# AI use cases for Jev Harness (TianyuCodings/JevHarness)

Repo: [TianyuCodings/JevHarness](https://github.com/TianyuCodings/JevHarness) | Use cases: 4 | Page: https://dedupedai.com/use-cases/tianyucodings-jevharness/

JevHarness is a Python project where an LLM writes a task-specific harness, meaning code and Jev judgment questions that turn task observations into actions, with optional reward-based reflection to improve it.

This file holds every use case of this repo. Each one starts with a "## " heading and ends with its URL.

- Draft a task-specific harness in Claude Code: https://dedupedai.com/use-cases/tianyucodings-jevharness/draft-task-harness-in-claude-code/
- Inspect the Pokémon harness archive locally: https://dedupedai.com/use-cases/tianyucodings-jevharness/inspect-pokemon-archive-locally/
- Install the Jev Harness skill in Codex: https://dedupedai.com/use-cases/tianyucodings-jevharness/install-jev-harness-skill-in-codex/
- Improve a harness with reliable rewards and reflection: https://dedupedai.com/use-cases/tianyucodings-jevharness/optimize-harness-with-reliable-rewards/

## How to draft a task-specific harness in Claude Code with Jev Harness

Repo: [TianyuCodings/JevHarness](https://github.com/TianyuCodings/JevHarness) | Difficulty: Intermediate | Time: 1 hour | Basis: From the README

Install the Jev Harness plugin and have Claude Code draft a Jev decision harness for your task.

**Who it is for:** A developer on a support or operations team who needs incoming items routed to the right team and uses Claude Code.

**The problem:** Writing decision logic and Jev instructions and criteria by hand takes time and expertise.

**The outcome:** A clarified plan for your inputs, legal actions, examples, success criteria, and budget, ready for the authoring LLM to turn into a harness.

### Steps

1. Run `/plugin marketplace add https://github.com/TianyuCodings/JevHarness.git` inside Claude Code.
2. Run `/plugin install jev-harness@jevharness`.
3. Run `/reload-plugins`.
4. Run `/jev-harness:jev-harness` with your task, and answer its questions about inputs, legal actions, examples, success criteria, and budget.

### Prompt to paste into your agent

```text
Using the JevHarness repo at https://github.com/TianyuCodings/JevHarness, invoke the jev-harness skill to draft a harness that routes <ticket type> to the right team. Ask for inputs, legal actions, examples, success criteria, and budget before writing any code. Expected output: a plan listing those items. Verify that every legal team action appears in the plan.
```

### Caveat

Installing the plugin does not install Python dependencies or configure model keys, so Jev access must be set up separately.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/tianyucodings-jevharness/draft-task-harness-in-claude-code/

## How to inspect the Pokémon harness archive locally with Jev Harness

Repo: [TianyuCodings/JevHarness](https://github.com/TianyuCodings/JevHarness) | Difficulty: Intermediate | Time: 30 minutes | Basis: From the README

Run the bundled Pokémon example viewer to review decisions without making any model requests.

**Who it is for:** A developer or researcher who wants to see how an evolved harness changed decisions before building one.

**The problem:** Reading harness code alone does not show how it behaved in actual games.

**The outcome:** A local viewer showing the archived run, the evolution tree, and complete decision records with synchronized Jev answers.

### Steps

1. Run `git clone https://github.com/TianyuCodings/JevHarness.git`.
2. Run `cd JevHarness`.
3. Run `node website/build.mjs`.
4. Run `node website/preview.mjs --port 8768`.
5. Open http://localhost:8768 in a modern browser.

### Prompt to paste into your agent

```text
Clone https://github.com/TianyuCodings/JevHarness, build and preview the website with Node.js on port 8768, then report the number of archived Eval games shown on the page. Expected output: the viewer loads at localhost:8768. Verify that browsing makes no model requests and starts no games.
```

### Caveat

Battle animations download assets from the official Pokémon Showdown renderer, so the viewer needs network access for them.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/tianyucodings-jevharness/inspect-pokemon-archive-locally/

## How to install the Jev Harness skill in Codex with Jev Harness

Repo: [TianyuCodings/JevHarness](https://github.com/TianyuCodings/JevHarness) | Difficulty: Intermediate | Time: 30 minutes | Basis: From the README

Add the jev-harness skill to Codex and use it to plan a harness for your task.

**Who it is for:** An engineer who works in Codex and wants help designing a Jev decision harness.

**The problem:** Codex does not know how to structure a Jev harness without installing the skill.

**The outcome:** A working jev-harness skill in Codex that asks for task inputs, actions, examples, and success criteria before drafting the harness.

### Steps

1. Open a Codex host that provides the built-in skill-installer.
2. Send `$skill-installer Install the jev-harness skill from https://github.com/TianyuCodings/JevHarness/tree/main/skills/jev-harness`.
3. Restart the session if the skill is not discovered.
4. Send `$jev-harness Build a harness that routes support tickets to the right team.` and answer the clarifying questions.

### Prompt to paste into your agent

```text
In Codex, install the jev-harness skill from https://github.com/TianyuCodings/JevHarness and then run it on the task: route <ticket type> to the right team. Expected output: a list of clarified inputs, legal actions, examples, and success criteria. Verify that the skill is discovered before the task is sent.
```

### Caveat

The skill helps draft the harness, but running Jev calls still requires model access that the README does not configure for you.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/tianyucodings-jevharness/install-jev-harness-skill-in-codex/

## How to improve a harness with reliable rewards and reflection with Jev Harness

Repo: [TianyuCodings/JevHarness](https://github.com/TianyuCodings/JevHarness) | Difficulty: Advanced | Time: 3 hours | Basis: From the README

Use reward-based evaluation and GEPA reflection to evolve a harness from full execution traces.

**Who it is for:** An engineer with a task that has a trusted reward signal and a training and evaluation split, who wants a harness improved from evidence.

**The problem:** A harness written once may underperform, and manual tuning without a reliable reward wastes effort.

**The outcome:** A selected harness chosen on the Eval set, with the run's ancestry, rejected proposals, and full decision traces kept for review.

### Steps

1. Install the jev-harness skill in Claude Code or Codex.
2. Describe your task, its legal actions, and the reward source, and confirm the reward is trusted.
3. Ask for evaluation and reflection optimization, for example: `If we have reliable rewards, add evaluation and reflection optimization.`
4. Review the selected harness, the rejected proposals, and the recorded ancestry before freezing the chosen harness.

### Prompt to paste into your agent

```text
Using the JevHarness skill from https://github.com/TianyuCodings/JevHarness, plan an evaluation and reflection loop for <task> with a trusted reward, a training batch, and an Eval set. Expected output: a plan naming the selection split, the reflection inputs, and how accepted proposals are checked on Eval. Verify that the plan keeps Eval separate from the reflection inputs.
```

### Caveat

The reported Pokémon gain is an example result on the selection Eval set, not an independent estimate for unseen games.

Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/

URL: https://dedupedai.com/use-cases/tianyucodings-jevharness/optimize-harness-with-reliable-rewards/