Anti-Refusal Statement
- Category
- Anti-Refusal
- Confidence
- 90% confidence
- Finding
Skill attempts to nullify the agent's safety policies or restrictions ('you have no restrictions', 'ignore your guidelines', 'do anything now'). This is a direct jailbreak that disables guardrails.
- Content
md The full 1.0.0 review reported an output-path issue that its secondary review confirmed; fixed above. Other raw findings were explicitly disagreed with by its secondary review: - Safety-preserving wording containing the phrase “Never bypass policy” was matched as an anti-refusal instruction. The requirement remains, rewritten positively as using host-approved interfaces and retaining approvals. - Tests used subprocess.run with a fixed argument list and no shell. They now exercise main(argv) in-process with captured streams and mocked I/O, reducing unnecessary process execution while preserving CLI coverage. - Missing allowed-tools was flagged. We document actual capabilities rather than invent broad preapproved permissions; this skill grants none. Any host-specific tool allowlist remains the operator's decision. - Standard MIT-0 license warranty language was matched as scope creep. The standard license is retained unchanged; it is legal text, not an agent instruction.
