agentplay CLI
agentplay turns local Claude Code, Codex and Cursor sessions into a file map and a replayable step timeline.
- 352 ★ on the repository
- Last commit
- Apache-2.0
Minimal access
Why this level
- Reads local agent transcripts; the only network call, an optional email signup, is disabled by an environment variable.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/agentplayDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- Reads local agent transcripts; the only network call, an optional email signup, is disabled by an environment variable.
This is third-party code. Review the repository files before installing.
What it does
agentplay finds agent transcripts on the machine, builds a redacted map of each session and opens a local viewer. The map shows which files the agent read and edited, which commands failed and which MCP servers and subagents took part. The timeline plays back at 1x, 4x and 16x. Claude Code, Codex, Cursor, GitHub Copilot, OpenCode, Kiro, Gemini CLI, Goose and others are supported. `agentplay stats` sums cost, time and errors per agent over a chosen period.
Who it is for. Developers who run agents on real tasks and want to see what a session did, where it retried commands and where the context went.
Good fit when
- An agent made a surprising edit and you want to recover what it read before it.
- A session ran suspiciously long and you need to find the repeating failing command.
- A team compares weekly spend on Claude Code and Codex.
- A session result must be shown to a colleague without prompt text.
Not a fit when
- You need a cloud viewer for transcripts without running anything locally.
- You want to analyse prompt and reply text, which the map does not store.
- You need tokens and cost for Cursor, which records neither.
Example request
Build a map of my latest session and show which files the agent read before editing, which commands failed and how often they repeated.Limitations
Python 3.11 or newer is required. For Cursor, Copilot, OpenCode, Kiro, Gemini CLI, Goose and others the map shows only the main lane; separate subagent lanes exist for Claude Code and Codex. Cost is an estimate from public list prices. On first launch the viewer offers an email signup for Entelligence updates; declining sends nothing, and `AGENTPLAY_NO_NETWORK=1` disables the network call entirely.
How to disable. Stop the viewer with Ctrl+C. Remove the package the way it was installed, and delete `~/.agentplay` with the index and maps.
FAQ
What does a map contain?
Repo names, repo-relative file paths, tool, MCP server and skill names, models, tokens, timings and error classes. There is no prompt text, reply text or command arguments.
How do I share a session?
Run `agentplay export <session>` and rebuild with `--path-mode dirs_only` for public posting. Repo names stay, so remove the `repos[].remote` field by hand if that is sensitive.
Does agentplay use the network?
Only for the optional email signup in the viewer. Session data stays on the machine, and `AGENTPLAY_NO_NETWORK=1` switches that call off.
Is an account or API key required?
No. The command reads transcripts already on disk and works without signup.
Which agents are supported?
Claude Code, Codex, Cursor, GitHub Copilot, OpenCode, Kiro, Gemini CLI, Antigravity CLI, Command Code and Goose. `agentplay agents` shows which of them were found on the machine.
Discussion
No questions or reviews yet.
Related
CLI toolsSession monitor for Claude Code, Cursor, Codex and six more agents in a terminal, window or HTTP API, with chat cleanup and search
A Claude Code and Codex plugin with skills for code research, tests, refactoring and independent review of finished changes
One desktop window for Claude Code, Codex, Gemini CLI and OpenCode: terminals, pull requests, diffs and approvals
A Python terminal coding agent in the manner of Claude Code: reads the repository, edits files, runs commands, with memory, sub-agents and MCP
A coding-agent monitor with a project component map and an experimental 3D kitchen view of ongoing tasks
Over twenty skills for coding agents: MCP and hooks, plan mode, a bug-fix protocol, prompting and a bash safety hook
README in short
Maps are rebuilt from transcripts, so later runs only rebuild sessions that changed. The privacy contract is enforced by tests described in `docs/format.md`.