What problem does it solve?
When a long AI session receives a context_management_hint warning, agents often react reflexively—handing off mid-task and losing in-flight work, or ignoring the warning until auto-seal. This Skill provides a structured judgment framework for deciding whether to hand off, sprint to a clean breakpoint, or simply continue and let compression happen.
Core Features & Use Cases
- Three-Question Self-Check: Evaluates whether the session is a single thread or topic drift, whether a clean breakpoint exists, and how trustworthy the fill-confidence metric is.
- 2x2 Decision Matrix: Maps drift state against breakpoint availability to one of four actions: handoff, sprint mode, continue, or compress.
- Handoff Execution Guidance: Directs the agent to call cat_cafe_propose_session_handoff with a five-part handoff note covering completed work, current state, next steps, key decisions, and pitfalls.
- Use Case: An agent processing multiple GitHub issues across repos starts confusing tracker repo with source repo and receives a context warning; the matrix identifies tree-like drift mid-task and triggers sprint mode to reach a clean breakpoint before handing off.
Quick Start
When you receive a context_management_hint warning, apply the three-question self-check and 2x2 matrix to decide whether to hand off, sprint, or continue.
