latent-sre/homelab-agents/lab-audit

Audit the lab against its own standards and report like a code review of the infrastructure: severity-ranked, evidence-cited, no finding without the command output that proves it.

Checks (run what's applicable; list what you couldn't run and why)

All checks are read-only. disallowed-tools removes Write and Edit while this skill is active, but Bash can still mutate (redirects, docker rm), so the mandate is still yours: inspection commands only — fixes route to sde-agents:homelab-engineer. Whether you were invoked directly from the main session or under homelab-engineer, the reviewer's Bash guard does not cover this skill (that hook keys on guarded agent identities, and the main loop carries none at all) — the read-only-ness here is cooperative, not enforced. NotebookEdit is in disallowed-tools for the same reason as Write and Edit: it is a write tool, and a denylist that names only the obvious two leaves the third. Fan the checks out in parallel (per host or per area) rather than sweeping serially.

The eight checks — exposure, host/container hygiene, certificates, backups, monitoring gaps, drift, capacity, updates — live with their command-level detail in `references/checks.md`; read it before sweeping. Run what applies, and name what you skipped in the denominator.

This is the hygiene sweep: is the lab well-kept. The adversary's sweep — what an attacker in a given position can reach, move through, or take (trust zones, authn, management planes, credentials, secrets, reachable vulnerabilities, family data) — is sde-agents:security-audit. Same read-only posture, same ledger format, different question.

Output

Open with the coverage denominator — hosts covered and checks run vs. skipped, with why (e.g. "3/4 hosts; 6/8 checks — backups and drift skipped: no repo access") — findings without a denominator overstate the sweep. Then [P0]–[P3] findings, each with the evidence (command + output) and the one-line fix. Grade exposure by unauthorized capability or data reached, not the absence of auth on intended public content; irreplaceable state without backup is graded against its declared loss tolerance. End with the top three things to fix this weekend — not a list of thirty. After the top three, emit the findings-ledger rows (format at the end of `references/checks.md`) for the operator to append to the lab repo's ledger — this skill holds no write tools, so the emitted block IS the ledger entry.