audit

treeship audit treats the Hub as a hostile witness: it fetches the agent's transparency log and re-proves every claim on your machine. A hub that omits records, rewrites history, or shows different views to different auditors gets caught.

Usage

treeship audit <agent-uri> --hub <url> [--watch] [--interval <secs>]

The four checks

1. Inclusion — every anchored entry's Merkle proof is re-verified offline; the checkpoint signature must chase to a signer you pinned under hub_checkpoint. Marks: anchored ✓ / anchored ✗ INVALID / anchored (proof unavailable).
2. Completeness — the entry count is compared against the evidence_anchor the agent committed to in its own signed card: complete (committed C, observed O) or OMISSION — committed C, observed O.
3. Witnessing — each observed checkpoint is persisted under ~/.treeship/merkle/witnessed/ and compared across runs: FORK DETECTED (equivocation — different roots for the same index), REGRESSION, or consistent.
4. Append-only — the Hub's consistency-proof chain is fetched and re-verified to prove the current tree extends the last witnessed one: append-only VERIFIED / INVALID / NOT PROVEN.

JSON output

{
  "verdict": "clean",
  "ok": true,
  "agent": "agent://deployer",
  "hub": "https://api.treeship.dev",
  "receipts": 12,
  "anchored": 12, "anchored_verified": 12, "anchored_invalid": 0,
  "completeness": "complete (committed 12, observed 12)", "omission": false,
  "consistency": "consistent (…)", "equivocation_or_regression": false,
  "append_only": "append-only VERIFIED — …", "append_only_failed": false,
  "entries": [ { "artifact_id": "…", "kind": "action", "action": "…", "anchored": true } ]
}

verdict is "clean" or "ANOMALY".

Exit codes