The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the @runcomfy/cli Node.js package for its core functionality. It instructs installation via the standard npm registry or execution via npx, which are recognized and safe methods for obtaining vendor-provided tooling.
[COMMAND_EXECUTION]: The skill uses the Bash(runcomfy *) tool to execute video processing commands. These operations are limited to the specific runcomfy CLI tool as defined in the skill's frontmatter configuration, adhering to the principle of least privilege.
[CREDENTIALS_UNSAFE]: The documentation references authentication via a local configuration file (~/.config/runcomfy/token.json) or an environment variable (RUNCOMFY_TOKEN). This follows standard security practices for CLI tools and does not involve hardcoding secrets within the skill itself.
[PROMPT_INJECTION]: The skill ingests untrusted data via video_url, audio_url, and image_url parameters. This presents an indirect prompt injection surface where instructions could theoretically be embedded in the metadata or content of the linked media. However, the risk is mitigated by the following:
Ingestion points: Asset URLs provided by the user in the CLI input.
Boundary markers: Parameters are passed within a structured JSON object to the CLI.
Capability inventory: Capabilities are restricted to generating video files via the authorized runcomfy binary.
Sanitization: The skill explicitly notes that the CLI does not perform shell expansion on input content, reducing the risk of command injection through these fields.

lipsync