Rashomon
An open-source AI agent observability tool that independently records the actual execution process of AI agents and cross-validates it against the final summary. It is used to detect inconsistencies between summaries and actual execution results in multi-agent architectures.
Timeline (last 90 days)
Rashomon is an open-source AI coding agent monitoring tool for tracking the actual behavior of agents during execution.
The author of Rashomon solicited community feedback on demand for and improvements to the tool in a Reddit post.
Rashomon records execution details such as command execution, test failures, file changes, sub-agent activity, and discrepancies between what agents say and do.
The author of Rashomon positions it as "Sentry for AI coding agents."
Rashomon is named after the 1950 film Rashomon by Japanese director Akira Kurosawa.
rashomon --timeline can flag suspicious patterns where a test command first fails and then passes after only test files have been modified.
Rashomon does not store test names, test output, prompts, or file contents during detection, and relies entirely on the captured execution history and classification of commands and files to make inferences.
The rashomon spend feature is used to estimate Claude Code usage costs based on standard API pricing, and can break down costs by model, agent, and session dimensions.
The cost statistics of rashomon spend include expenditures incurred by rounds that failed but were not mentioned in the final summary.
The Rashomon project has been open-sourced on GitHub, with the repository at altrace-dev-role/rashomon.
8 more timeline events
All Facts (15)
Rashomon is an open-source AI coding agent monitoring tool for tracking the actual behavior of agents during execution.
50%UnverifiedThe author of Rashomon positions it as "Sentry for AI coding agents."
50%UnverifiedRashomon records execution details such as command execution, test failures, file changes, sub-agent activity, and discrepancies between what agents say and do.
50%UnverifiedThe approach of cross-validating agent claims with an external, independent execution record represents a pragmatic direction for improving agent reliability.
50%UnverifiedRashomon is named after the 1950 film Rashomon by Japanese director Akira Kurosawa.
50%UnverifiedThe rashomon spend feature is used to estimate Claude Code usage costs based on standard API pricing, and can break down costs by model, agent, and session dimensions.
50%UnverifiedThe cost statistics of rashomon spend include expenditures incurred by rounds that failed but were not mentioned in the final summary.
50%UnverifiedRashomon does not store test names, test output, prompts, or file contents during detection, and relies entirely on the captured execution history and classification of commands and files to make inferences.
50%UnverifiedRashomon is an execution record system independent of an agent's self-reporting that reconstructs the truth from actually observed command executions, file changes, test runs, and sub-agent activities, then compares this against the agent's self-report to flag inconsistencies.
50%Unverifiedrashomon --timeline can flag suspicious patterns where a test command first fails and then passes after only test files have been modified.
50%UnverifiedRashomon is an open-source tool whose core idea is to independently record what an AI agent actually executed and flag inconsistencies when the actual record diverges from the summary.
50%UnverifiedThe Rashomon project repository is located at github.com/altrace-dev-role/rashomon.
50%UnverifiedThe tool name Rashomon is taken from Akira Kurosawa's 1950 film of the same name, alluding to different narrative versions of the same event.
50%UnverifiedThe author of Rashomon solicited community feedback on demand for and improvements to the tool in a Reddit post.
50%UnverifiedThe Rashomon project has been open-sourced on GitHub, with the repository at altrace-dev-role/rashomon.
50%