The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data which creates a surface for indirect prompt injection attacks.
Ingestion points: Untrusted data enters the context through image_urls, images, and image parameters in model invokes, as well as via the enable_web_search feature (SKILL.md).
Boundary markers: The skill advises agents to only ingest URLs explicitly provided by the user and notes that prompt content is passed as JSON to avoid shell-expansion (SKILL.md).
Capability inventory: The skill uses Bash(runcomfy *) to execute image generation and editing tasks which include network requests and file writing (SKILL.md).
Sanitization: The skill acknowledges the risk and recommends that agents suspect reference assets if generation diverges from the prompt (SKILL.md).
[EXTERNAL_DOWNLOADS]: The skill requires the installation of the @runcomfy/cli package from the NPM registry to function.
Evidence: The documentation instructs the operator to run npm i -g @runcomfy/cli or npx -y @runcomfy/cli (SKILL.md).
[COMMAND_EXECUTION]: The skill executes shell commands via the runcomfy CLI to perform its primary functions.
Evidence: The frontmatter restricts execution to Bash(runcomfy *), and the skill body details various runcomfy run commands (SKILL.md).

ai-image-generation