create-skill

Pass

Audited by Gen Agent Trust Hub on Aug 12, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses shell commands (ln -s, ls) to manage symbolic links within the .claude/skills directory and verify the successful creation of new skill folders. These operations are restricted to the local workspace structure.\n- [PROMPT_INJECTION]: The skill accepts user-provided requirements and roles to generate new SKILL.md instructions, creating an indirect prompt injection surface.\n
  • Ingestion points: User input describing the desired skill is passed to the skill-author sub-agent in Step 2.\n
  • Boundary markers: The workflow relies on specialized review agents (Step 4) rather than text-based delimiters to validate the safety of generated content.\n
  • Capability inventory: The skill possesses the capability to write files and create symbolic links within the repository.\n
  • Sanitization: A skill-reviewer agent is explicitly instructed to scan the generated output for hardcoded API keys and attempts to bypass pre-commit hooks.\n- [EXTERNAL_DOWNLOADS]: The skill references sub-agent configurations from the Fandhe-AI/agent-cli-skills repository to load its functional components. This is a standard vendor-provided resource used to support the scaffolding workflow.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 12, 2026, 01:17 AM
Security Audit — agent-trust-hub — create-skill