watchmen plugin
watchmen watches Claude Code and Codex sessions and builds skill bundles and workspace briefs for later runs.
- 295 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- Reads the full agent session history and sends fragments to an outside model provider.
- Installs a background daemon and hooks into the system scheduler.
- Reads Claude Code login tokens from the macOS keychain in OAuth mode.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/watchmenDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Reads the full agent session history and sends fragments to an outside model provider.
- Installs a background daemon and hooks into the system scheduler.
- Reads Claude Code login tokens from the macOS keychain in OAuth mode.
This is third-party code. Review the repository files before installing.
What it does
watchmen reads session history from `~/.claude/projects/`, finds repeated actions and writes skill bundles plus `CLAUDE.md` and `AGENTS.md` workspace briefs. The same skills reach Claude Code, Codex and pi.dev, so switching agents on one repository needs no re-explaining. Setup runs through the `watchmen init` wizard: it asks for a model provider, previews analysis cost and installs a background daemon with a local viewer. Plugins attach through `/plugin` in Claude Code and `/plugins` in Codex.
Who it is for. Developers working heavily with coding agents who are tired of repeating the same procedures to them.
Good fit when
- The agent rediscovers how to run tests and builds every session.
- The project switches between Claude Code and Codex.
- Skills should come from your own habits instead of hand-written files.
- A daily workspace digest is wanted before a session starts.
Not a fit when
- Session history holds private code that cannot go to an outside model.
- A background process and local database are unwelcome.
- One hand-written `CLAUDE.md` is enough.
Example request
Show a project digest and which skills watchmen built from my recent sessions.Limitations
Analysis costs money: calls go to the chosen model provider, and the wizard shows the price first. Fragments of session history are sent to that provider. Keys are stored in `~/.config/watchmen/.env`. Subscription login through Claude Pro and ChatGPT works only on macOS and is experimental for ChatGPT. The last commit in the repository dates from August 2026. The first pass over history takes from thirty minutes to ninety per project. On Windows and Linux login through a Claude Code subscription does not work, so an API key is needed.
How to disable. Disable the plugin in the agent, stop the daemon and viewer in the system scheduler, then delete `~/.watchmen/` and `~/.config/watchmen/`.
FAQ
Which agents are supported?
Claude Code, Codex and pi.dev. Codex has no status line, so the `brief` command is used instead.
What should be checked after setup?
`watchmen doctor` shows key, corpus, daemon, viewer and hook status on one screen.
Where is data stored?
In `~/.watchmen/`. The `WATCHMEN_HOME` variable sets another location.
Which model providers are supported?
OpenRouter, OpenAI and Anthropic by API key, plus Claude Pro and ChatGPT by subscription login on macOS.
What is needed to run it?
Python 3.11 or newer, the `uv` manager, credentials for one provider and at least one supported agent.
Discussion
No questions or reviews yet.
Related
PluginsA plugin that syncs Claude Code, Codex CLI and Antigravity on one project and lets them hand work to each other.
The /codex-workflows skill: Claude Code writes a script, dozens of Codex agents run it, and the run shows on a map
Claude Code supervises while Codex, opencode and pi work on your quota: delegated edits, independent review, model discussions
ECX ties Claude Code and Codex together with shared memory so each sees the other latest messages, plus 19 skills and 21 hooks.
An operating layer for Claude Code and Codex: one /do entry point, project memory across sessions, safety hooks, cost telemetry and parallel agents
A template sets up a multi-agent environment in a repository: Claude Code conducts Sonnet and Opus subagents and calls Codex CLI.
README in short
The tool uses real sessions as its source and writes the result as skills both agents can read.