Vibeleaderboard
Index / tool
Visit github.com
Category
Developer Tools
Rank
No. 3785Tools index

Previous survey · No. 3592 ·

Pricing
Open Source
Type
TOOL
Use case
Model & Agent Evaluation · Software Testing & Quality
GitHub
22 stars
Latest release
v1.2.0-alpha.2
Date

About

An open-source execution recorder for AI coding agents that keeps its own record of what happened and compares it against the agent's account. For Claude Code it records shell commands, exit codes, tool calls, test outcomes and subagent ordering, and stores keyed digests rather than prompts, responses, file contents or tool output. It flags patterns such as a failing test that later passes after only test files changed, which can mean the agent edited the test instead of fixing the bug.

At a glance

Rashomon records Claude Code tool calls and reports whether a check passed after the last edit. It is worth evaluating if an agent’s claim that work is done often sends you back through its transcript looking for a test result.

Good fit

Claude Code users who want local evidence of edits and checks, plus repository-specific guidance before an agent starts.

Consider something else when

Use a separate enforcement layer if your requirement is to block commands or prevent edits.

Start here

Ask for two changes in a scratch repo. Run a test after the first change and leave the second untested. See whether the end-of-turn report distinguishes the two.

Cost and requirements

Apache 2.0. The documented plugin uses Claude Code hooks. Delivery of some notes depends on the installed Claude Code version and session surface.

Limits to know

It writes a local event store and can inspect transcripts to verify note delivery. Review recording and retention settings for sensitive repos. A passing command still needs to cover the behavior that changed.

Tags

claude-codeobservabilityai-agentsauditclitesting

Tech Stack

Go

Comments (0)

No comments yet

Editorially curated, with community endorsements as a secondary signal. Corrections welcome.