Invariance Battery
PassAudited by ClawScan on May 12, 2026.
Overview
This instruction-only safety skill is coherent and not suspicious, though users should be aware it describes blocking agent actions and keeping persistent audit records.
This skill appears safe as an instruction-only guide for agent safety checks. Before relying on it, decide which invariants are allowed to block actions, keep those rules under user control, and avoid logging sensitive information indefinitely.
Findings (2)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
If invariants are too broad, incorrect, or influenced by untrusted content, the agent could block legitimate user-requested actions.
The skill intentionally describes a control gate that can decide whether an agent action proceeds or halts.
Agent Action Invariance Check PASS (proceed) / FAIL (halt + report)
Use this only with explicit, user-approved invariants and make any halt or failure report visible to the user.
A future implementation could store sensitive task details or incorrect failure records for a long time.
The skill describes persistent audit logging of invariant checks, which could retain sensitive context if implemented without limits.
Every check is written to the append-only spine
Define what gets logged, where it is stored, who can read it, and how long it is retained; redact sensitive data where possible.