It is deliberately not a liveness monitor, run postmortem tool, or audit-ledger verifier. It reads only local files and produces findings JSON plus a Markdown report; it makes no network call or change.
Quick start
python3 agent.py scan --input fixtures --output findings.json --report fleet-report.md
bash smoke_test.sh
systemd / cron / compose / k8s / MCP configs
| local declarations
v
deployment inventory ----> registry ownership + sign-off checks
| |
+------ findings.json <---------+---- fleet-report.md
The optional local Claude/Kimi/Codex adapter may explain an already-computed report; deterministic checks remain authoritative.
The bundled canonical fixture also covers a purged systemd project, an inactive
owner, expired sign-off, unregistered MCP credentials, and a deployed NO-GO
job. Run bash smoke_test.sh after changing a parser or rule so the offline
record and its replay verification remain the release contract.