The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data which could potentially contain malicious instructions intended to influence the AI's behavior.
Ingestion points: User-provided prompt, tags, lyrics, and audio URLs (referenced in SKILL.md).
Boundary markers: Data is encapsulated within a JSON object passed to the --input argument of the CLI tool.
Capability inventory: The skill executes shell commands via the runcomfy CLI using the Bash tool.
Sanitization: The skill uses structured JSON input to define boundaries, though it does not explicitly mention content-level sanitization for steganographic instructions in audio files.
[COMMAND_EXECUTION]: The skill utilizes the allowed-tools frontmatter to restrict command execution to the runcomfy CLI, implementing the principle of least privilege.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of the @runcomfy/cli package via NPM. This is a standard dependency for the intended functionality and targets the official vendor package.
[CREDENTIALS_SAFE]: The skill provides secure guidance for handling API tokens, recommending the use of environment variables or a configuration file with restricted permissions (0600), and explicitly warns against logging or echoing tokens.

ai-music