# AI use cases for Ant Omni Evo (ant-research/AntOmniEvo)

Repo: [ant-research/AntOmniEvo](https://github.com/ant-research/AntOmniEvo) | Use cases: 4 | Page: https://dedupedai.com/use-cases/ant-research-antomnievo/

AntOmniEvo is a Python framework that evolves a system's tunable files (skills, prompts, workflow configs, or single-file algorithms) through a loop where AI coding agents propose edits scored by your own evaluator.

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

- Evolve an agent skill directory against your eval set: https://dedupedai.com/use-cases/ant-research-antomnievo/evolve-agent-skill-directory/
- Optimize a retrieval workflow's pipeline.json and node code: https://dedupedai.com/use-cases/ant-research-antomnievo/optimize-workflow-pipeline-config/
- Optimize a single-file algorithm and its description: https://dedupedai.com/use-cases/ant-research-antomnievo/optimize-single-file-algorithm/
- Set up an experiment with the bundled agent skill template: https://dedupedai.com/use-cases/ant-research-antomnievo/start-experiment-with-agent-skill-template/

## How to evolve an agent skill directory against your eval set with Ant Omni Evo

Repo: [ant-research/AntOmniEvo](https://github.com/ant-research/AntOmniEvo) | Difficulty: Advanced | Time: 1 day | Basis: From the README

Let agents rewrite SKILL.md, references and scripts until an evaluator scores the agent higher.

**Who it is for:** An engineer maintaining an AI agent whose skill files underperform on a known set of test tasks and who wants the files improved without hand-editing each failure.

**The problem:** Skill files (SKILL.md, references, scripts) are tuned by hand, and each fix needs a person to read failures and guess which file to change.

**The outcome:** A directory of optimized skill files with a change lineage that ties each edit to the failure evidence behind it.

### Steps

1. Write a System that runs your agent on one eval instance and returns its output.
2. Write an Evaluator that scores each output from 0 to 1; its scoring criteria become the optimization objective.
3. Prepare train and validation instances that define what good means for your agent.
4. Describe the skill directory as a TunableArtifactSchema: the file tree and what each file is for.
5. Provide the starting skill files as initial tunable artifacts and start the evolution loop so the Proposer edits files from failure trajectories.
6. Review the best candidate's directory and its change lineage before deploying the files.

### Prompt to paste into your agent

```text
Using the AntOmniEvo repo (ant-research/AntOmniEvo), set up an evolution run for the skill directory at <skill-dir> with an Evaluator that scores outputs from 0 to 1. Expected output: the best candidate's tunable artifacts as a directory plus its change lineage. Verify that the CandidateStore holds every candidate, run, analysis, and changelog.
```

### Caveat

The README describes the framework and its docs but this excerpt shows no install commands, so check docs/quickstart.md for setup details before planning the run.

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/ant-research-antomnievo/evolve-agent-skill-directory/

## How to optimize a retrieval workflow's pipeline.json and node code with Ant Omni Evo

Repo: [ant-research/AntOmniEvo](https://github.com/ant-research/AntOmniEvo) | Difficulty: Advanced | Time: 1 day | Basis: From the README

Have agents edit pipeline config and node scripts until a workflow scores better on your data.

**Who it is for:** A developer running a multi-step pipeline, such as a retrieval DAG, whose result quality varies and who has a repeatable way to score outputs.

**The problem:** Tuning a pipeline means editing a config file and several node scripts together, and it is hard to tell which change caused a gain or loss.

**The outcome:** An optimized pipeline.json and set of node scripts, with the lineage showing which edits came from which failures.

### Steps

1. Define the System as the function that runs the pipeline on one instance and returns its result.
2. Write an Evaluator that scores each pipeline result from 0 to 1.
3. Gather train and validation queries that reflect the real workload.
4. Map pipeline.json and the nodes/*.py files into a TunableArtifactSchema so the Proposer knows which file does what.
5. Start the evolution loop from the current pipeline files and let the framework keep the strongest candidates.
6. Compare the best candidate's files with the originals and deploy only the changes you accept.

