River Review

Turn review into an organizational judgment asset.

River Review is an OSS framework that codifies review standards as versioned, repo-owned Skills and applies the same team judgment from local Claude Code / Codex reviews through post-PR GitHub Actions checks.

River Review in 30 seconds

Remove noise locally before the PR, then re-check the PR with shared team standards.
AI produces findings and decision inputs; the caller or a human keeps ownership of GO / NO-GO and final approval.

• Try River Review in five minutes: Quickstart
• Start from reusable review knowledge: Starter Cookbook
• Run the pre-PR + post-PR two-stage pattern: Two-stage review gate
• Compare with other AI review tools: AI code review comparison
• Measure quality, regressions, and cost: Run store / regression, Cost estimation, and Dashboard

Fastest trial

Claude Code

/plugin marketplace add s977043/river-review
/plugin install river-review@river-review-marketplace
/reload-plugins
/river-review:review-local

Normal plugin usage needs no additional River Review LLM API key. Claude Code's own model applies the Skills.

Codex

codex plugin marketplace add s977043/river-review

After adding the marketplace, River Review's specialist review skills are available to Codex. See the Quickstart for the full path.

What changes in practice

A generic instruction such as "check error handling" can produce vague advice and leave the team to re-decide what matters. A River Review Skill carries scope, evidence expectations, severity, and false-positive avoidance rules.

• Make vague findings concrete: logging-observability points to the changed location and returns a finding about swallowed errors, observability impact, and a concrete fix.
• Reduce review-lens drift: fixture + golden output tests pin expected behavior and regression evals verify changes.
• Control low-value repetition: confidence / severity / suppression memory / review coverage reduce noise.

See Representative Skills for real fixtures and expected outputs.

Quality and operational capabilities available after adoption

River Review already provides the following foundations for team use:

• Skill quality: fixture + golden output tests, regression evals, and per-skill false-positive evaluation.
• Noise control: severity (critical / major / minor / info), confidence, suppression memory, and review coverage.
• Multi-review synthesis: consensus, Team Lead synthesis, and blind-spot reporting.
• Operational measurement: Run store, regression comparison, usage telemetry, cost estimation, and dashboard views.
• Adoption evidence: competitive comparison, known limitations, and FAQ.

Start with existing Skills and measurement. Codify new team-specific judgment only after real usage shows a gap.

Understand the concept

• Concept: turning review into an organizational judgment asset
• Welcome to River Review
• What is River Review
• Human Judgment Focus

Documentation structure

The official docs follow Diátaxis.

• Tutorials: step-by-step first success.
• Guides: task-oriented procedures.
• Reference: CLI, schemas, and output contracts.
• Explanation: concepts and design decisions.

Japanese is the source of truth, with English companions in corresponding .en.md files. 日本語版はこちら.