skill-curator
Pass
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes multiple other SKILL.md files which act as untrusted data sources. A malicious skill within the target library could contain instructions designed to subvert the curation process.
- [COMMAND_EXECUTION]: The skill uses a shell loop (for d in ///; do grep ... done) to inventory skills by reading their frontmatter. This is a legitimate functional requirement for the skill's stated purpose of auditing local directories.
- [INDIRECT_PROMPT_INJECTION]: Evidence Chain:
- Ingestion points: Reads all SKILL.md files located in the directory structure specified by ///.
- Boundary markers: The skill instructions do not specify any delimiters or safety markers to differentiate between the data being audited and the agent's instructions.
- Capability inventory: File system read (via grep and file reading) and file system write (implied by the Fix and verify workflow step).
- Sanitization: No sanitization or validation of the content of the audited skills is performed before processing.
Audit Metadata