create-agent-skill

Pass

Audited by Gen Agent Trust Hub on Aug 26, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill generates new agent instructions by ingesting user-provided source material (prompts, contexts, and files), which creates an attack surface for instructions hidden in data. \n
  • Ingestion points: User inputs such as prompts, context files, and workflows defined in SKILL.md (Step 1). \n
  • Boundary markers: Confirmation requests in Step 1 and a final draft review in Step 5. \n
  • Capability inventory: File writing (SKILL.md, scripts) and script execution (Scripts section). \n
  • Sanitization: Relies on human-in-the-loop review of the draft before finalization (Step 5).\n- [COMMAND_EXECUTION]: The workflow instructs the agent to execute newly created scripts for validation purposes before finalizing the skill. \n
  • Evidence: SKILL.md section 'Scripts' ('Run bundled scripts and exact commands end to end once before delivery').
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 26, 2026, 09:26 PM
Security Audit — agent-trust-hub — create-agent-skill