The Agent Skills Directory

[SAFE]: The skill uses readonly: true for all subagent spawns, ensuring that the agents cannot modify the codebase during exploration or explanation phases. It focuses on using standard tools like Glob, Grep, and Read for architectural analysis.
[INDIRECT_PROMPT_INJECTION]: The skill processes user questions and codebase content. While this creates an ingestion surface for untrusted data (e.g., malicious comments in the code), the skill employs structural boundaries (explainer/explorer templates) and operates in a readonly environment, which effectively mitigates common injection risks such as file modification or unauthorized command execution.

how