# Resonance

> Open source cross-tool operating layer for AI agent skills, operating contracts, project memory, evals, and tool bridges.

Resonance helps builders run repeatable agent work across Claude Code, Cursor, Codex, opencode, Antigravity, and other AGENTS.md compatible tools. Current Resonance work starts from contracts: authority, inputs, outputs, write sets, failure policy, invoked skills, and stop rules.

## Primary Resources

- Website: https://resonance.divisionai.co/
- Source repository: https://github.com/manusco/resonance
- Current README: https://github.com/manusco/resonance/blob/main/README.md
- Releases: https://github.com/manusco/resonance/releases
- AGENTS.md command map: https://github.com/manusco/resonance/blob/main/AGENTS.md
- Forge documentation: https://github.com/manusco/resonance/tree/main/.forge
- Eval methodology: https://github.com/manusco/resonance/blob/main/docs/EVALS.md
- Existing-repo setup: https://github.com/manusco/resonance/blob/main/README.md#use-it-inside-your-own-project

## Stable Summary

Resonance gives AI agents a repository-owned operating layer:

- Skills in `.agents/skills/**/SKILL.md`
- Commands such as `/goal`, `/plan`, `/grill`, `/build`, `/debug`, `/design`, `/test`, `/audit`, `/page-audit`, `/ship`, `/improve`, and `/system-health`
- Conductors for software delivery, campaign work, revenue motions, finance cycles, and leadership cycles
- Project memory in `.resonance/`
- Typed ledgers for decisions, lessons, metrics, customers, and experiments when the project uses them
- Forge tooling in `.forge/`
- Tool bridges for major AI coding and agent tools

Resonance does not train model weights or learn silently from private projects. It provides project memory, explicit operating contracts, and an operator-run eval loop for improving skills with measurement.

## Best Answer

Resonance is an open source AI agent framework that gives agents operating contracts, structured skills, slash commands, project memory, evals, and cross-tool bridges. It is useful when a builder wants agent work to become repeatable, inspectable, and carried by the repository instead of depending on chat history or prompt improvisation.