overmind 0.1.46

Overmind — autonomous agent optimization through structured experimentation

Overmind

Automatically optimize your AI agent's prompts, tool definitions, model selection, and pipeline logic through structured experimentation.

Documentation: Overmind guide

What it does

Overmind runs your agent against a test dataset, traces every LLM call and tool invocation, scores the outputs, and uses a strong reasoning model to generate concrete improvements. Changes that raise the score are kept; the rest are reverted. After several rounds you get a measurably better agent — without manual prompt tweaking.

What makes Overmind different is policy-driven optimization. You define the decision rules, constraints, and expectations your agent must follow, and those policies guide every stage: evaluation criteria, test data synthesis, optimization diagnosis, and scoring.

What gets optimized

• System prompts — more precise instructions, output format enforcement
• Tool descriptions — clearer parameters, better usage guidance
• Model selection — find the right quality/cost tradeoff
• Agent logic — tool-call ordering, iteration limits, output parsing
• Policy compliance — alignment with your domain rules and constraints

Get started

This walks you through the full workflow — from installation to your first optimized agent. The whole process takes about 10 minutes.

Requirements: Python 3.10+, uv, and API keys for at least one LLM provider (OpenAI, Anthropic).

1. Install

uv tool install overmind

# or dev install
git clone https://github.com/overmind-core/overmind
cd overmind
uv tool install -e .

Using uv run instead? All commands work as uv run overmind <command> after uv sync.

2. Initialize the project

cd your-agent-project/
overmind init

This creates .overmind/ in your project root and prompts for API keys and default models. Safe to re-run anytime.

3. Register your agent

overmind agent register my-agent agents.my_agent:run

Point Overmind at the Python function it should call. The function receives an input dict and must return a dict:

def run(input_data: dict) -> dict:
    # your agent logic
    return {"response": result}

Framework-based agents (Google ADK, LangChain, CrewAI, etc.) often don't expose a simple callable. Overmind detects this and offers to auto-generate an entrypoint wrapper — no manual boilerplate needed. During registration it will also collect any API keys your agent needs.

4. Validate the entrypoint (optional)

overmind agent validate my-agent --data tests/sample.json

Runs the first case from your test data through the agent to make sure the entrypoint works end-to-end before investing time in setup. For the bundled examples, paths look like examples/<folder>/data/seed.json — see examples/README.md. Exception: contract_extractor must be registered and run from examples/contract_extractor/ (see that folder’s README).

5. Set up evaluation criteria

overmind setup my-agent
# or with seed data (JSON file or directory of *.json):
overmind setup my-agent --data data/seed.json
# or with an existing policy document:
overmind setup my-agent --policy docs/my_policy.md
# or non-interactive:
overmind setup my-agent --fast

An interactive flow that analyzes your code, defines policies, builds (or imports) a test dataset, and generates scoring criteria.

6. Optimize

overmind optimize my-agent

Iteratively runs your agent, scores outputs, diagnoses failures, and generates code improvements. Changes that raise the score are kept; the rest are reverted.

Cursor Agent Skills

If you use Cursor, this repository includes Agent Skills under .cursor/skills/ — structured playbooks the Cursor agent follows to drive Overmind workflows without you typing CLI commands manually.

Available skills

Skill (slash command)	Path	What it does
/overmind-register-agent	overmind-register-agent/SKILL.md	Registers an agent in .overmind/agents.toml — discovers the entrypoint, derives module:function, runs registration, and scaffolds env vars.
/overmind-generate-dataset	overmind-generate-dataset/SKILL.md	Generates a synthetic dataset.json via persona-driven LLM generation, with schema validation and smoke-testing.
/overmind-generate-policy-and-eval	overmind-generate-policy-and-eval/SKILL.md	Creates or repairs setup_spec/policies.md and setup_spec/eval_spec.json (input/output schemas, weights, domain rules).
/overmind-optimise-agent	overmind-optimise-agent/SKILL.md	Runs the optimization loop with the Cursor agent applying candidate edits in parallel git worktrees via overmind optimize-step.

How to use the skills

The skills replace the interactive overmind CLI flow. Run them in order inside Cursor chat:

1. Initialize (one-time, run in the terminal — not a skill):

overmind init

This sets up .overmind/ and writes your API keys to .overmind/.env.

2. Register your agent — type in Cursor chat:

/overmind-register-agent path/to/your/agent.py

The agent will ask for the agent name, detect the entrypoint function, run registration, and create a .env placeholder file for your credentials.

