The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the runcomfy CLI tool through a restricted Bash tool configuration (Bash(runcomfy *)). This ensures the agent can only execute the intended vendor-specific commands.\n- [EXTERNAL_DOWNLOADS]: Includes instructions to install the official @runcomfy/cli package from the NPM registry, which is the standard distribution method for the tool.\n- [PROMPT_INJECTION]: The skill provides a surface for processing untrusted media URLs provided by the user. It incorporates safety instructions directing the agent to validate user intent and refuse requests involving non-consensual or harmful content.\n
Ingestion points: Media URLs (image, audio, video) are ingested via the --input JSON argument in SKILL.md.\n
Boundary markers: Data is passed to the CLI as a structured JSON string, maintaining a clear boundary.\n
Capability inventory: Subprocess execution and network operations are limited to the runcomfy CLI as described in SKILL.md.\n
Sanitization: Content validation is handled at the model API level, with explicit agent instructions to refuse malicious use cases.

face-swap