The Agent Skills Directory

[SAFE]: The skill performs targeted codebase scans using patterns defined in its reference documentation (e.g., searching for auth.ex, config/runtime.exs, and Dockerfile). These scans are used to ground the interview questions in the project's actual architecture and do not involve exfiltration of sensitive data.
[SAFE]: The skill incorporates an external research phase that utilizes a platform-provided web research agent. This process is designed to find community best practices and library options, sharing only a high-level requirements summary with the research tool.
[SAFE]: The skill manages project state by writing artifacts (such as interview.md and research summaries) to a local .claude/plans/ directory. This allows for persistent context across development sessions without using malicious persistence mechanisms.
[SAFE]: The instructions explicitly mandate human-in-the-loop decision-making, requiring the agent to use the AskUserQuestion tool at key checkpoints before proceeding with research or moving to the planning phase, ensuring the user maintains control over the agent's actions.

brainstorm