lians-audit

/lians-audit

Run Lians operations that produce technical evidence about memory state. Parse the operation from $ARGUMENTS and execute the matching call. Always print exactly what was returned, including limitations. Do not claim that a technical result establishes legal or regulatory compliance.

Operations

snapshot <agent_id> <YYYY-MM-DD>

Exhaustive knowledge-state reconstruction - every fact valid at that instant, not a ranked top-k. The one-call demo for "show me everything the agent knew on 2025-03-14."

snap = mem.snapshot(agent_id="<agent>", as_of=datetime(...))
print(snap["total"], "facts valid at", snap["as_of"])

verify

Verify the tamper-evident SHA-256 hash chain. Report any broken link exactly as the tool returns it.

print(mem.verify_chain())   # {"status": "ok", "rows_checked": N, "violations": []}

backtest <agent_id> <YYYY-MM-DD>

Lookahead-bias detection - flags any fact the agent held that it could not have known at the simulation date. A clean report shows that this check found no lookahead flags; it does not validate the entire simulation.

r = mem.backtest_check(agent_id="<agent>", simulation_as_of=datetime(...))
print("clean" if r["is_clean"] else f"{len(r['flags'])} contamination flags")

erase <subject_id> <request_ref>

Destroys the subject's per-subject key so their encrypted content becomes unreadable while the audit hash chain survives. This is irreversible. Confirm with the user before running, and record the request_ref.

print(mem.erase(subject_id="<subject>", request_ref="<ref>"))

If the operation isn't recognized, list these four and ask which one.