Retrospective

Retrospective

Analyze evaluation patterns across completed plans, identify recurring failures, and auto-apply checklist evolution — the downstream consumer of executing-plans Phase 4 "Checklist Evolution Candidates". The user reviews post-commit via git show docs/retros/checklists/.

Recommended: run wrapped in /goal

Launch it under Claude Code's built-in /goal (v2.1.139+) — a retrospective can span multiple turns:

/goal "Claude has narrated a successful checklist-evolution commit (with commit hash) and stated the retrospective is complete" /superpowers:retrospective <plan-paths>

/goal is a user-typed outer wrapper (a skill cannot enable it for itself mid-run), and its evaluator judges only what Claude narrates in the transcript — phrase the condition against narrated output (the commit-hash line, an explicit completion statement), never filesystem state. Full semantics and condition phrasing: ../../skills/references/goal-wrapper.md.

Pre-Check (run first, in order)

A. INSUFFICIENT-POST-PLAN advisory (informational)

Read the most recent plan_completed event from docs/retros/plans-completed.jsonl. If hours_since_completion < 24h AND bash "${CLAUDE_PLUGIN_ROOT}/lib/post-plan-diff.sh" summary <completion_commit> <files...> returns total == 0, output the INSUFFICIENT-POST-PLAN reminder verbatim (see ./references/post-plan-diff.md §Pre-Check A) and proceed to Phase 0 — do NOT pause. Skip silently when completion_commit is missing.

B. Recall persistent memory (calibration priors)

CRITICAL — do this before Phase 0, do NOT skip it as setup. Persistent memory holds the human judgments the evolution-log cannot capture. Recall is a SECONDARY calibration signal — docs/retros/evolution-log.jsonl (Phase 1 step 5) stays authoritative; when they disagree, the log wins.

The MEMORY.md index hooks are ALREADY in your context (injected at session start) — read nothing from disk. Scan them now and select the ones bearing on this run: prior evolution/rejection decisions and debt trackers, harness-design principles, and working-style feedback. Carry them forward as priors into Phase 1 step 5 (calibration history), Phase 3 (suppression of weakly-justified ADDs), and Phase 4 (self-reject a proposal that contradicts a recalled prior, citing the memory entry). If no calibration-relevant hook exists, log Pre-Check B: no calibration-relevant memory and proceed. Never resolve a memory path, read a topic file, block, or ask — the index hooks carry the signal on their own.

When a recalled hook cited as evidence for an approved Phase 3 proposal proves project-specific and durable, the Phase 4 step 3.5 draft records Promoted from private assistant memory hook: <hook-name>, <date> in its ## Why section — cross-project harness-design stances are NOT promoted, and the private hook is never deleted or modified.

Phase 0: Bootstrap (run only when no checklists exist)

Before Phase 1, check whether docs/retros/checklists/ contains {mode}-v1.md for each mode (design / plan / code). Phase 0 runs per-mode independently — seed only the modes missing a v{N} file; if all three are present, log Phase 0: all checklists present, skipping seed and proceed. Full procedure (Path A vs. Path B Full History Bootstrap, exit codes, --force reset): ./references/bootstrap.md.

Phase 1: Data Collection

1. Resolve inputs: Parse $ARGUMENTS for plan paths. If --across-all, scan docs/plans/ for all *-plan/ directories with evaluation reports. If no argument is given, read docs/retros/plans-completed.jsonl and auto-scope to plans completed after the most recent retrospective_run event in docs/retros/evolution-log.jsonl. Complement the log via the docs index: bash "${CLAUDE_PLUGIN_ROOT}/lib/docs-index.sh" list --kind plan --status implemented scopes shipped plans; bash "${CLAUDE_PLUGIN_ROOT}/lib/docs-index.sh" list --status expired surfaces prior expirations as calibration input (read an expired plan's retro report before re-proposing anything that touches it). Also consult memory: bash "${CLAUDE_PLUGIN_ROOT}/lib/docs-index.sh" list --kind memory --status active; fold matches into Phase 2 analysis as calibration input.
2. Resolve evals: For each plan path, read evaluation reports (evaluation-round-*.md, evaluation-design-round-*.md, evaluation-plan-round-*.md) from the plan directory, or from a sibling *-evals/ directory when one exists.
3. Read checklists: Scan docs/retros/checklists/ for each mode's latest {mode}-v{N}.md (highest N).
4. Read reports: Read each plan's evaluation reports; extract per-item results (Item ID, Result, Evidence) and rework items.
5. Read evolution history (calibration input): Read docs/retros/evolution-log.jsonl if present; build a history table keyed by item_id (most recent event, timestamp, rationale). This feeds Phase 3 — do NOT re-propose an ADD for an item REMOVEd in the most recent retrospective unless the new evidence is materially different; cite the historical entry in any such proposal.
6. Read post-plan diff: For each plan with a completion_commit field, run bash "${CLAUDE_PLUGIN_ROOT}/lib/post-plan-diff.sh" list <completion_commit> <completion_modified_files...> and pass the classified commit list to Phase 5a (classification rules and skip conditions: ./references/post-plan-diff.md).