The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes Python's subprocess module in scripts/plan-chunks.py and scripts/validate-bloat-output.py to execute local commands. These commands are limited to git ls-files for document discovery and calling the internal doclifecycle engine via sys.executable. These are legitimate, scoped operations necessary for the skill's core functionality.- [PROMPT_INJECTION]: The skill processes external repository content (markdown files), creating a surface for indirect prompt injection. * Ingestion points: Document content is read by subagents during the 'Judge each chunk' step defined in SKILL.md. * Boundary markers: The SWEEP_PROMPT in scripts/plan-chunks.py explicitly defines the document list as the subagent's 'entire scope' and provides a rigid 'Definition of done'. * Capability inventory: Subagents use tools for file reading and segmentation; the orchestrator uses subprocess for planning and validation scripts. * Sanitization: The skill relies on structured JSON output validation and mandatory evidence quotes rather than textual sanitization to mitigate adversarial content.- [SAFE]: No suspicious external network calls, hardcoded credentials, or obfuscated payloads were identified. The tool architecture prioritizes confinement, writing artifacts to temporary directories outside the repository tree, and enforces a human-in-the-loop review process before any changes are applied.

detecting-doc-bloat