Pytest Optimizer 03 Execute

03-execute

Land the plan as separate, verified commits. This is the only phase that edits the suite and writes history.

$ARGUMENTS may pass --from=<index> to start at a specific plan item, --dry-run to apply and verify without committing (reverting after each), and --memory-dir to override the memory location.

Step 1: Preconditions

Read plan.json, baseline.json, and the project quality-check + test commands. Refuse to run unless:

• the working tree is clean (one commit per speedup needs a clean base), and
• plan.json exists and was approved in the pytest-optimizer-02-plan skill.

Read execution-log.json if present and resume from last_applied_index + 1 (or --from). Never re-apply an item that already has a commit_sha.

Step 2: Apply each item in order

For each plan item not yet applied:

1. Apply the single change (the heuristic's action). Touch only the files the plan lists for this item.

2. Verify green. Run the project quality checks (formatter, linter, type checker) and the test suite, as declared in the project's AGENTS.md/ CLAUDE.md — substitute that command wherever this file writes pytest:

   pytest -p no:cacheprovider -q
   

   For a typing/parametrize item (goals 9, 10), the type checker must cover tests/ and pass.

3. On success — commit this change alone with the drafted why/what message (templates/commit-message.tmpl, adapted to the target project's convention). Record the commit_sha.

4. On failure — revert the change (git checkout -- <files> / discard), mark the item skipped with the reason. Do not let one failure block the rest.

5. Checkpoint — append the item's outcome to execution-log.json immediately and update state.json (phase=execute, last_applied_index). This makes the loop crash-safe: a re-run continues from the last durable checkpoint.

Under --dry-run, perform 1–2 then revert and log the verify result without committing.

Step 3: Re-measure and report

After the plan is exhausted, re-run the baseline measurement (serial, cache disabled, N runs) and compare total wall-time to baseline.json. Emit the 03-execute sections from references/output-contract.md: hero block, then ## Applied (commit SHAs + measured verify), ## Skipped (revert reasons), ## Result (post-suite wall-time vs baseline, total recovered).

Step 4: Offer another pass

Close with an ask-user-choice panel: re-scan for a second pass (the suite is now leaner, so new candidates may surface), scaffold the opt-in cache/timing plugin, or stop. Re-scanning restarts the loop at the pytest-optimizer-00-scan skill with a fresh baseline.

Portability notes

• ask-user-choice — present the listed options and wait for the user to pick one. Hosts with a structured multiple-choice tool (Claude Code's AskUserQuestion) should use it; otherwise print a numbered list and wait for a numbered reply. Never proceed on an assumed answer.
• $ARGUMENTS — the text the user passed when invoking this skill. If your host does not substitute it, read it as the user's request in the current turn, and ask when there is none.
• Bundled files — every relative path in this skill points at a file shipped inside this skill directory. Read them from here, not from the host's plugin tree.