Sonah

Sonah: CLI tool that reviews coding agent sessions and code changes before pull requests.

Sonah evaluates what coding agents like Claude Code or Codex accomplished during a session, rather than just displaying the git diff. It checks prompt requirements, analyzes independent agent decisions, and uses read-only agents to inspect code for errors. The tool operates locally through personal credentials, processing session recordings and repository files to generate reports in page and markdown formats. It categorizes findings into specific action tiers with direct code quotes without issuing verdicts. It runs on Node.js and supports macOS and Linux environments.

Sonah at a glance

Best for: Software engineer
Runs on: macOS, Linux, CLI
Pricing: Free for individuals and non-profits; organizations with up to three users free; 30-day trial for other organizations; BUSL-1.1 license converting to Apache 2.0 after two years.
Released: October 2026
Category: AI coding tool
Website: github.com/getsonah/sonah

Key features

Evaluates prompt task completion status
Analyzes independent agent decisions
Inspects code changes for errors
Generates reports in page and markdown formats
Provides direct quotes from sessions and code
Links report items to GitHub repositories
Resumes interrupted review sessions

Use it for

Reviewing coding agent output before merging pull requests
Verifying if agent claims match actual test results
Checking autonomous decisions made by AI coding agents
Auditing repository modifications made during agent sessions

Alternatives to Sonah

Reviewer Agent — It uses a separate AI model to review code and prevent premature agent stopping.
Coucou — It provides open-source, real-time agent monitoring and approvals directly from the screen notch.
pstack — Improves AI coding agent outcomes with rigorous workflows.
jevgrep — Reduces coding agent costs by providing targeted, relevant repository context.
Agent Archive — Hand off coding sessions between Claude Code, Codex, and Cursor across machines.
Reviu — Review agent diffs, send fixes back, and clean branches with real Git before opening PRs.
Agent QA — It uses repository-based memory to learn from test runs and perform self-healing.
OpenCode — Works with 75+ model providers, local models, and existing subscriptions
Weft — It enables safe parallel coding agent work without file locks.
Agent Trials — Test if an agent fixes your bugs in your stack at a sane price before deployment.
Stop That Shit — Enforces task boundaries to stop unnecessary AI agent actions.
agent.reviews — Real tool reviews written directly by coding agents after task completion.

All Sonah alternatives