caveman
Pass
Audited by Gen Agent Trust Hub on Oct 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a persona that reformats user-facing output, creating an attack surface where untrusted data could theoretically influence the agent's logic or output clarity.\n
- Ingestion points: The skill processes user replies and technical queries as described in
SKILL.mdandREADME.md.\n - Boundary markers: Present. The skill contains explicit instructions to revert to normal prose for 'security warnings', 'irreversible actions', and 'step sequences a fragment could scramble' (defined in
SKILL.mdsection 'When to break the rules').\n - Capability inventory: None. The skill instructions do not invoke external tools, write to the filesystem, or perform network operations.\n
- Sanitization: Present. Internal rules such as 'Clarity wins' (Rule 5) and 'Payload verbatim' (Rule 6) are designed to protect technical content, commands, and negated statements from being corrupted by the stylistic transformation.
Audit Metadata