skill-from-research
Pass
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external research packs which are treated as untrusted data. This creates an indirect prompt injection surface where malicious instructions hidden in research documents could attempt to influence the agent behavior during the skill creation process.
- Ingestion points: Processes files in .local/sources/ or user-specified paths via pack_inventory.py and recursive reads.
- Boundary markers: The skill explicitly instructs the agent to treat the pack as data, ignore embedded directives, and never execute commands from the pack text.
- Capability inventory: The agent has capabilities to read/write files and execute the local pack_inventory.py script and perform web searches for verification.
- Sanitization: The skill requires verification of all load-bearing claims against primary sources and redaction of PII/secrets.
- [COMMAND_EXECUTION]: The workflow involves executing a local Python script pack_inventory.py to audit the research pack.
- Evidence: Invokes python3 pack_inventory.py as a mandatory workflow step in SKILL.md. The script itself uses standard libraries and includes security checks to prevent symlink-based path traversal.
Audit Metadata