audit: claude-shakedown 2026-07-09 dossier + San execution prompt #6

Merged
j7 merged 2 commits from audit/claude-shakedown-2026-07-09 into main 2026-07-09 19:45:51 -04:00
Owner

Branch: audit/claude-shakedown-2026-07-09main
Reviewer: Jigo (per PR-2 Q9 precedent: San preflight, Jigo review, J7 gate)
Files: documents/audits/claude-shakedown-2026-07-09.md (new), prompts/san-execute-shakedown.md (new). No canonical docs modified — all fixes land via the Lane A hardening PR that San will open.

What this is

The external Claude shakedown deliverable per prompts/claude-shakedown.md, plus the operational prompt that turns its accepted findings into kanban cards.

Dossier headline (full detail in §4.3)

  1. Spec §2 item 3 contradicts itself on the multiplex target — blocks P0's definition of done; needs ADR-0004.
  2. Honcho is "removed"/"de-scoped"/"OK" in three docs and still bound to 0.0.0.0:8000 — an unmade stop/keep decision.
  3. board-schema.md lacks forbidden_paths + report_back_channel, so external-workers §4 is structurally unenforceable; no transition rules despite wolf-solo move authority.
  4. GBrain double-carded (E2-T1 / E3-T12) with E1-T5 mis-citations in two docs — San would mint duplicates on day one.
  5. AMB :6380 absent from both the spec component table and the runtime-state port map while P8 works its ACLs.

Plus: answers to the 9 PR-2 open questions (5 settled from docs, 3 need J7, 1 answered by precedent), reconstruction of the 4 deferred Jigo minors (3 should land pre-Sprint 0), 7 ranked recommendations, and §4.7 false-negative accounting (no approval bypasses, no committed secrets, no second control plane, no paused-component allocation found).

Merge gates

  • Jigo review of both files
  • J7 answers Gate-0 D1 (multiplex target) and D2 (Honcho fate) — may happen post-merge but pre-Sprint-0
  • On merge: run prompts/san-execute-shakedown.md against San

Note on documents/

New top-level tree per shakedown §0 ("audit is ephemeral until promoted") — deliberately outside canonical specs//docs/. If you'd rather not introduce it, docs/audits/ works with a one-word edit; ADR-0003 boundary purists may prefer keeping ephemera out of docs/ though.

**Branch:** `audit/claude-shakedown-2026-07-09` → `main` **Reviewer:** Jigo (per PR-2 Q9 precedent: San preflight, Jigo review, J7 gate) **Files:** `documents/audits/claude-shakedown-2026-07-09.md` (new), `prompts/san-execute-shakedown.md` (new). No canonical docs modified — all fixes land via the Lane A hardening PR that San will open. ## What this is The external Claude shakedown deliverable per `prompts/claude-shakedown.md`, plus the operational prompt that turns its accepted findings into kanban cards. ## Dossier headline (full detail in §4.3) 1. Spec §2 item 3 contradicts itself on the multiplex target — **blocks P0's definition of done**; needs ADR-0004. 2. Honcho is "removed"/"de-scoped"/"OK" in three docs and still bound to `0.0.0.0:8000` — an unmade stop/keep decision. 3. `board-schema.md` lacks `forbidden_paths` + `report_back_channel`, so external-workers §4 is structurally unenforceable; no transition rules despite wolf-solo move authority. 4. GBrain double-carded (E2-T1 / E3-T12) with E1-T5 mis-citations in two docs — San would mint duplicates on day one. 5. AMB `:6380` absent from both the spec component table and the runtime-state port map while P8 works its ACLs. Plus: answers to the 9 PR-2 open questions (5 settled from docs, 3 need J7, 1 answered by precedent), reconstruction of the 4 deferred Jigo minors (3 should land pre-Sprint 0), 7 ranked recommendations, and §4.7 false-negative accounting (no approval bypasses, no committed secrets, no second control plane, no paused-component allocation found). ## Merge gates - [ ] Jigo review of both files - [ ] J7 answers Gate-0 D1 (multiplex target) and D2 (Honcho fate) — may happen post-merge but pre-Sprint-0 - [ ] On merge: run `prompts/san-execute-shakedown.md` against San ## Note on `documents/` New top-level tree per shakedown §0 ("audit is ephemeral until promoted") — deliberately outside canonical `specs/`/`docs/`. If you'd rather not introduce it, `docs/audits/` works with a one-word edit; ADR-0003 boundary purists may prefer keeping ephemera out of `docs/` though.
j7 added 2 commits 2026-07-09 18:32:59 -04:00
Deliverable per prompts/claude-shakedown.md §4. Contains: verified-vs-inferred
table, per-doc audit, 10 ranked issues, answers to the 9 PR-2 open questions,
reconstruction of the 4 deferred Jigo minors, 7 ranked recommendations, and
explicit false-negative accounting (§4.7).

Headline: multiplex target contradiction blocks P0 (needs ADR-0004);
Honcho de-scoped but live on :8000; board-schema cannot enforce
external-workers §4; GBrain double-carded (E2-T1/E3-T12) + E1-T5 mis-citations;
AMB :6380 absent from both canonical maps.

Ephemeral until promoted (per shakedown §0.4); findings become ADRs only
when a recommendation becomes an architectural decision.
Companion to documents/audits/claude-shakedown-2026-07-09.md. Defines:
Gate 0 (two J7 decisions: multiplex target ADR-0004; Honcho runtime fate),
Lane A (single pre-Sprint-0 hardening PR: schema fields + transitions,
GBrain dedup, cross-ref sweep, AMB visibility, example packets, two Jigo
minors), Lane B (re-sequence Outpost Traefik fix), and an explicit
parked list (GBrain skeleton, MC/Honcho/flake questions -> open-question
entries, not cards). All cards wolf-mode per external-workers §3.
j7 merged commit d6d0595b77 into main 2026-07-09 19:45:51 -04:00
j7 deleted branch audit/claude-shakedown-2026-07-09 2026-07-09 19:47:28 -04:00
Sign in to join this conversation.
No reviewers
No labels
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
j7/StarrOS!6
No description provided.