/optimize

name: optimize
description: Minimal codex-native optimization loop. Use for metric-driven improvements with guardrails and measurable gates.

Optimize

Metric-driven optimization with explicit guards, rollback criteria, experiment log.

Input Schema

{
  "goal": "required measurable improvement objective",
  "mode": "single|campaign",
  "metric_cmd": "required command that emits or validates the target metric",
  "metric_direction": "higher|lower",
  "guard_cmd": "required command that must continue to pass",
  "max_iterations": "optional integer, default 1",
  "min_delta": "optional practical significance threshold",
  "scope_files": [
    "paths the optimization may edit"
  ],
  "done_when": "metric improves without guard regression"
}

Workflow

01: Create run directory

Run `python PLUGIN_ROOT/shared/create_run.py --skill optimize` once. Retain its single printed path as `<run-directory>` and substitute that literal path into every later artifact path and helper argument. Never store or reuse the path through a shell variable; shell variables do not persist across tool calls.

02: Validate metric and guard commands

Require:

• Repeatable `metric_cmd` producing comparable value or pass/fail.
• Known `metric_direction`.
• `guard_cmd` fails on unacceptable regressions.
• Bounded `scope_files`.
• Explicit, bounded `max_iterations` for `campaign`.
• Protect files/scripts used by `metric_cmd`/`guard_cmd` unless user explicitly scopes them and accepts measurement-integrity risk.

Dry-run both before edit:

Execute the configured `metric_cmd` and `guard_cmd` separately with the host-native command runner. Write complete combined output to `<run-directory>/metric-baseline.txt` and `<run-directory>/guard-baseline.txt`; retain both exit codes and stop before editing if either command cannot run.

03: Record baseline and hypothesis

Write `<run-directory>/hypothesis.md`:

• metric to improve
• expected mechanism
• files allowed to change
• guard risk
• rollback condition

For `campaign`, noisy metrics, GPU/ML performance, or correctness-sensitive code, apply `../../shared/specialist-orchestration.md`. Write `<run-directory>/specialist-optimization-plan.md` with narrow context packs for:

• `squeezer`: profiling mechanism, bottleneck hypothesis, measurement plan.
• `qa-specialist`: guard coverage and regression risk.
• `data-steward`: data pipeline or reproducibility impact.
• `scientist`: metric validity, ablation design, statistical noise.
• `challenger`: overfitting to the metric or weakening guard checks.

No fan-out for one small measured change with stable metric/guard. Never let specialist change metric/guard scripts unless explicitly in `scope_files` and measurement-integrity risk recorded.

**Structural context (optional)**: when `scope_files` resolves to a Python module/symbol, also probe codemap-py once for callers, coupling, and test impact before the first iteration: `python PLUGIN_ROOT/shared/codemap_adapter.py context --category develop --target <qname> --out <run-directory>/codemap-context.json`. Per `../../shared/codemap-contract.md`, absence/incompatibility is non-fatal — continue with the hypothesis above. Persist the result once here, before step 04 applies any change; any triggered specialist consumes `<run-directory>/codemap-context.json`, never a fresh query.

Initialize machine-readable iteration log:

Create an empty `<run-directory>/experiments.jsonl` with the filesystem tool before the first iteration.

04: Apply one minimal optimization change per iteration

One independent hypothesis per iteration. Do not optimize unmeasured paths. Before each, write `<run-directory>/iteration-<n>-before.patch` with scoped-file diff. If iteration fails and only its patch is present, revert with `git apply -R` against iteration diff; otherwise fail run when clean reversal cannot be proven. Never use `git reset --hard`.

05: Re-measure

Re-run the same retained `metric_cmd` and `guard_cmd` separately with the host-native command runner. Write complete combined output to `<run-directory>/metric-after.txt` and `<run-directory>/guard-after.txt`; retain both exit codes.

06: Compare baseline and after results in `<run-directory>/comparison.md`

Required fields:

• baseline value
• after value
• delta
• guard status
• confidence
• noise caveats

Append one JSON object/iteration to `<run-directory>/experiments.jsonl`:

{
  "iteration": 1,
  "hypothesis": "one-line mechanism",
  "metric_before": 0.0,
  "metric_after": 0.0,
  "delta": 0.0,
  "guard": "pass|fail",
  "decision": "kept|reverted|inconclusive|failed",
  "rollback_evidence": "path or reason"
}

07: Decide keep/revert

• Keep only with intended metric movement and passing guards.
• With `min_delta`, keep only if delta meets/exceeds practical-significance threshold.
• Revert or fail on guard regression.
• For noisy measurement, repeat or mark inconclusive.
• In `campaign`, stop at first kept result unless user asked continued exploration; otherwise continue only while `max_iterations` remains and each rejected iteration has rollback evidence.

08: Run shared quality gates

Inspect `python PLUGIN_ROOT/shared/run_gates.py --help`. Tests runs configured test or guard command; give real commands or explicit reasons for other gates.

09: Write and validate the mandatory result artifact

Follow `../../shared/helper-cli-contract.md` and authoritative help. Write `OPTIMIZE_METADATA`, validate as `optimize`, promote only validated candidate.

Fail-Fast Rules

1. Missing metric or guard command => fail. 2. Baseline cannot be captured => fail. 3. Scope is unbounded => fail. 4. Guard regression after change => fail unless reverted. 5. Metric/guard script changed without explicit scope and measurement-integrity note => fail. 6. Campaign iteration rejected without rollback evidence or unresolved-risk note => fail. 7. Claimed improvement below `min_delta` without explicit