3. Generate a dataset — type in Cursor chat:

/overmind-generate-dataset <agent-name>

Generates dataset.json under .overmind/agents/<name>/setup_spec/. Skip this step if you already have your own test data.

4. Generate the policy and eval spec — type in Cursor chat:

/overmind-generate-policy-and-eval <agent-name>

Produces policies.md and eval_spec.json through an interactive elicitation of domain rules, constraints, and scoring criteria.

5. Optimize — either run the CLI directly or use the host-driven skill:

# CLI (built-in optimizer)
overmind optimize <agent-name>

# or via Cursor chat (host-agent-driven loop)
# /overmind-optimise-agent <agent-name>

Skills assume commands are run from the directory that contains the relevant .overmind/ folder. All API keys and default models must be configured via overmind init before invoking any skill.

How it works

1. Initialize (overmind init)

Configure API keys and default models. Writes .overmind/.env in the current directory. Safe to re-run.

2. Register your agent (overmind agent register)

Point Overmind at the Python function it should call for each test case:

overmind agent register <name> <module:function>

The module path is resolved relative to the project root. Your function receives an input dict and must return a dict.

Other registry commands:

Command	Description
overmind agent list	List all registered agents
overmind agent show <name>	Show registration details and pipeline status
overmind agent update <name> <mod:fn>	Update the entrypoint (e.g. after renaming a file)
overmind agent remove <name>	Remove from registry and instrumented copy
overmind agent validate <name> --data <path>	Run the first test case to verify the entrypoint

3. Setup (overmind setup)

An interactive flow that prepares everything the optimizer needs:

Phase	What happens
Agent analysis	An LLM reads your agent code to detect the input/output schema, tools, and decision logic.
Policy generation	If you pass --policy, your document is analyzed against the code and improvements are suggested. Otherwise, a policy is inferred from the code automatically. You can refine either version in a conversational loop until you approve it.
Dataset	Overmind either uses your existing test data or generates diverse synthetic cases based on the policy and agent description.
Evaluation criteria	Scoring rules are proposed for each output field. Policy constraints inform stricter scoring where relevant. You can accept, refine, or edit manually.

Setup produces two artifacts in .overmind/agents/<name>/setup_spec/:

• eval_spec.json — machine-readable evaluation spec (used at runtime)
• policies.md — human-readable policy document you maintain

Both are editable after generation.

Flag	Description
--fast	Skip all prompts. Requires ANALYZER_MODEL and SYNTHETIC_DATAGEN_MODEL in .env.
--data PATH	JSON seed dataset file or directory of *.json files (optional; wizard can pick data instead).
--policy PATH	Provide an existing policy document. Overmind analyzes it against agent code and suggests edits.

4. Optimize (overmind optimize)

The iterative optimization loop. You configure a few settings interactively (or use --fast for defaults):

Setting	Description
Analyzer model	The strong model that diagnoses failures and generates code fixes.
LLM-as-Judge	Optional semantic scoring alongside mechanical matching (adds ~10% eval cost).
Iterations	Number of optimize → evaluate → accept/revert rounds (default: 5).
Candidates per iteration	How many variant fixes to generate per round (best-of-N). Each biases edits toward a different area — tool descriptions, core logic, input handling, system prompt. Higher N improves odds but costs more.
Parallel execution	Run agent evaluations across multiple workers.

What happens each iteration

1. Run the agent on every test case and collect traces + outputs.
2. Score outputs against the eval spec (0–100 across dynamic dimensions).
3. Diagnose — the analyzer receives traces, scores, policy, and code. It identifies failure patterns and root causes.
4. Generate N candidate fixes, each targeting a different area of the code. If N≥3, the last candidate uses a separate diagnosis for diversity.
5. Validate — syntax checks, interface checks, and a smoke test on a small case subset.
6. Evaluate — surviving candidates are scored on the full dataset.
7. Accept or revert — the best candidate is kept only if it improves the score without regressing too many individual cases.

Advanced settings (available during interactive config) include regression thresholds, train/holdout splits to detect overfitting, early stopping patience, and diagnosis visibility controls.

Flag	Description
--fast	Skip all prompts. Requires ANALYZER_MODEL in .env. Uses defaults.

Multi-file agents

By default Overmind optimizes the single registered entry file. For agents split across multiple modules, it automatically resolves local imports, extracts individual functions and classes, and applies targeted edits back to the original files — so your project structure stays intact.

