audit-3: append-only audit log + strict legacy-prefix rule + anchor cross-check
ci / lint-test-leaks (pull_request) Canceled after 0s

- verify_audit_chain(): rows with entry_hash IS NULL are only tolerated
  BEFORE the first hashed row (documented legacy prefix). A NULL hash
  AFTER the chain started is tampering and fails verification at that
  row id — previously any NULL row anywhere was silently skipped, so an
  attacker with UPDATE could erase intermediate evidence undetected.
- migration 0004: audit_log becomes append-only. SQLite: BEFORE UPDATE/
  DELETE triggers RAISE(ABORT). Postgres: BEFORE triggers RAISE EXCEPTION.
  Closes the hole for the table owner and direct SQL connections.
- monitor.verify_chain_against_anchors(): compares the chain tail against
  anchors published to Nostr. Explicit statuses: no_anchors (never
  anchored / notes missing on relays), local_only (no fetcher), ok
  (relay note attests the tail or the last local anchor), mismatch.
- tests: NULL-before-genesis ok, NULL-in-middle fails at that id, NULL
  late fails, UPDATE rejected, DELETE rejected, INSERT still allowed,
  no-anchors explicit, local-only explicit, mismatch detected, match ok.

Gate: ruff 0 · 81 unit + 1 PG integration · alembic 0001→0004 on PG ·
gitleaks clean · make check exit 0.
This commit is contained in:
2026-10-28 19:56:50 +00:00
parent 029e4f77d7
commit 09fe9199b3
4 changed files with 406 additions and 4 deletions
+117
View File
@@ -251,3 +251,120 @@ def build(cfg: Config | None = None) -> int:
return 0
finally:
engine().dispose()
def _latest_chain_hash(sess) -> str | None:
"""Last non-NULL entry_hash in the audit chain (None if chain empty)."""
from cayceops.models import AuditEntry
last = (
sess.query(AuditEntry)
.filter(AuditEntry.entry_hash.isnot(None))
.order_by(AuditEntry.id.desc())
.first()
)
return last.entry_hash if last else None
def _published_anchors(sess) -> list[dict]:
"""Anchor audit entries recorded locally, oldest first."""
from cayceops.models import AuditEntry
rows = (
sess.query(AuditEntry)
.filter(AuditEntry.action == "anchor")
.order_by(AuditEntry.id.asc())
.all()
)
return [
{
"event_id": (r.detail or {}).get("event_id"),
"last_hash": (r.detail or {}).get("last_hash"),
"last_entry": (r.detail or {}).get("last_entry"),
"ts": r.ts.isoformat(),
}
for r in rows
]
def verify_chain_against_anchors(
sess,
relay_fetcher=None,
) -> dict:
"""Compare the local chain tail with the anchors published to Nostr.
relay_fetcher(deployment) -> list of note contents (strings) found on
the relays for this deployment's anchor tag. When None, only local
coherence is checked and the result says so.
Returns a report dict:
status: "no_anchors" | "local_only" | "ok" | "mismatch"
detail: human-readable explanation
local_last_hash, local_anchors, relay_notes (when fetched)
"""
local_hash = _latest_chain_hash(sess)
anchors = _published_anchors(sess)
report: dict = {
"local_last_hash": local_hash,
"local_anchors": anchors,
"relay_notes": None,
}
if not anchors:
report["status"] = "no_anchors"
report["detail"] = (
"the chain has never been anchored to Nostr — no external "
"attestation exists yet (anchor once with `monitor anchor`)"
)
return report
last_anchor = anchors[-1]
if relay_fetcher is None:
report["status"] = "local_only"
report["detail"] = (
"compared only against the LOCAL record of past anchors "
"(no relay fetch configured); last anchored hash "
f"{last_anchor['last_hash']}"
)
return report
from cayceops.config import load as _load
deployment = _load().deployment_id
notes = relay_fetcher(deployment) or []
report["relay_notes"] = notes
if not notes:
report["status"] = "no_anchors"
report["detail"] = (
"local anchor records exist but no anchor notes were found on "
"the relays — the external attestation is missing or was deleted"
)
return report
# the freshest relay note must attest the current chain tail
expected_fragment = f"last_hash={local_hash}"
latest_note = notes[-1]
if expected_fragment in latest_note:
report["status"] = "ok"
report["detail"] = "latest relay anchor matches the current chain tail"
return report
# maybe the chain advanced after the last anchor — that's fine ONLY if
# the note matches the hash recorded in the last local anchor entry
if last_anchor["last_hash"] and last_anchor["last_hash"] in latest_note:
report["status"] = "ok"
report["detail"] = (
"relay anchor matches the last local anchor; the chain has "
"advanced since (anchor again to attest the new tail)"
)
return report
report["status"] = "mismatch"
report["detail"] = (
f"relay anchor does not attest chain tail {local_hash} nor the last "
f"local anchor {last_anchor['last_hash']} — the published anchor "
"does not correspond to this database"
)
return report
+11 -4
View File
@@ -209,14 +209,21 @@ def _build_entry(
def verify_audit_chain(sess) -> tuple[bool, int | None]:
"""Re-walk the chain. Returns (ok, first_broken_entry_id or None).
Legacy rows (entry_hash IS NULL, written before migration 0002) are
skipped and documented: the chain begins GENESIS at the first hashed row,
so a migrated database verifies clean without any tampering.
Rows with entry_hash IS NULL are ONLY tolerated BEFORE the first hashed
row (the documented legacy prefix from before migration 0002). A NULL
hash AFTER the chain has started is tampering — an attacker erasing the
hash of an intermediate row to hide an edit — and fails verification at
that row's id.
"""
prev_hash: str | None = None # None = still in legacy prefix
chain_started = False
for entry in sess.query(AuditEntry).order_by(AuditEntry.id).all():
if entry.entry_hash is None:
continue # legacy row: outside the chain by documented genesis
if chain_started:
# tampering: a hash was erased in the middle of the chain
return False, entry.id
continue # legacy prefix row: outside the chain by genesis
chain_started = True
if prev_hash is None:
expected_prev = "GENESIS" # first hashed row anchors the chain
else: