# Izri

> AI QA for AI-generated code. When a coding agent (Claude Code, Cursor, Aider, Devin, your own bot) writes a PR, Izri answers one question: is this delta safe to merge? It does that by computing four independent signals — scope, tests, hallucination, visual — and aggregating them into a single `izri/quality` umbrella verdict.

Izri is a quality gate, not a code-review tool. Findings are addressed to the human — we never propose patches.

## Getting started

- [What is Izri](https://izri.ai/docs/start/what-is-izri): the four signals, the umbrella verdict, and what Izri deliberately does not do.
- [Quickstart](https://izri.ai/docs/start/quickstart): add the GitHub Action, mint a token, first verdict in five minutes.
- [Your first verdict](https://izri.ai/docs/start/your-first-verdict): reading the check run and the sticky PR comment.

## Install

- [GitHub Action](https://izri.ai/docs/install/github-action): all 8 inputs, 4 outputs, and branch-protection wiring. The recommended path.
- [CLI](https://izri.ai/docs/install/cli): `izri run`, `izri check-scope`, `izri status` — terminal surface for the four signals.
- [MCP server](https://izri.ai/docs/install/mcp): mount `@izri/mcp` in Claude Code / Cursor / Windsurf. Nine tools for in-agent scoring.
- [Self-hosting](https://izri.ai/docs/install/self-hosting): Docker Compose, environment, and the runner you must not forget.

## The four signals

- [Signals overview](https://izri.ai/docs/signals/overview): every weight, every hard rule, and what each severity does. Composite threshold is 70.
- [`izri/scope`](https://izri.ai/docs/signals/scope): does the diff match the PR's stated intent, and does it touch paths it shouldn't?
- [`izri/tests`](https://izri.ai/docs/signals/tests): did the relevant tests run and pass against the diff?
- [`izri/hallucination`](https://izri.ai/docs/signals/hallucination): did the passing tests actually exercise the changed lines? Heaviest signal at weight 0.40.
- [`izri/visual`](https://izri.ai/docs/signals/visual): visual regression on changed routes. Semantic diff is Team-tier and above.

## Configure

- [`.izri/config.yml`](https://izri.ai/docs/configure/config-yml): full schema and the in-repo › dashboard › defaults precedence chain.
- [Projects and repositories](https://izri.ai/docs/configure/projects): GitHub App installation and how a diff resolves to a project.
- [Organizations and roles](https://izri.ai/docs/configure/organizations-and-roles): OWNER / ADMIN / MEMBER / VIEWER and their scopes. All permissions are org-scoped.
- [API tokens](https://izri.ai/docs/configure/api-tokens): minting, the `izri_` prefix requirement, scoping, rotation.
- [Webhooks](https://izri.ai/docs/configure/webhooks): three event types, `sha256=` HMAC verification, retry model.

## Reference

- [REST API](https://izri.ai/docs/reference/rest-api): the HTTP surface, authentication, and error codes.
- [Verdicts and exit codes](https://izri.ai/docs/reference/verdicts-and-exit-codes): every status and its CI exit code. Read this before scripting against Izri.
- [Plans and quotas](https://izri.ai/docs/reference/plans-and-quotas): Free / $9 / $29 / $99 per organization, flat. What each tier unlocks.
- [Troubleshooting](https://izri.ai/docs/reference/troubleshooting): common failure modes, organized by symptom.

## Optional

- [llms-full.txt](https://izri.ai/llms-full.txt): the entire docs corpus concatenated. Use when you need everything in one fetch.
- [Pricing](https://izri.ai/pricing): the plan comparison matrix.
- [Integrations](https://izri.ai/integrations): the three install paths with copy-pasteable snippets.
- [Repository](https://github.com/legendify-dev/izri): source code (open-core under AGPL-3.0).

Every page under https://izri.ai/docs/ is also reachable as raw markdown by appending `.md` — for example https://izri.ai/docs/install/cli.md.