Every agent PR should explain itself.
Our primary project is Shale: agent PR evidence that captures what an AI coding agent was asked to do, what it touched, and what checks ran — then renders that evidence as a card on the pull request.
| Repo | Role | License |
|---|---|---|
| shale | Agent PR evidence: intent capture, session evidence, and PR cards — the primary project | Apache-2.0 |
| prism | Graph-ranked context delivery for AI coding agents (CLI + MCP) | MIT |
| grove | Persistent code-graph engine: Tree-sitter parsing, SQLite, impact and test queries | MIT |
| fuse | Symbol-aware Git merge driver built on Grove | MIT |
brew install provasign/shale/shale
cd your-repo
shale init
git add . && git commit -m "chore: enable shale" && git pushNo account, no server, no GitHub App, no token paste. The next agent-authored PR carries a card — see the live demo pull requests for real examples, or start with the getting-started guide.
- Shale — you want every agent-authored PR to explain itself to reviewers.
- Prism — you want agents to retrieve graph-ranked code context instead of
chasing
rgresults across a dozen file reads. - Grove — you want the code graph itself: indexing, symbols, blast radius, test coverage queries, embeddable Go API.
Everything is local-first and fail-open by design. Docs live at provasign.dev.