The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the runcomfy CLI tool through the Bash tool. The configuration includes allowed-tools: Bash(runcomfy *), which restricts the shell execution environment to only the intended command, adhering to security best practices for least privilege.
[EXTERNAL_DOWNLOADS]: The documentation provides instructions to install the @runcomfy/cli tool from the NPM registry. This is a standard method for distributing official software packages and utilizes a well-known service.
[PROMPT_INJECTION]: The skill contains an indirect prompt injection surface because it ingests untrusted data from external audio URLs which are then processed by the agent.
Ingestion points: The audio parameter in the audio-inpaint and audio-outpaint routes accepts HTTPS URLs to external media.
Boundary markers: Parameters are encapsulated in a structured JSON object passed to the CLI, providing some delimiter protection.
Capability inventory: Across the defined tasks, the skill has access to subprocess execution (via Bash), file-system writes for audio outputs, and network operations conducted by the CLI.
Sanitization: The skill assumes the underlying CLI correctly handles the JSON-formatted input string to prevent command injection at the shell boundary.

ace-step