What is Sonah?
Sonah reviews coding-agent sessions alongside the resulting code, helping developers check whether an agent followed instructions, what it decided independently, and whether its claims match the work. It supports Claude Code and Codex, runs locally, and produces a review page and Markdown report. The command-line tool is free and open source; Sonah Team adds shared branch reviews for $10 per user per month, plus tax.
Sonah Features:
Prompt Ask Tracking: Sonah lists each request from the session and marks whether it was completed, partly completed, missed, or not checked.
Independent Decisions: The report surfaces choices the agent made without asking, including work it chose not to do, so reviewers can assess them.
Code Review: A read-only review agent examines the repository, checks changes against the user's requests and the coding agent's claims, and quotes code for its findings.
Session Claim Checks: Sonah compares the agent's recorded results with statements it made, such as a claim that tests passed despite a recorded failure.
Checks and Omissions: The review can flag commands that failed or never ran, as well as planned checks that did not happen.
Review Sharing: Sonah Team lets teammates confirm, override, or comment on review items and open comments in Claude Code to address them.
Sonah Benefits:
Clearer Merge Decisions: Reviewers can judge agent work against the original request and session context, rather than relying on the code diff alone.
Fewer Hidden Surprises: Surfacing unapproved decisions and skipped work makes potential issues visible before they are accepted into a branch.
Evidence for Findings: Quoted passages from the session or code give reviewers specific evidence to evaluate instead of an unsupported score or verdict.
Flexible Review Handoff: A local report can be shared as a Markdown file or page, while teams can discuss review items together when they need to.
Control Over the Call: Because Sonah leaves the final judgment to the developer, teams can decide which findings matter and whether to act on them.
Use Cases:
Pre-Merge Agent Review: Check a Claude Code or Codex session before merging its pull request, especially when the diff does not explain the full task.
Public API Changes: Verify that an agent respected instructions to ask before changing existing public interfaces or other sensitive code.
Test Run Verification: Inspect a session where an agent reported passing tests, to see whether the recorded commands support that claim.
Daily Startup Shipping: Founders and engineers shipping frequently with coding agents can review agent contributions before they reach a shared branch.
Team PR Follow-Up: Engineering teams can review branch findings together, leave comments, and send a chosen comment into Claude Code for a fix.

