The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes codebase files to generate documentation, which creates a surface where malicious instructions embedded in code files could attempt to influence the agent's research summary.
Ingestion points: Files in the project root and subdirectories are read via the Read, Glob, and Grep tools as part of the research workflow defined in scripts/deep_research.sh.
Boundary markers: The skill relies on natural language instructions for subagents to provide summaries and does not use specific data delimiters or ignore-instructions for the processed content.
Capability inventory: The skill utilizes the Bash tool to run the claude CLI and a local Python script for managing .roomodes configuration.
Sanitization: External codebase content is processed without explicit sanitization steps.
[COMMAND_EXECUTION]: The automation script scripts/deep_research.sh executes shell commands and an inline Python script to manage project metadata and invoke research agents. These actions are consistent with the skill's purpose for project automation and documentation.
[DYNAMIC_EXECUTION]: The research workflow involves dynamically generating JSON-based subagent definitions at runtime. These configurations are tailored to the detected project structure, allowing for specialized analysis of different directory types.

codebase-researcher