The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill automatically fetches content from user-provided URLs using web_fetch. It specifically targets sites like skills.sh and GitHub to retrieve Markdown content and metadata.
[COMMAND_EXECUTION]: In Phase 1.2, the skill instructs the agent to use local tools such as git clone and the GitHub CLI (gh api) to acquire repository content if web fetching fails. These commands operate on external, user-supplied repository paths.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (SKILL.md files) which serves as a significant attack surface.
Ingestion points: External content is ingested from URLs via web_fetch, git clone, or gh api and loaded into the agent's context for analysis.
Boundary markers: The instructions do not define specific delimiters or "ignore" headers to prevent the agent from following instructions embedded within the analyzed skills.
Capability inventory: The skill has the capability to write multiple report files to the docs/skill-report/ directory and perform network operations.
Sanitization: No explicit sanitization or filtering of the fetched content is mentioned before it is processed by the LLM for scoring and reporting.

skill-comparator