The Agent Skills Directory

[Data Management]: The skill manages artifacts within the .copilot-tracking/ directory. This is standard behavior for maintaining project context and state across sessions.\n- [Content Sanitization]: The implementation of a "Sanitize the content first" procedure in the handoff process ensures that internal coaching metadata and calibration data are removed before information is shared with downstream research phases.\n- [Structured Handoff Contract]: The use of a formal handoff contract with confidence markers (validated, assumed, unknown, conflicting) promotes data integrity and allows downstream agents to assess the reliability of inherited information.\n- [Indirect Prompt Injection Surface]: Like most skills that process user-generated inputs (such as interview notes and research data), there is a potential surface for indirect prompt injection. The skill mitigates this by using structured YAML schemas and explicit sanitization steps. Ingestion points: reads files from .copilot-tracking/dt/{project-slug}/ (e.g., coaching-state.md). Boundary markers: uses structured YAML for handoff artifacts. Capability inventory: file reads/writes and subagent dispatch for assessment and compilation. Sanitization: mandates stripping internal metadata and calibration data before handoff.

dt-rpi-integration