The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external JSON fixtures and user-provided descriptions for custom art workflows, creating a surface for potential injection.
Ingestion points: The fixture command reads local JSON files (fixture.json) to populate skill manifests, and the art command processes user-supplied descriptions to generate thumbnail images.
Boundary markers: The skill instructions strictly require the agent to preview every mutation and obtain explicit user approval for a SHA-256 hash representing the planned changes before any write operation is performed.
Capability inventory: The associated CLI script performs extensive filesystem operations (creating directories, writing/updating markdown, JSON, and YAML files) and network requests for image generation.
Sanitization: String inputs are validated to remove control characters and are restricted by maximum lengths. The skill includes a custom PNG validator (scripts/png.mjs) that enforces exact dimensions (1024x1024) and forbids ancillary or metadata chunks that could contain hidden instructions.
[COMMAND_EXECUTION]: The core functionality of the skill involves executing a local Node.js script (scripts/create-skill.mjs) to modify the repository. This script automates the creation of skill structures and the updating of multiple registration files including README.md, marketplace.json, and GitHub workflow files.
[EXTERNAL_DOWNLOADS]: The skill facilitates the download of image assets from OpenAI (api.openai.com) and Azure OpenAI endpoints. To mitigate risks associated with remote calls, the skill implements a strict fetch wrapper (scripts/providers.mjs) that validates the request origin and rejects all HTTP redirects.

create-skill