The Agent Skills Directory

[SAFE]: The skill implements a structured codebase analysis workflow using parallel subagents for exploration and a synthesis agent for explanation. This design follows principle of least privilege by using read-only subagents for the exploration phase.
[SAFE]: The skill explicitly instructs agents to avoid file modifications, focusing purely on prose output ("The explainer only reads and writes prose. Tell it in the prompt not to edit files").
[SAFE]: The architecture utilizes multiple reasoning models (such as Fable, Opus, and Sonnet) for its critique mode to provide diverse perspectives, which is a standard pattern for improving analysis quality without introducing security risks.
[SAFE]: All tool usage (Glob, Grep, Read) is restricted to local codebase analysis within the agent's provided environment. No external network exfiltration or suspicious third-party dependencies were identified.
[INDIRECT_PROMPT_INJECTION]: The skill has an inherent attack surface as it processes codebase content (files, directory structures) which could contain adversarial instructions. However, the risk is mitigated by the read-only nature of the explorers and the synthesis-focused prompts which reduce the impact of potential injection in the analyzed code.

how