Skip to content

docs(audits): Pass-7 adversarial audit of the #77-#140 wave + spec truth-up#141

Open
Reederey87 wants to merge 1 commit into
mainfrom
docs/pass7-audit
Open

docs(audits): Pass-7 adversarial audit of the #77-#140 wave + spec truth-up#141
Reederey87 wants to merge 1 commit into
mainfrom
docs/pass7-audit

Conversation

@Reederey87

Copy link
Copy Markdown
Owner

Pass 7 — seventh design & implementation audit

First adversarial audit of the #77#140 wave (no prior audit checkpoint): the OS agent
sandbox (P4-GIT-03, #107#129), the AD-1 zero-infra carrier hub (#96#106), the
supply-chain/distribution pipeline (#105#120), and the multi-device env/trust/service wave
(#131#139). Trunk audited: d667530 (#140).

Method

Verification-driven nine-dimension workflow (9 dimension reviewers → per-finding adversarial
verification
), mirroring Pass 6. The three P1 candidates were additionally cross-verified by a
second model family
(gpt-5.5 via Codex) — the project's dual-review norm applied to the audit
itself. 22 candidates → 21 confirmed (P1=3, P2=11, P3=7), 1 refuted. Two severity corrections at
verification (P7-SANDBOX-01 P2→P1, P7-SUPPLY-01 P1→P3).

Headline findings (open — for the Phase-B implementation wave)

  • P7-SANDBOX-01 (P1) — the default-on Seatbelt/bwrap/Landlock sandbox grants writes only to the
    worktree+tmp dirs, but a DevStrap agent worktree is a git linked worktree whose .git lives in
    the parent clone → agent git commit EPERMs and agent pr has nothing to push. The core loop
    silently breaks on the common Mac dev host, and the only e2e canary uses a fresh nested repo so it's
    untested. Reproduced against a real linked worktree by both model families.
  • P7-DATA-01 (P1) — the revoke-triggered env-blob rewrap is the one env-profile writer that skips
    the EnvProfileSourceCoords/envCoordLess CAS guard the other two writers apply → a concurrent
    newer profile is clobbered with stale content.
  • P7-SEC-01 (P1) — the snapshot carries no device-trust state, so a device.revoked compacted
    below the retention floor is lost on recovery → the stale-approved recovering replica re-grants the
    fresh WCK epoch to the revoked device (defeats TRUST-01 forward-secrecy-on-revoke).
  • P2/P3: no rewrap-on-approve (P7-SYNC-02), stranded blobs on transient GET (P7-SYNC-01),
    folder-carrier torn writes + non-atomic stale-lock break (P7-HUB-01/02), service exec-path
    validation / Cellar-symlink path / global label (P7-CLI-01/P7-SVC-01/02), .snowflake deny-set
    drift (P7-SANDBOX-02), property-gen coverage (P7-SYNC-03), and the curl|sh cosign gap
    (P7-SUPPLY-01).

What's in this PR

  • docs/audits/AUDIT_RECOMMENDATIONS_2026-07-07_PASS7.md — the full audit (exec summary,
    at-a-glance table, prioritized roadmap, per-finding detail with file:line evidence + verifier
    rulings, exa anchors, Appendix A/B).
  • docs/audits/README.md — Pass-7 index row + "17 open of 21" open table (header count == rows) +
    dated note. (The ledger was already current through docs: unattended-operation wave close-out (PRs #136–#139) #140; Pass 7 adds only its own rows.)
  • Doc truth-up — the four confirmed P7-DOC-* findings, fixed here: ARCHITECTURE.md, spec/00,
    and docs/quickstart.md re-synced to the shipped OS sandbox + service installers.
  • spec/18_WORK_LOG.md — newest-first entry.

Gates

  • spec-drift passed (20 specs, 6 changed files); TestEveryCommandIsDocumented +
    TestMigrationsDocumented pass; ledger header-count invariant holds (17 rows).
  • Docs-only PR (no Go source changed) → single-reviewer per the PR-cycle convention.

🤖 Generated with Claude Code

…uth-up

Seventh-pass verification-driven audit (9 dimension reviewers -> per-finding
adversarial verification; the 3 P1s cross-verified by a second model family).
22 candidates -> 21 confirmed (P1=3, P2=11, P3=7), 1 refuted.

Adds AUDIT_RECOMMENDATIONS_2026-07-07_PASS7.md + the Pass-7 ledger rows, and
fixes the four confirmed doc-staleness findings (P7-DOC-01/02/03/05): ARCHITECTURE.md,
spec/00, and docs/quickstart.md are re-synced to the shipped OS sandbox + `service`
installers. 17 of 21 findings remain open for the Phase-B implementation wave.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant