genesis

agent
Security Audit
Pass
Health Pass
  • License — License: Apache-2.0
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Community trust — 52 GitHub stars
Code Pass
  • Code scan — Scanned 12 files during light audit, no dangerous patterns found
Permissions Pass
  • Permissions — No dangerous permissions requested

No AI report is available for this listing yet.

SUMMARY

Turn your codebase into a business flow map and see which agent changes are untested or risky. An agent skill for Claude Code and Codex.

README.md

Genesis icon   Genesis

Visualize your agent's changes.
See where they can break.

License: Apache-2.0 Python 3.12 GitHub stars

English  ·  简体中文

A checkout flow drawing itself step by step, then its four changes landing with test verdicts; the card charge is marked as a risk
▶ Watch the 37-second demo

Genesis turns your codebase into a business flow map and shows which of your agent's
changes are untested or risky. It runs as a skill inside Claude Code and Codex.

Quick start

git clone https://github.com/GenesisAgentInfra/genesis.git
cd genesis
uv sync --python 3.12 --package genesis-ci --no-dev --frozen
uv run --package genesis-ci --no-dev --frozen genesis agent install --agent claude

Use --agent codex or --agent both if you work in Codex.

Now open your own project in Claude Code and ask:

/genesis map this project
/genesis what does my uncommitted work change, and is it tested?
/genesis review this branch against main

In Codex the skill is $genesis. When it finishes, open .genesis/view.html in a
browser. The summary.md next to it is ready to paste into a pull request.

What you get

Say you asked your agent for guest checkout. It built that, and changed three other
things on the way. Genesis draws the checkout flow and puts all four changes on it:

The first half of the Place an order flow, branching on guest or signed in, stock, and the free-shipping threshold

What changed for shoppers Is it tested?
A card charge that times out is retried up to 3 times, with no idempotency key. A shopper can be charged twice. 🔴 Not reached. No test makes the payment time out.
Stock is no longer held while the card is charged. Two shoppers can pay for the last unit. 🟡 Reached only. Tests run this code but none buys the last unit twice.
Shoppers can check out as a guest by leaving an email. ⚪ Unknown. The guest test stubs out the email check.
Free shipping starts above $35 instead of $50. ✅ Covered. Tests place orders at $34.99, $35.00 and $35.01.

Click a step to see the code behind it and the tests that touch it:

The Charge the card step opened: the risk, what changed for shoppers, and the test evidence behind the verdict

Can I trust it?

Your agent does the reading, but it doesn't get the last word. Every step it writes has
to quote the lines of code it stands for, and the Genesis CLI checks each quote against
a frozen snapshot of your repo. If one doesn't match, the whole description is rejected
and the agent tries again. If the agent skips a branch, the map is marked partial.

The CLI calls no model, sends nothing anywhere, and never runs your code.

Limits

  • Python repositories only, for now.
  • Test verdicts come from reading your tests. Genesis doesn't run them.
  • Risks are what your agent noticed while reading. There is no rule engine.

License

Apache-2.0. Bundled components keep their own licenses, listed in
NOTICE and THIRD_PARTY_NOTICES.md. Want to help?
See CONTRIBUTING.md.

Star history

Star history of GenesisAgentInfra/genesis

Reviews (0)

No results found