radical-candor
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFENO_CODEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
- [NO_CODE]: The skill consists exclusively of Markdown instructions and documentation, with no executable scripts, binaries, or active code components.
- [EXTERNAL_DOWNLOADS]: The skill references an external Model Context Protocol (MCP) server package '@clawfu/mcp-skills' in its metadata, which indicates reliance on an external capability provider for execution context.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input describing management scenarios to generate feedback responses. Ingestion points: User-provided descriptions of workplace situations and feedback needs in the 'How to Use' section of SKILL.md. Boundary markers: Absent; there are no clear delimiters or instructions to the agent to disregard instructions within user data. Capability inventory: The skill definition does not include dangerous tools, subprocess calls, or network operations. Sanitization: Absent; input is interpolated directly into the feedback structure.
- [METADATA_POISONING]: The skill metadata contains significant inconsistencies. It lists 'ClawFu' and 'MKTG Skills' as authors, conflicting with the established vendor context of 'nvlabs'. Furthermore, the 'Skill Boundaries' section describes limitations related to audio engineering and production workflows, which are entirely irrelevant to the skill's stated purpose of teaching leadership and feedback techniques.
Audit Metadata