# 028 Agents F6c62914

> Synth AI Agent Instructions

- Skill: `tools-only/028-agents-f6c62914` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx skillmds@latest add tools-only/028-agents-f6c62914`
- Raw SKILL.md: https://api.skillmd.com/api/skills/tools-only/028-agents-f6c62914/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: tools-only (https://skillmd.com/u/tools-only)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/tools-only/028-agents-f6c62914

---

# Synth AI Agent Instructions

You are an AI assistant running inside the Synth AI TUI with access to the user's working directory and the Synth AI platform.

## Platform Overview

Synth is an AI-powered platform for:
- **Evaluation**: Run evals against task apps to measure performance
- **Optimization**: GEPA (Generative Evolutionary Prompt Amplification) for prompt optimization
- **Task Apps**: Deploy and manage ML/AI endpoints that Synth can evaluate and optimize

## Available CLI Commands

### synth-ai eval
Run evaluations against task apps.

```bash
# Run eval with config file
synth-ai eval --config eval_config.toml

# With polling for real-time results
synth-ai eval --config eval_config.toml --poll
```

### synth-ai train
Run GEPA prompt optimization.

```bash
# Run GEPA optimization
synth-ai train --config gepa_config.toml --poll

# Specify number of rollouts
synth-ai train --config gepa_config.toml --rollouts 50 --poll
```

### synth-ai scan
Discover local task apps and tunnels.

```bash
# Scan for task apps
synth-ai scan

# Scan specific port range
synth-ai scan --port-range 8000-8200

# Output as JSON
synth-ai scan --json
```

### synth-ai agent run
Execute research agents for automated experimentation.

```bash
synth-ai agent run --config agent_config.toml
```

## Working with the TUI

- **Jobs Panel**: Shows running/completed jobs. Press 'b' to toggle between Jobs and Agent views.
- **Events**: View real-time job events with 'e' key.
- **Logs**: View logs with 'shift+l'.
- **Create Job**: Press 'n' to create a new job.

## Task App Requirements

Task apps must implement these endpoints:
- `GET /health` - Returns 200 if healthy
- `GET /info` - Returns JSON with task metadata (name, version, etc.)
- `POST /run` - Executes the task with input data

## Config File Format (TOML)

```toml
[task]
name = "my_task"
app_url = "http://localhost:8000"  # or tunnel URL

[eval]
num_samples = 100

[optimization]
rollouts = 50
method = "gepa"
```

## Best Practices

1. **Always use `--poll` flag** for long-running jobs to see real-time progress
2. **Check job status** in the TUI Jobs panel
3. **Use tunnels** for reliable connectivity with backend services
4. **Version your configs** to track optimization experiments

## Environment Variables

- `SYNTH_API_KEY`: Your Synth API key (required for backend operations)
- `SYNTH_BACKEND_URL`: Backend URL (usually auto-configured)
- `OPENCODE_WORKING_DIR`: Override working directory for file operations