source-eval

Pass

Audited by Gen Agent Trust Hub on Apr 22, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to list files in .claude/skills/ or reference management files to check for duplicate functionality.
  • [EXTERNAL_DOWNLOADS]: The skill is designed to fetch and process content from external URLs (technical blogs, official documentation, GitHub repositories) for evaluation.
  • [INDIRECT_PROMPT_INJECTION]: The skill's primary function involves ingesting and analyzing untrusted data from external sources. There is a risk that instructions hidden within this content (e.g., in HTML comments or markdown) could attempt to bypass the evaluation criteria or trigger unintended agent actions.
  • Ingestion points: External URLs and local file paths provided by the user.
  • Boundary markers: None specified in the instructions to prevent the agent from obeying instructions found within the source content.
  • Capability inventory: File system read (local files/URLs), file system write (creating evaluation notes in sources/), and directory listing.
  • Sanitization: No explicit sanitization or validation of the external content is mentioned before processing.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 22, 2026, 09:33 PM