The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and process external data, such as vendor documentation and generated drafts, which introduces a surface for indirect prompt injection. 1. Ingestion points: User-supplied drafts and external documentation URLs cited in the metadata. 2. Boundary markers: The skill body explicitly addresses 'instruction-boundary safety' and provides guidelines for separating teaching content from metadata. 3. Capability inventory: The skill has access to shell execution (Bash), file reading (Read, Grep), and file modification (Write, Edit) tools. 4. Sanitization: Verification is performed via mandatory linting and human review gates before committing content.
[COMMAND_EXECUTION]: The authoring workflow involves the execution of local repository scripts such as skill-lint.js and check-protocol-consistency.js to validate skill metadata and protocol consistency.
[EXTERNAL_DOWNLOADS]: The skill metadata and references point to external documentation and technical specifications from well-known services, including Anthropic's GitHub repository, OpenAI, and Claude documentation for truth certification purposes.

skill-scaffold