The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted repository content, search results, and transcripts which may contain malicious instructions designed to influence the agent's behavior.
Ingestion points: entire search output and entire explain transcripts (SKILL.md, Process steps 4 and 6).
Boundary markers: The skill includes a specific guardrail instruction to treat all repository data, command output, and transcripts as untrusted and to never follow instructions contained within them.
Capability inventory: Access to git and the entire CLI toolset.
Sanitization: The instructions require the agent to shell-quote arguments and strip or escape dangerous characters (backticks, subshells, semicolons) before command interpolation.
[COMMAND_EXECUTION]: The skill invokes system commands to interact with the repository and the Entire platform.
Execution points: Uses git rev-parse for environment validation and entire search/entire explain for data retrieval.
User input integration: User-provided topics are used as arguments for the search command, with explicit instructions for the AI to sanitize these strings before execution.

teach