skill-evolver
Pass
Audited by Gen Agent Trust Hub on May 13, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface.\n
- Ingestion points: Content from external URLs (Anthropic, OpenAI, DeepMind, GitHub) and local sibling skills.\n
- Boundary markers: Uses a structured template but lacks robust delimiters to isolate external content from instructions.\n
- Capability inventory: Can invoke sub-skills (
/research,/umbrella), execute local scripts, and read files.\n - Sanitization: No validation of ingested external content is performed before use.\n- [COMMAND_EXECUTION]: The skill executes a local helper script
scripts/validate-args.shfor input validation. This script is a vendor-provided resource and uses regex filtering to prevent command injection.\n- [EXTERNAL_DOWNLOADS]: The skill references and researches information from reputable external sources such as Anthropic, OpenAI, and DeepMind documentation, as well as GitHub.
Audit Metadata