The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data by instructing the agent to read project files, code, and documentation to extract facts for analysis.
Ingestion points: The agent is directed to read the actual code, config, docs, or data related to the user's current decision or bug.
Boundary markers: No specific technical delimiters are mentioned, but the skill enforces a logical boundary by requiring facts to be verified and distinguished from inferences.
Capability inventory: The agent can read files and delegate research tasks to subagents.
Sanitization: The instructions mandate that all gathered information must be explicitly tagged as verified facts or reasoned inferences, which mitigates the risk of blindly following instructions embedded in the data.
[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The documentation provides commands to install the skill from a remote GitHub repository.
The installation instructions use npx skills add https://github.com/leoluyi/skills.
This reference targets the author's own repository on a well-known service, which is a standard and safe practice.

breakdown