The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill documentation instructs users to execute npx skills add agentspace-so/runcomfy-skills, which downloads and executes code from a remote repository during the skill installation process.
[EXTERNAL_DOWNLOADS]: The skill depends on the @runcomfy/cli package, which must be installed globally from the NPM registry to enable model interaction.
[COMMAND_EXECUTION]: The skill operates by invoking the runcomfy CLI tool via shell commands, passing user-supplied prompts and image URLs as arguments.
[INDIRECT_PROMPT_INJECTION]: The skill is a conduit for user-supplied content to reach a generative model, creating a surface for indirect prompt injection.
Ingestion points: User-provided prompt string and image URL processed in SKILL.md.
Boundary markers: The skill passes parameters as a JSON object to the CLI to prevent shell injection, though it lacks specific mechanisms to filter malicious instructions within the model prompt itself.
Capability inventory: The skill has the ability to execute the runcomfy CLI, which performs network operations to model-api.runcomfy.net and downloads output files.
Sanitization: The skill instructions rely on the CLI to transmit data directly via HTTPS without shell expansion, but do not describe sanitization of the content of the user prompt or image data.

flux-kontext