The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents an attack surface where source video URLs processed by the CLI are untrusted and could contain hidden instructions. It provides a clear evidence chain and mitigations:
Ingestion points: Untrusted video_url enters the context via the runcomfy run command arguments in SKILL.md.
Boundary markers: The documentation instructs the agent to ingest only user-provided URLs and provides guidelines for detecting divergence from prompts.
Capability inventory: The skill uses runcomfy run (SKILL.md) to perform network operations and local file writes to an output directory.
Sanitization: The skill explicitly states that the CLI does not shell-expand prompt content, mitigating shell injection.
[COMMAND_EXECUTION]: The skill uses the runcomfy CLI to execute video extension tasks. This usage is confined to the specific vendor tool declared in the allowed-tools frontmatter and is consistent with the skill's primary purpose.
[EXTERNAL_DOWNLOADS]: The documentation references installing the @runcomfy/cli package via npm. This is a standard installation procedure for the vendor's own verified tools and is documented for operator setup.
[CREDENTIALS_UNSAFE]: The skill describes how the CLI manages API tokens, including the storage path at ~/.config/runcomfy/token.json. It provides best practices for managing these secrets, such as using environment variables and avoiding echoing them in logs.

video-extend