You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
validate-coverage-ledger accepts an archived verifier's check; seven CLI tests fail where safe-open flags are missing #135
Hi, we vendored the security-audit skill and ran its validators. We found two issues. Both are still present at c1c8a8c.
1. An archived verifier's check can come back into the live state (validate-coverage-ledger.cjs)
priorOwners collects attempt.agent_id from archived attempts, and later attempts may not reuse those ids. Archived checks are not covered: the agent_id on an archived check (the verifier) is never added to any set. So a check made by a verifier whose attempt was archived can be copied into the live state, and the ledger still validates. That defeats the independence the archive is meant to enforce.
Suggested fix: collect archived check owners in their own set (for example priorCheckOwners). Reject any of those ids when it appears on a later attempt or a live check, and use the same error style as the existing priorOwners rejection. A fixture with an archived check whose agent_id reappears on a live check would pin it.
2. Seven CLI tests have no HAS_SAFE_INPUT_OPEN skip (validate-findings.test.cjs)
HAS_SAFE_INPUT_OPEN (line 18) gates the tests that depend on O_NOFOLLOW and related flags, and some tests already use { skip: !HAS_SAFE_INPUT_OPEN }. These seven do not, but they also go through readFileWithinLimit. Where the flags are missing, that function rejects the input before the behavior under test is reached, so each test fails with a generic error instead of skipping:
accepts a producer-shaped findings document through the CLI (line 164)
CLI rejects input above the byte limit without an exception trace (375)
CLI rejects invalid UTF-8 without replacement or an exception trace (383)
CLI rejects lone-surrogate prose without changing payload semantics (445)
CLI rejects Unicode format controls in source paths (455)
CLI rejects input above the nesting-depth limit without an exception trace (507)
CLI rejects an oversized array without an exception trace (516)
Suggested fix: add { skip: !HAS_SAFE_INPUT_OPEN } to each, the same way lines 403 and 416 already do.
Activity
Jack-Park-2026 commentedon Nov 7, 2026
Great