The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill includes instructions to download implementation guides and reference material using curl. These downloads target the official Anthropics skills repository and the official Model Context Protocol documentation, which are trusted or well-known service sources for agent capabilities.
[INDIRECT_PROMPT_INJECTION]: The skill requires the agent to analyze third-party API documentation and semantic data to design tools. This creates a surface where malicious instructions embedded in external documentation could influence tool implementation. The skill mitigates this by instructing developers to implement strict input validation and security reviews for every tool.
Ingestion points: Step 1 in SKILL.md (researching API semantics) and the exploration phase in references/evaluation.md.
Boundary markers: The skill relies on human-in-the-loop confirmation and manual verification of generated answers as specified in the evaluation instructions.
Capability inventory: The skill uses Bash, Write, and Edit tools to generate and modify server code based on the researched documentation.
Sanitization: Documentation in references/design-and-implementation.md explicitly mandate that all model-supplied arguments are never trusted and must be validated at the tool's boundary.
[REMOTE_CODE_EXECUTION]: The installation example in the SKILL.toon metadata file uses npx to fetch and add the skill from the author's (akillness) repository. This is a standard distribution pattern for the vendor and is used for skill management.
[COMMAND_EXECUTION]: The skill provides a local utility script scripts/validate-evals.mjs used to verify the integrity and safety coverage of evaluation suites. The script performs read-only filesystem operations and regex-based validation on local files and does not perform network operations.

mcp-builder