AgentLedger
Files AI agent reports into a private git store, renders them as annotatable pages, and routes questions back to the session that wrote them.
At a Glance
Self-hosted AgentLedger CLI and skill under the Apache-2.0 license. Reports are stored in your own private git store.
Engagement
Available On
Listed Nov 2026
About AgentLedger
AgentLedger is an open-source Python CLI and agent skill for Claude Code and Codex, published by SecretSettler under Apache-2.0. It files the long reports an agent writes into a private git store, renders them as reviewable pages, and sends reviewer questions back to the session that produced the report.
What It Is
AgentLedger is a report review tool for agent-written work such as benchmark comparisons, audits, and investigations. The agent writes Markdown and publishes it with the CLI. Each publish becomes the next version (v001, v002, and so on), records the repo, branch and commit it describes, and attaches the gzipped transcript of the writing session.
How Review Works
Running agentledger open serves the store on localhost. Every ## section is a module that can be commented on, linked to, and compared across versions. Selecting text lets a reviewer add a comment, saved to the version's comments.csv, or press Ask. Ask sends the question, quote and thread to the session that wrote the report. Claude Code resumes a fork of that session, while Codex resumes it in place. Without the original session, a fresh run answers from the trace. Answers are saved as agent replies in comments.csv.
The agent then lists waiting threads, replies, and publishes a new version with --resolves, flagging changed modules and pointing resolved threads at the fixing version.
Sharing and Privacy
A single report can be packed with agentledger share into a zip another person opens on their own machine and asks their own AI about. Teammates can also read and comment through serve --share, or clone the private store from GitHub. Pages load no external scripts, and report HTML passes an allow-list sanitiser. Ask runs the user's own agent CLI inside that agent's sandbox, is served only to requests from the local machine, and is never exposed in share mode.
Setup Path
An install script installs the CLI, creates a private store at ~/agentledger-store, and installs the skill for Claude Code and Codex. A Claude Code plugin adds /agentledger:report, /agentledger:review and /agentledger:open. The CLI requires Python 3.9+ and has no dependencies.
Community Discussions
Be the first to start a conversation about AgentLedger
Share your experience with AgentLedger, ask questions, or help others learn from your insights.
Pricing
Open Source
Self-hosted AgentLedger CLI and skill under the Apache-2.0 license. Reports are stored in your own private git store.
- Apache License 2.0
- Python 3.9+, zero dependencies
- Claude Code plugin and Codex skill
- Private git store for agent reports
- Local review UI with line comments and Ask
Capabilities
Key Features
- Files agent reports as versioned entries in a private git store
- Attaches session transcripts as traces
- Local review server with section-level modules
- Comment on any selected text
- Ask the originating agent session about a passage
- Version comparison with changed-module flags
- Thread resolution tracked across versions
- Share a single report as a zip for another person's AI
- Dashboard and project map
- Claude Code plugin and Codex skill
- Sandboxed agent execution for answers
- HTML sanitiser for report pages
