Skills
Navigate
skills/runcomfy-com/skills/ace-step/Gen Agent Trust Hub

ace-step

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests audio content from user-provided URLs for inpainting and outpainting operations, which is a potential surface for indirect prompt injection. \n
  • Ingestion points: The audio field defined in the audio-inpaint and audio-outpaint route schemas in SKILL.md. \n
  • Boundary markers: The skill uses structured JSON to pass input parameters to the CLI tool. \n
  • Capability inventory: The skill performs model processing using the runcomfy CLI (invoked via Bash(runcomfy *) in SKILL.md). \n
  • Sanitization: Input content is passed as a JSON string to the CLI to prevent shell command injection. \n- [COMMAND_EXECUTION]: The skill executes the runcomfy command to interact with AI models. This execution is scoped and restricted via the allowed-tools frontmatter field. \n- [EXTERNAL_DOWNLOADS]: The skill documentation recommends installing the @runcomfy/cli package, which is a vendor-owned resource for accessing the RunComfy API.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 04:18 PM
Security Audit — agent-trust-hub — ace-step