About Scavi
Clean context.
Better agents.
Scavi is an open-source context linter for AI coding agents. It checks whether repository instructions still match the code they describe.
What it catches
- stale paths and removed files
- invalid package scripts
- package-manager mismatches
- conflicting agent instructions
How it works
Scavi reads files such as AGENTS.md, CLAUDE.md, Cursor rules, and Copilot instructions, then validates their claims against repository evidence. Deterministic answers stay local and require no LLM.
Optional semantic verification
When static checks stop, evidence takes over.
Scavi indexes repository text locally, retrieves only the code related to one claim, and asks OpenAI or a local Ollama model for a conservative verdict: consistent, stale, or uncertain.
The full repository is never sent to the model. Results below your confidence threshold stay uncertain and do not create a warning.
AGENTS.md:3Configuration is persisted in JSON files.STALE · 92%Evidence:
src/storage.ts:1-8SQLite storage foundCore commands
npx scavi-cli checkcheck repository contextscavi initcreate a configurationscavi fixpreview minimal safe fixes