backport-review-fix-into-plan-of-recordlisted

# Backport a Review Fix into the Plan-of-Record ## Overview In subagent-driven development the plan-of-record is executable: a task's implementer prompt reproduces the authored code verbatim. When a per-task review then catches a real bug and you fix it in the working tree, the *plan* still contains the buggy version. Any later re-run of that task — a resumed session, a re-dispatch after a limit reset, a second pass over the same plan — will faithfully re-author the original bug over your corrected tree. The fix is only durable once it lives in the plan, not just in the commit. ## When to Use - A per-task review (or `verify-subagent-completion`) catches and you fix a genuine bug in code an execution plan authored. - The plan-of-record still holds executable code / verbatim snippets the implementer replays. - The plan or its remaining tasks may be re-run, resumed, or handed to another session. ## When NOT to Use - The plan is purely narrative ("implement auth middleware") with no reproducible code snippet — there's nothing to reinstate. - One-shot work with no possibility of a re-run of that task. ## Steps 1. **Fix the bug in the working tree first** and confirm it (test passes, behaviour correct) per normal review flow. 2. **Locate the same code in the plan-of-record** — the task step whose implementer prompt authored the buggy line. 3. **Backport the exact fix into the plan** so the plan's snippet now matches the corrected tree. Fix it where the defect was authored,