The Agent Skills Directory

[SAFE]: The skill follows a well-defined process for codebase research, including establishing user preferences and autonomy levels upfront.
[SAFE]: Data handling is restricted to reading codebase files and writing research documentation to a local directory (thoughts/). No evidence of unauthorized network transmission or secret exfiltration was found.
[SAFE]: The skill uses internal sub-agents (e.g., codebase-locator, codebase-analyzer) and the context7 MCP for documentation fetching, which are standard components of the vendor's ecosystem.
[SAFE]: Indirect prompt injection risk is minimal and inherent to the tool's purpose of reading codebase files; the instructions emphasize documenting existing state rather than executing code or evaluating suggestions.

researching