Skip to content

Six untracked audit findings from #2644: two shim-check universe gaps and four run-idempotency defects #2664

Closed as not planned

Description

@noahgift

Consolidating the remaining #2644 audit findings that lived only in a scratch file. Filing them so the record is on GitHub. All were skeptic-confirmed; the gap findings additionally carry a no-op discrimination control (the guard was verified GREEN on pristine input before the mutation was believed).

Shim-check universe gaps

SHIM-2644-02 — the checker keys on a path, the shadowing mechanism keys on a name.
check_dogfood_shim.sh check 4's universe is the single hardcoded path ~/.claude/skills/dogfood/SKILL.md. But skill shadowing keys on the frontmatter name: field — the checker's own #2332 comment says so. A user-scope SKILL.md under any other directory carrying name: dogfood shadows the repo skill and passes the check.

SHIM-2644-03 — check 1's universe is git ls-files, so an UNTRACKED copy passes.
An untracked full copy of the runner inside the working tree still yields ok 1 exactly one tracked dogfood.sh and an overall PASS. The pre-#2640 defect state was precisely an untracked copy — the guard cannot see the state it was built to prevent.

Run-idempotency defects (found by the twice-run, order 2b)

Two consecutive full runs at the same commit, receipts diffed gate-by-gate.

TWICE-1 — the runner dirtied the tree it then measured. Every run mutated .pv/contracts.idx, .pv/contracts.idx.mtime and .pv/lint-previous.json — tracked files — flipping run 2's git-clean from PASS to WARN. (Fixed in QUAL-013 (#2651) by untracking .pv and gitignoring the artefact dirs; recorded here for completeness.)

TWICE-2 — a run materializes .cargo/config.toml in the repo (absent at worktree creation, present after run 1), changing later runs' cargo behavior. Run 2's clippy note cites warning: using config file.

TWICE-3 — byte-identical consecutive receipts are impossible by construction: the receipt embeds its timestamp and the filename encodes it. The determinism standard must therefore be stated as "identical modulo timestamp" — stated, not assumed.

TWICE-4 — gate() note capture is nondeterministic across identical-result runs. It grabs the first incidental line matching error|fail|warning, so environment noise rewrites receipt notes between runs that agree on every verdict. (Same root cause as #2663, which carries the fix.)

Verdict was stable across both runs (NO-GO/NO-GO, same 8 RED), so these are integrity defects in the evidence, not verdict flips.

Refs #2644 audit · #2663 (RCPT-1) · #2651 (QUAL-013) · #2640

Activity

  1. added this to the 0.69.0 milestone on Sep 10, 2026
  2. added
    automationBot-maintained dashboard, not a defect — skip in triage
    P1High priority
    on Sep 11, 2026
  3. modified the milestones: 0.69.0, 0.71.0, 0.72.0 on Sep 17, 2026
  4. noahgift commented on Sep 25, 2026

    @noahgift
    MemberAuthor

    Parked by APR-EPIC-001: over S3 budget (rank 17/142). Reopen = pull. Applied by aprender-d1 under a 3-lane quorum receipt (handoff/balance-quorum-rev4.json).

  5. modified the milestones: 0.72.0, backlog on Sep 25, 2026
  6. added
    parkedClosed over epic budget (APR-EPIC-001 rule 5). Reopen = pull.
    on Sep 25, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    P1High priorityautomationBot-maintained dashboard, not a defect — skip in triageparkedClosed over epic budget (APR-EPIC-001 rule 5). Reopen = pull.

    Type

    No type

    Projects

    No projects

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions