mancode CLI
Work modes for AI coding agents: a light solo mode for daily tasks and /man with a plan, checks and review for big changes
- 359 ★ on the repository
- Last commit
- AGPL-3.0
Broad access
Why this level
- Writes agent instruction files into the project: AGENTS.md, Cursor rules, skills
- The secrets vault injects macOS keychain values into commands you approved
- The .mancode folder accumulates plans and verification results and may capture sensitive context, so review it before committing
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/mancodeDetects 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.
- Writes agent instruction files into the project: AGENTS.md, Cursor rules, skills
- The secrets vault injects macOS keychain values into commands you approved
- The .mancode folder accumulates plans and verification results and may capture sensitive context, so review it before committing
This is third-party code. Review the repository files before installing.
What it does
mancode is a CLI from npm that adds files to a project so that Claude Code, Cursor, Codex and other agents can work in one of several modes. The solo mode is on by default: before an edit the agent checks whether existing code, the standard library or an installed dependency already does the job, and after the edit it reviews its own diff. For tasks that need a plan there is /man: the agent studies the project, asks only the questions that change the decision, proposes options and saves the plan in .mancode/local/workflows. Implementation starts after you confirm the plan. The other modes cover neighbouring cases. /manba reproduces a bug, finds the cause and walks the real user path to verify the fix. /manteam keeps shared decisions and task handoffs between people and agents in .mancode/shared. /manps builds a project report: stale TODOs, unused and risky dependencies, and hardcoded design values. Tasks, plans and verification results are stored under a stable TaskRef identifier. A new session picks the same task up after the agent restarts or you switch to another client. A separate part of the CLI scans a log, config or document for keys, emails, phone numbers and absolute paths and writes a redacted copy. mancode keeps secrets in the macOS keychain and injects them into approved commands by reference.
Who it is for. For developers and teams whose agent adds needless abstractions, skips the plan or loses context between sessions.
Good fit when
- On a simple edit the agent creates a new component, a style file and color variables although the project already has a Button and design tokens
- Before a large change you want a saved plan with options and acceptance criteria that you confirm before any code is written
- A task spans several sessions or several agents and its state has to survive a restart
- A team needs shared decisions, a glossary and task handoffs between people and agents in the .mancode/shared folder
Not a fit when
- You need the coding agent itself: mancode runs on top of Claude Code, Cursor, Codex or Copilot and does nothing without them
- The task is small and one-off: plain solo is enough, and /man with a plan and review is overkill
- You work in Windsurf, Cline or Roo Code: their adapters are on the author's roadmap
- You need a guarantee that a log holds no sensitive data: redaction rules rely on formats, so an empty scan proves nothing
Example request
/man add CSV export for the report: study the project, propose two options and show the plan, do not write code yetLimitations
- Requires Node.js 22.5.0 or newer. Git is optional: without it the team mode falls back to solo automatically.
- The author still treats the ZCode, Kimi Code, Qoder and DeepSeek Harness adapters as provisional: the skill discovery path in those hosts is being verified.
- The secrets vault relies on the macOS keychain. The author has not finished testing a locked keychain or a denied access prompt.
- The license is AGPL-3.0; check its terms separately before embedding in a closed product.
- The main documentation is in Chinese; an English README sits next to it.
How to disable. Delete the .mancode folder and the files init created: the man commands under .claude/skills, the mancode block in AGENTS.md, .cursor/rules and .agents/skills. npm uninstall -g mancode removes the package itself. Look through .mancode first: it holds your plans and verification results.
FAQ
How do I install mancode in Claude Code?
Run npm install -g mancode, go to the project folder and run mancode init --platform claude-code. After Claude Code restarts, the project has the /man, /manba, /manteam, /manps and /mansolo commands. Hooks are not needed by default; only the --legacy flag installs the old hook-based setup.
How is mancode different from CLAUDE.md?
CLAUDE.md holds static instructions. mancode leaves only a short bootstrap block in it, while sessions, plans, requirements and verification evidence live in the structured .mancode folder. On top of that come the modes, persistent workflows and separate review subagents.
How do I call /man in Codex?
Codex reads skills from .agents/skills, so the modes are invoked with a dollar sign: $man, $manba, $manteam. In the ChatGPT desktop app an enabled man skill also shows up in the command list as /man. DeepSeek Harness reads skills from .dsh/skills and calls them as /man.
Can mancode push or deploy on its own?
No. The publication command only reads upstream state and sends nothing. Before a force push, a schema migration or a bulk delete the modes ask for explicit human confirmation.
How do I check a log for secrets before sharing it?
Run mancode privacy scan --file app.log --json: it returns rules, categories and offsets without the values. mancode privacy preview --output app.redacted.log writes a masked copy and leaves the original alone. Exit code 0 means no matches.
Discussion
No questions or reviews yet.
Related
CLI toolsA Python terminal coding agent in the manner of Claude Code: reads the repository, edits files, runs commands, with memory, sub-agents and MCP
A CLI that gives six coding agents project context, over 80 skills, verification and resumable tasks across sessions
MiniCode is a compact TypeScript terminal coding agent like Claude Code: tool loop, TUI, skills, MCP and sub-agents, built for study and tinkering
An Emacs package that gives Codex, Claude Code, Gemini CLI and other agents one shared menu and the same commands
An MCP server and CLI in Go: the agent keeps an index of code and database schema in the repository, versions it in Git and reads it whole before each task
The author's personal set: shared agent rules and dozens of skills for GitHub, the browser, macOS, messaging and releases, plus a sync script
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Repository stars | Main difference |
|---|---|---|---|
| mancode (this entry) | no | 359 | |
| Superpowers | no | 307,845 | a skill set for the development process without a task store or data redaction |
| OpenSpec | no | 73,368 | spec-driven development with proposals, requirements and tasks kept as repository files |
| BMad Method | no | 54,748 | roles and stages from idea to architecture, a different structure from agent modes |
README in short
The main docs are in Chinese and cover setup for eight agents, and an English translation sits in README.en.md. The author separately lists what is still being verified and explains which data stays on the machine.