planner-reflect

To extract lessons learned from a planning execution, compare planned outcomes with actual results using multiple evidence sources. This updates your planner context with gap-fill, model-strength signals, user-correction patterns, and cost-calibration deltas to improve future AI agent workflows.

A planning loop retrospective aligns plans with reality by comparing planned outcomes against actual session results. It extracts actionable insights from five evidence sources and writes them into planner-context.md to continuously calibrate and improve subsequent planning sessions.

You update planner-context.md by running a retrospective process after a session that produced PLANNER_OUTPUT.md. The system reads evidence sources, extracts four learning types, and automatically writes the actionable lessons into the file for future reference.

Yes, you need a PLANNER_OUTPUT.md file to run a retrospective. The process specifically applies to Claude-driven project workflows where this file exists, serving as the baseline to compare against actual execution results and generate lessons learned.

When a planning retrospective finds no new lessons learned, the system always emits a designated "Lessons learned" block anyway. This ensures retrospective visibility is maintained across sessions, confirming the evidence sources were reviewed even if no actionable insights were extracted.

Yes, you can use planning context updates for cross-session project retrospectives. The process guides the update of planner-context.md with historical learning types, ensuring that insights from previous planning loops inform and calibrate subsequent planning cycles.