The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the user's local codebase, including README files, manifest files, and source code comments, to populate its documentation templates. \n
Ingestion points: The scripts/scan.py script reads previews of manifest files and searches for TODO/FIXME comments. The agent is further instructed to read PRD, TRD, and SPEC files. \n
Boundary markers: Instructions explicitly direct the agent to 'Only document what is verifiable from files or terminal output — never infer or assume' and to mark unknowns as [TODO]. \n
Capability inventory: The skill can execute bundled Python scripts, run git commands via a subprocess, and write documentation files to the local docs/codebase/ directory. \n
Sanitization: No explicit content sanitization logic is present for data read from the codebase. \n
[COMMAND_EXECUTION]: The scripts/scan.py utility executes system commands using the subprocess module to gather repository metadata. \n
Evidence: The script calls git log and git rev-parse to retrieve commit history, file churn data, and repository status. These calls use hardcoded argument lists, which is a secure implementation that minimizes command injection risks. \n
[DATA_EXFILTRATION]: The skill is designed to discover and document project configurations, which involves accessing potentially sensitive file paths. \n
Evidence: scripts/scan.py searches for and reads previews of manifest files (e.g., package.json, pyproject.toml) and environment variable templates (e.g., .env.example, .env.template). This behavior is strictly for local documentation purposes and no external data transmission patterns were found.

acquire-codebase-knowledge