The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to ingest untrusted data from various sources to generate new agent instructions (SKILL.md files), making it vulnerable to indirect prompt injection.
Ingestion points: Untrusted data enters the context via URLs (processed by WebFetch), file contents (processed by Read, Grep, Glob), and conversation history.
Boundary markers: The skill instructions do not specify any delimiters or safety warnings to ensure the agent ignores malicious instructions embedded within the source material.
Capability inventory: The skill uses Read, Grep, Glob, WebFetch, and Write tools across its workflow files.
Sanitization: There is no evidence of content validation or sanitization before the 'distilled' content is written to new skill files.
[EXTERNAL_DOWNLOADS]: The skill utilizes the WebFetch tool to retrieve content from arbitrary URLs provided by the user. While essential for its documented purpose of learning from documentation, this involves network operations to non-whitelisted external domains.

learn