistency passed); docs + OKF bundles regenerated; 18 approval/flagship tests pass; prove_update_platform.py and validate_wiring.py green
Confidence: high
| TASK-003 test typing pass |`harness-migration`|**Done** — all 586 test files typed (contract/behavior/adversarial/lifecycle) per `scripts/audit_test_quality.py`|
39
+
| TASK-004 flagship tests off deprecated approval |`governance-gap`|**Done (2026-07-01)** — G-P2-2 removed call-id preapproval; flagship uses `--approve-scoped`; see `work-log/task-004-blocked-2026-06-13.md` resolution|
Copy file name to clipboardExpand all lines: docs/retrospective/06-action-register.md
+4-3Lines changed: 4 additions & 3 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -64,6 +64,7 @@ Legend: ✅ Done (committed) · 🟡 In progress · ⬜ Not started · 🔵 Alre
64
64
| G-P2-5 | Governance | Make `max_skill_md_lines` an error for installed (candidate-provenance) skills; keep it a warning during development. |`teaagent/skill_review.py:187-193`| Error/warning split by provenance. | ✅ | Candidate-provenance installed skills receive an error; development skills receive a warning; `tests/test_skill_review_installed_strictness.py`|
65
65
| G-P2-6 | Governance | Ratify the shell-mutation reversibility boundary for SEC-11 in an ADR: decide the two-tier undo model and the explicit out-of-scope external-effect non-goal. |`docs/adr/0042-shell-mutation-reversibility-boundary.md`; `docs/adr/README.md`| ADR-0042 accepted; SEC-11 register row cites it; two-tier model (`GitBranchSandbox` transactional rollback for in-worktree; disclosed `UndoJournal` fallback) documented; external effects declared out of scope by design. | ✅ | This session — ADR-0042 authored; SEC-11 boundary decided and cross-referenced; no code change (decision record). |
66
66
| G-P2-7 | Governance | Reconcile docs-truth drift surfaced by the origins/dormant-surfaces analysis: correct the README documented-but-uncalled claims (D1–D6) and the invalid audit/hook surfaces in user-facing guides (§3c). | `README.md`; `docs/cli.md`; `docs/api/integration-guide.md`; `docs/onboarding-security-engineers.md`; `docs/onboarding-ml-researchers.md`; `docs/governance/trust-and-audit-whitepaper.md`; `docs/guides/getting-started-{security-reviewer,team-operator}.md` | README claims match code (failure warnings on-demand not auto-injected; TUI-only pin refresh; false pin-count indicator removed; static-analysis not LSP; post-run not pre-commit; RAG archive is a helper, not auto-invoked); guides use the real audit CLI (`audit export <run_id>`, `audit verify --root .`, no `analyze`/`archive`) and the real hooks API (`register_pre_hook`/`register_post_hook`, `HookEvent` names). D7 (constitution-tier 650→663) deferred to Human Review. | ✅ | This session — verified each claim against code (`get_failure_warnings`/`archive_to_rag`/`ValidationRunner` uncalled; `_prompt()` has no pin count; real `audit` subcommands; `hooks.py` API) then corrected 8 files. |
67
+
| G-P2-8 | Governance | Reconcile roadmap docs that lagged the code: resolve the stale TASK-004 "blocked" work-log, correct the H6 "no update proof" exit-evidence, and move done TASK provenance rows off "Promote". |`docs/work-log/task-004-blocked-2026-06-13.md`; `docs/roadmap-status.md`; `docs/backlog-priority.md`; `docs/work-log/roadmap-verification-2026-07-01.md`| task-004-blocked carries a RESOLVED banner (G-P2-2 shipped `--approve-scoped`; flagship migrated, 18 tests pass); H6 exit-evidence cites the verified single-platform update proof; TASK-002/003/004/006 rows read Done; a dated verification note records the full enumeration. | ✅ | This session — each status verified against code (grep/reads, `prove_update_platform.py`, `audit_test_quality.py`, flagship tests) before editing. |
67
68
| A-P2-1 | Architecture | Consolidate optional-extra aliases: fold `crypto`/`oauth`/`audit-encryption` into a single `crypto` extra. |`pyproject.toml:60-68`| Single `crypto` extra, aliases removed. | ✅ |`pyproject.toml` exposes one `crypto` extra; obsolete aliases are absent |
68
69
| A-P2-2 | Architecture | Split CLI handler god modules: `_ergonomics.py:1407`, `_doctor.py:1022`, `_agent/run.py:999`, `chat_repl.py:894`. | As listed | Each module under 800 lines. | ✅ |`_doctor`/`_ergonomics`/`_agent/run.py` split; `chat_repl.py` (906) eliminated by retiring it (U-P2-1); `_ergonomics/approval.py` (1,014) split — extracted `approval_doctor_command`→`_approval_doctor.py` (306) and `approval_preset_command`→`_approval_preset.py` (168), leaving `approval.py` at 567. `check_god_modules` clean at threshold 800. |
69
70
| A-P2-3 | Architecture | Expand property-based testing with Hypothesis strategies for permission-mode transitions, budget enforcement, approval-hash exactness, and audit-chain invariants. |`tests/`| Hypothesis tests exist for the four areas. | ✅ | All four invariant families are covered in `tests/test_hypothesis_invariants.py`|
@@ -83,14 +84,14 @@ Legend: ✅ Done (committed) · 🟡 In progress · ⬜ Not started · 🔵 Alre
| H6 | Owner packaging and local distribution | Desktop/client-server and local release channels have supply-chain, update, rollback, and support plans for owner-operated use | TBD | Partially fixed — unwired | Low | WDA-005 |`update/*`package implemented but unwired; no owner-platform proof yet |
36
+
| H6 | Owner packaging and local distribution | Desktop/client-server and local release channels have supply-chain, update, rollback, and support plans for owner-operated use | TBD | Partially fixed — unwired | Low | WDA-005 |Single-platform update proof is reproducible via `scripts/prove_update_platform.py` — emits `artifact_sha256`, `delta_sha256`, and `rollback_ok` for a from→to upgrade (ran green 2026-07-01; its JSON output is a regenerated, machine-local artifact, not committed). `update/*`remains intentionally not CLI-wired (no `teaagent update` daily surface); no desktop packaging / session-attach proof yet |
0 commit comments