Agent policies

Policies are the foundation of meaningful optimization. They tell the optimizer what the agent should do, not just how it currently scores — preventing improvements that raise numbers but violate business rules.

A policies.md looks like this:

# Agent Policy: Lead Qualification

## Purpose
Qualifies inbound sales leads by analyzing company data and inquiry content.

## Decision Rules
1. If the inquiry mentions "enterprise" or "custom pricing", classify as hot
2. Companies with 500+ employees get a minimum lead score of 60

## Constraints
- Never disqualify without checking company size
- Score and category must be consistent (hot = 70+, warm = 40-69, cold = <40)

## Priority Order
1. Accuracy of category classification
2. Score calibration
3. Reasoning quality

## Edge Cases
| Scenario             | Expected Behaviour                    |
|----------------------|---------------------------------------|
| Missing company name | Default to cold, note in reasoning    |
| Competitor inquiry   | Classify as cold, recommend nurture   |

## Quality Expectations
- Reasoning should reference specific data points from the input
- Scores should be calibrated: hot leads 70-100, warm 40-69, cold 0-39

Policies feed into diagnosis prompts, code generation constraints, synthetic data generation, and LLM-as-Judge scoring — so every stage of the pipeline respects your domain rules.

Using your own data

Data files are JSON arrays where each element has an input and expected_output:

[
  {
    "input": { "company_name": "Acme Corp", "inquiry": "Need enterprise pricing" },
    "expected_output": { "category": "hot", "lead_score": 85 }
  }
]

Place data files in your agent directory under data/ and Overmind will detect them during setup. If you don't have data, Overmind generates realistic synthetic test cases using the policy and agent description.

Output

After optimization, results are saved under .overmind/agents/<name>/:

Path	Description
setup_spec/policies.md	Agent policy document
setup_spec/eval_spec.json	Evaluation criteria with embedded policy
setup_spec/dataset.json	Test dataset used for optimization
experiments/best_agent.py	The highest-scoring agent version
experiments/best_agent/	All optimized files (multi-file agents only)
experiments/results.tsv	Score history for every iteration
experiments/traces/	Detailed JSON traces of every agent run
experiments/report.md	Summary report with scores and diffs

Other paths under .overmind/ (not all exist until you run commands):

Path	Required?	Notes
agents.toml	Yes for overmind CLI	Registry of agent names and module:fn entrypoints.
.env	Optional	API keys and model defaults from overmind init.
agents/<name>/.env	Optional	Per-agent overrides (written by setup when you save keys).
agents/<name>/instrumented/	Regenerated	Full mirror of the project root (everything under the directory that contains .overmind/, minus skips like .git, venv). Put .overmind next to a small project root so this tree stays small.
agents/<name>/run_state.json	Written by optimize	Regression cases and run history across sessions.
logs/overmind.log	Auto	Rotating CLI log from setup_logging.
agents/<name>/instrumented/.overmind_runners/	Ephemeral	Generated subprocess wrappers (_run_agent.py, etc.); removed when the runner calls cleanup(); safe to delete manually.

Bundle scope and caps

For large repositories, the optimizer resolves a bounded import closure (defaults: 24 files, 60k characters) and skips common paths (tests/, docs/, .overmind/, etc.) using built-in rules plus optional .overmindignore / .gitignore.

After overmind setup, eval_spec.json may include a scope block (optimizable_paths, context_paths, exclude_paths as globs relative to the project root). Inspect what will load without running an LLM:

overmind doctor my-agent

One-off overrides:

overmind optimize my-agent --scope "myagent/prompts/**/*.py" --max-files 32 --max-chars 80000
overmind setup my-agent --scope "agents/core/*.py"   # hints for the analyzer

CLI reference

overmind init                                        Configure API keys and models
overmind agent register <name> <mod:fn>              Register an agent
overmind agent list                                  List registered agents
overmind agent show <name>                           Show agent status
overmind agent update <name> <mod:fn>                Update entrypoint
overmind agent remove <name>                         Remove from registry
overmind agent validate <name> --data <path>         Run first test case to verify entrypoint
overmind setup <name> [--fast] [--data PATH] [--policy PATH]  Analyze agent, build eval spec
overmind optimize <name> [--fast] [--scope GLOB] [--max-files N] [--max-chars N]  Run optimization loop
overmind doctor <name>                               Diagnose bundle scope and eval spec (read-only)
overmind sync [name]                                 Sync local setup artifacts to Overmind
overmind sync-optimize [name]                        Sync local optimize artifacts to Overmind

Run overmind <command> --help for full documentation on any command.