no_human plugin
A Claude Code plugin that hands a task to a local no_human and reviews a branch with a reviewer that never saw the coder's session
- 331 ★ on the repository
- Last commit
Broad access
Why this level
- The coder agent writes code and commits in your repository and opens a pull request
- Review uses your Claude token, and the GitHub Action needs an Anthropic key in repository secrets
- Tickets and pull request links pass between the tracker, the local server and agents
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/no-humanDetects 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.
- The coder agent writes code and commits in your repository and opens a pull request
- Review uses your Claude token, and the GitHub Action needs an Anthropic key in repository secrets
- Tickets and pull request links pass between the tracker, the local server and agents
This is third-party code. Review the repository files before installing.
What it does
The no_human plugin for Claude Code connects a session to the local no_human service. Two MCP tools send work to the local server: one files a task and the other shows the state, the number of attempts, the pull request link and token use. Three skills add to them: the first frames a request for a new task, the second runs `nh gate` over the current branch or a pull request, and the third builds release notes from tags and commits. no_human itself takes a task from ticket to a reviewed pull request on your own machine. The coder agent writes a plan from the ticket and the code, then makes the change. Next a second model in a separate session, which never saw the coder's transcript, tries to refute the claim that it is done. The review result is a pass or fail checklist citing file and line. A pull request opens and the merge stays with a person. The tamper guard mechanically counts deleted tests, new skips and assertions turned into tautologies. Without justification the attempt stops. For bug fixes there is a reproduction gate: the tests offered as evidence must fail on the old code and pass on the new. Tickets arrive from Jira, Linear and monday.com, and GitHub and GitLab issues are imported by URL.
Who it is for. For developers and teams who hand tickets to coding agents and want an independent check of the result before merging.
Good fit when
- You want to pass a Jira or Linear ticket to an agent and get a pull request with a plan, tests and a review report
- You want a second-model review of the current branch with no running server or database, just the `nh gate` command
- You need to be sure the agent did not delete or weaken tests to get a pass
- You want proof that a bug fix really fixes it: the test fails before the change and passes after
- You need release notes from tags and commits that describe changes by their effect on the user
Not a fit when
- You want a hosted service with accounts: no_human runs only on your machine
- The task is a couple of lines: the full cycle with a plan and a separate review costs more than a manual edit
- You have no Claude credential: the review part uses your own token from the `claude setup-token` command
- You want automatic merging: no_human opens a pull request and stops
Example request
File the task "Rate-limit the login endpoint" with no_human for the repository /path/to/repo, then check its status.Limitations
The plugin does nothing without a running no_human install. Install the CLI with `uv tool install no-human` or `pipx install no-human`, run `nh init` and `nh doctor`, then start it with `nh start`. You need Python 3.12 or later, git and uv, plus Node.js when building from source.
The MCP bridge is fixed to the local address 127.0.0.1:8420 and has no authentication: the author treats one computer as a trusted domain. The branch review skill creates no config in the home folder, though the CLI's background update check may write a cache file.
The author gathered the statistics on how many attempts the second model sent back from their own task board. Check the behavior on your project before trusting it with critical changes.
How to disable. Remove the plugin with `/plugin uninstall no-human@no-human-ai` and, if you like, remove the marketplace. no_human itself stops when the `nh start` process ends and is removed with `uv tool uninstall no-human`. Its data lives in the ~/.no_human folder.
FAQ
How do I install the no_human plugin in Claude Code?
Run `/plugin marketplace add no-human-ai/no_human` and `/plugin install no-human@no-human-ai`. The repository itself acts as the marketplace, so cloning alone registers nothing.
What does `nh gate` do?
It runs the independent reviewer and the tamper guard over the diff of the current branch or a pull request. No server or database is needed, only the CLI and your Claude token. The command only reads and changes nothing in the branch.
How is the reviewer different from an agent's own self-check?
It runs on a different model in a session that never saw the coder's transcript and is told to refute the claim that the work is done. It does not accept a self-score: every blocking finding must cite a file and line.
Which trackers are supported?
Jira Cloud, Linear and monday.com are polled through their APIs, and issues from GitHub and GitLab are imported by URL. Writing back to the tracker is off by default and turns on with the `write_back` setting.
Can I run the check as a GitHub Action?
Yes. The repository has a GitHub Action, no-human-ai/no_human: it runs the same reviewer and tamper guard as a pull request check and leaves one checklist comment that it updates on every run. It needs your Anthropic key in secrets.
Discussion
No questions or reviews yet.
Related
PluginsA skills library that gives coding agents a development process: brainstorming, planning, TDD, subagents and code review
Skills for real engineers by Matt Pocock
Skills For Real Engineers
Small composable skills for engineering with agents: plan grilling, TDD, bug diagnosis, code review and architecture
Karpathy-inspired Claude Code guidelines
Karpathy-Inspired Claude Code Guidelines
Four principles in one CLAUDE.md against common LLM coding mistakes: silent assumptions, overengineering, drive-by edits
A plugin and skill that makes the agent write minimal code: reuse, stdlib and native platform features come first
GitHub toolkit for spec-driven development: the specify CLI adds agent commands and skills to a project, from principles to implementation
Garry Tan's skill suite that turns Claude Code into a virtual team: product, architecture, design, review, QA and shipping
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Repository stars | Main difference |
|---|---|---|---|
| no_human (this entry) | no | 331 | |
| SWE-AF | no | 1,036 | a team of AI agents that writes code in parallel branches itself, while no_human adds an independent reviewer and a tamper guard |
| Claude Code Security Reviewer | no | 6,490 | checks the security of pull request changes, while no_human judges task completeness and test integrity |
README in short
The main documentation has a separate section on tracker and CI integrations and publishes the author's statistics on rejected attempts. The plugin has its own short guide to installing from the marketplace.