### Prompt to paste into your agent

```text
Using the AntOmniEvo repo (ant-research/AntOmniEvo), evolve the workflow at <pipeline-dir> where pipeline.json and nodes/*.py are the tunable artifacts, scored by an Evaluator returning 0 to 1. Expected output: the best candidate's pipeline directory and changelog. Verify that each changed file in the best candidate traces back to a recorded failure.
```

### Caveat

The README gives the pipeline.json and nodes/*.py example, but it does not show how a retrieval DAG is wired in, so confirm the integration details in the docs before relying on it.

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/ant-research-antomnievo/optimize-workflow-pipeline-config/

## How to optimize a single-file algorithm and its description with Ant Omni Evo

Repo: [ant-research/AntOmniEvo](https://github.com/ant-research/AntOmniEvo) | Difficulty: Intermediate | Time: 3 hours | Basis: From the README

Let agents rewrite one .py or .ts file and its description until scores improve.

**Who it is for:** A developer with one algorithm file and a clear way to measure its output who wants it improved without manual trial and error.

**The problem:** Manual rounds of editing one algorithm file and rerunning tests are slow, and gains are hard to attribute.

**The outcome:** A better version of the algorithm file and its description, with a record of each edit and the evidence for it.

### Steps

1. Write a System that runs the algorithm on one eval instance and returns its result.
2. Write an Evaluator that scores each result from 0 to 1.
3. Prepare train and validation instances that reflect the target task.
4. Map the algorithm file and its description into a TunableArtifactSchema as a small two-file directory.
5. Start the evolution loop from the current file and review the best candidate when the run ends.

### Prompt to paste into your agent

```text
Using the AntOmniEvo repo (ant-research/AntOmniEvo), evolve the single-file algorithm at <file-path> and its description, scored by an Evaluator returning 0 to 1. Expected output: the best candidate's file and description plus its change lineage. Verify that the best candidate's score on the validation instances beats the starting file.
```

### Caveat

The README says the system under optimization need not contain an LLM, but the Proposer must be agents, so plan for agent access even for a simple algorithm.

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/ant-research-antomnievo/optimize-single-file-algorithm/

## How to set up an experiment with the bundled agent skill template with Ant Omni Evo

Repo: [ant-research/AntOmniEvo](https://github.com/ant-research/AntOmniEvo) | Difficulty: Intermediate | Time: 3 hours | Basis: From the README

Use the skills/ template so a coding agent interviews you and drives the experiment setup.

**Who it is for:** A team that wants a coding agent to handle the setup of an Ant Omni Evo experiment and to triage failures, starting from an existing example skill.

**The problem:** Setting up components, the entry script, and the launch, then diagnosing failures, takes repeated manual work for each new scenario.

**The outcome:** A launched experiment with the entry script written by the agent, plus a list of triaged failures checked against the pitfall log.

### Steps

1. Open the skills folder in the repo and read the eddy-antomnievo-experiment skill as a template.
2. Give that skill to your coding agent so it can interview you for the repo, dataset and metric.
3. Let the agent pick components, write the entry script, and launch the run.
4. Check the agent's failure triage against the built-in pitfall log.
5. Adapt the skill for your own scenario by distilling your best practices into a new skill.

### Prompt to paste into your agent

```text
Using the AntOmniEvo repo (ant-research/AntOmniEvo), hand its skills/eddy-antomnievo-experiment skill to the coding agent and set up an experiment for <target-repo> with dataset <dataset> and metric <metric>. Expected output: a written entry script and a launched run. Verify that the agent asked for the repo, dataset and metric before writing any files.
```

### Caveat

The bundled skill's content is in Chinese, as the README notes, so an English-speaking team may need to read it with translation help.

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/ant-research-antomnievo/start-experiment-with-agent-skill-template/