custom-skills-dev

Pass

Audited by Gen Agent Trust Hub on Aug 28, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the user to execute local validation scripts (python tests/validate_skills.py) and install development tools locally using uv. It also uses npx to run a specific version of a skill management tool (npx --yes skills@1.5.22).
  • [INDIRECT_PROMPT_INJECTION]: The skill processes skill files and repository content, creating an attack surface for instructions embedded in data.
  • Ingestion points: Files in the skills/ directory of the ai-dev-skills repository.
  • Boundary markers: No delimiters or warnings are specified for the agent when reading these files.
  • Capability inventory: The CLI tool uses subprocess calls for shell operations and git for remote repository interactions.
  • Sanitization: No specific sanitization or validation of skill file content is described in the provided references.
  • [DATA_EXPOSURE]: The skill accesses and manages resources in directories such as ~/.claude/skills/ and ~/.gemini/skills/. This behavior is a core function of the skill but involves direct access to platform-specific application data.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 28, 2026, 07:13 AM
Security Audit — agent-trust-hub — custom-skills-dev