planner-reflectlisted

# Planner-Reflect — post-task learning This skill compares what was planned against what actually happened, extracts lessons, and writes them back into `<project-root>/.claude/planner-context.md` so the next planning session starts smarter. It is the feedback loop that closes the gap described in feature README §35-44: plans drift from reality, gaps in the agent catalog go unnoticed, weak-model signals are forgotten, user corrections evaporate after the session ends. Run this skill explicitly via `/plan-reflect`, ideally in the same session as the work it reflects on — cross-session transcript availability is not guaranteed and the skill degrades gracefully if it is missing. The run is short and linear: gather the five evidence sources, classify findings into the four update types, mask any PII in the proposed additions, write the result to `planner-context.md`, and emit a chat summary that always includes a "Lessons learned" block. There is no looping, no escalation, and no chaining into other skills — reflection is observation, not action. ## Inputs (5 evidence sources) The skill reads exactly five sources, in this order. Each source is independent; missing one weakens the conclusions but does not block the run. 1. **`<feature-dir>/PLANNER_OUTPUT.md`** — the original plan: phases, agents chosen, model picks, cost estimate. Read with `Read`. This is the ground truth for "what was planned"; if it is missing, the skill exits (see next section). 2. **`git log --oneline <si