Check the rae fixture-provenance audit findings against this repository #444

Problem

OpenRAE/rae#1461 adds the fixture provenance audit for OpenRAE/rae#1344. Its record ends with findings for sibling repositories (docs/research/fixture-provenance/index.md L343-375 at 0a528fcf), observations about platforms and tools that hold wherever LilRAE, env-packs, adapters or hub use them. #1344's acceptance criteria require reporting each finding that applies to a sibling back to it. Filing them is outside #1461, and nothing in env-packs tracks them.

A first pass over env-packs' dev (132318f):

• Applies: closing keywords. The default branch is main and PRs target dev, so a closing keyword in a PR body closes nothing when the PR merges. fix(kits): release the Linux kits with an open compute substrate #434 ("Closes Kits require an exact virtual-machine substrate by accident #413", merged 2026-10-11 15:42Z) and fix: improve AI Escape participant guidance #431 ("Closes AI Escape Lab: eval-sandbox Claude needs IS_SANDBOX=1 + no-prompt alias + agent-discoverable briefing #430", merged 16:12Z) left Kits require an exact virtual-machine substrate by accident #413 and AI Escape Lab: eval-sandbox Claude needs IS_SANDBOX=1 + no-prompt alias + agent-discoverable briefing #430 open.
• Applies to any future verifier: release attestations accumulate per digest. .github/workflows/release-please.yml L353-361 attests the wheel and sdist (ADR 0015), and pack-distribution.yml signs pack releases with keyless Sigstore. Nothing in the repository counts or binds attestations today, so a verifier added later must not treat several attestations with one identity as ambiguous, nor bind to one run attempt across jobs.
• Does not apply on dev today: no workflow runs on pull_request: closed or workflow_run; no workflow or tool reads pull_request.body or run_started_at; the gh calls in release-please.yml run in jobs with a checkout (L100, L193) and build gh api paths from GITHUB_REPOSITORY (L130, L144); env-packs has no libvirt, http.server, OCI registry or conftest fixtures.
• Not yet checked: string comparison of datetime.isoformat() timestamps, which drops the fraction on a whole second, and tests that build the "real" side of an equality check through the code under test.

Scope

• Record, in this issue or in the repository, a disposition for each finding in the rae list: applies (with the fix or an issue), or does not apply (with the reason).
• For the closing-keyword finding, make the issue-closing route explicit: keep closing issues separately after the merge to dev, or close through the promotion to main with the keyword in the commit message.
• Check the tests for the two unchecked patterns and fix any match.

Done when

• Every finding in docs/research/fixture-provenance/index.md "Findings for sibling repositories" has a recorded disposition for env-packs.
• Each finding that applies is fixed or has its own open issue.

Related

• docs(research): record the fixture provenance audit and the boundary fixture convention rae#1461 (lists the sibling findings and leaves filing them outside the change)
• Audit the test suite for fixtures that assert shapes the real system never produces rae#1344