# Metaharness Optimize

> Run metaharness from an Omnigent agent to optimize harness files in the current workspace. Use when a host wants the metaharness kernel without wrapping a proprietary proposer.

- Skill: `superagenticai/metaharness-optimize` (Agent Skill)
- Install (CLI): `npx skillmds@latest add superagenticai/metaharness-optimize`
- Raw SKILL.md: https://api.skillmd.com/api/skills/superagenticai/metaharness-optimize/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: SuperagenticAI (https://skillmd.com/u/superagenticai)
- Updated: 2026-09-10
- Page: https://skillmd.com/skills/superagenticai/metaharness-optimize

---


# Optimize With metaharness

Call the packaged CLI from the host agent working directory. The kernel is
`optimize_harness`; this skill is only a wrapper around `metaharness run`.

```bash
metaharness run . --backend fake --budget 1
```

For a live proposer, pass `--backend codex` or `--backend omnigent` according
to the project's `metaharness.json`. Do not add an AgentSky or HarnessRouter
proposer backend; those hosts should embed the same Python or UHP contract.

Expected artifacts after a run:

- `runs/<name>/` with candidate workspaces and diffs
- `best_candidate_id` and `best_objective` on stdout
- ledger via `metaharness ledger runs/<name>`