pytest-optimizer-02-planlisted
Install: claude install-skill tony/ai-workflow-plugins
# 02-plan
Turn measured candidates into a reviewable commit plan. This phase **only reads**
the suite and writes `plan.json`; it makes no code changes.
`$ARGUMENTS` may pass `--max-commits=N` to cap the plan, `--min-score` to raise the
inclusion threshold, and `--force` to recompute.
## Step 1: Load and score
Read `benchmarks.json` and `baseline.json`. Score every `validated` candidate with
`references/scoring-rubric.md`:
```
total = 0.35*safety + 0.30*impact + 0.15*effort
+ 0.12*confidence + 0.08*reversibility
```
Apply the **hard gates** first: drop any candidate with `safety < 0.4` or
`impact == 0`. For each dropped item, capture the prerequisite refactor (if any)
as a separate, clearly-labeled follow-up — not as an auto-applied commit.
## Step 2: Order
Score sets priority; these constraints override it where they apply:
1. Safety-gate fixes first (make collection deterministic, prove order
independence) before any parallel/reorder speedup.
2. Typing and parametrize migrations early (low-risk, readable diffs).
3. Scope and consolidation before parallelism.
4. One speedup per commit — never bundle.
Honor `--max-commits` / `--min-score`.
## Step 3: Draft commits
For each planned item, draft a commit from
`templates/commit-message.tmpl`, adapting the
`type(scope)` prefix to the **target project's** convention (read from its
`AGENTS.md`/`CLAUDE.md`). Each entry records: order, heuristic id, the score
breakdown, target files, the draft subject/body, the ver