anywhere-agents CLI
One config for Claude Code and Codex in every repo: shared rules, skill packs, second-model review and a destructive-command guard
- 261 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- Hooks modify ~/.claude/settings.json and ~/.claude/hooks/, and SessionStart runs bootstrap every time a session opens
- Bootstrap pulls rules, skills and packs from external repositories, private ones included through your Git credentials
- Automatic /vet mode launches a second model CLI with broad permissions; manual relay remains a separate option
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/anywhere-agentsDetects 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.
- Hooks modify ~/.claude/settings.json and ~/.claude/hooks/, and SessionStart runs bootstrap every time a session opens
- Bootstrap pulls rules, skills and packs from external repositories, private ones included through your Git credentials
- Automatic /vet mode launches a second model CLI with broad permissions; manual relay remains a separate option
This is third-party code. Review the repository files before installing.
What it does
anywhere-agents is a command-line loader for a shared agent configuration. One AGENTS.md drives Claude Code, Codex and other agents in every repo on every machine. The bare `anywhere-agents` command does it all: pulls shared rules, deploys skills, hooks and settings, and regenerates CLAUDE.md and agents/codex.md from AGENTS.md. A pack is added with `anywhere-agents pack add <url>`, which fetches, installs and deploys it in one step. A pack is a small bundle of rules, skills or permissions. Two ship by default: agent-style and aa-core-skills. The first bans about 45 AI-tell words: a PreToolUse hook blocks writes to .md, .tex and .rst files that contain one. The second holds six skills, among them implement-review with the short alias `/vet`, which hands your diff to a second reviewer from another model family. Codex is the default, and `/vet agy` picks Gemini through the Antigravity CLI. The guard.py hook stops `git push`, `git commit`, `git merge`, `git rebase`, `git reset --hard`, `gh pr merge` and similar commands and asks for confirmation. Read-only calls such as `git status` and `git diff` pass silently. In Claude Code a SessionStart hook runs bootstrap each time you open a session, and a row at the bottom of the terminal shows remaining Claude, Codex and Gemini quota.
Who it is for. For developers and researchers who use several agents and repositories and want the same rules, second-model review and protection from destructive git commands.
Good fit when
- You use Claude Code at work and Codex at home and do not want two configurations to maintain by hand
- Every new repository repeats the same setup of writing rules, permissions and skills
- Before a push you want a second reviewer to look at the changes through `/vet` and catch what the first model missed
- You need the agent stopped before `git push --force` and other destructive commands and want to confirm them by hand
- You write articles and papers and AI-tell words in .md and .tex files should be blocked by a hook
Not a fit when
- You need a config for a single repository and a single agent: your own AGENTS.md is enough
- You do not want a script that pulls rules from an external repository to run at every session start
- You need to manage rule packs through the npm version without Python: pack commands exist only in the Python version, while the npm package only supports `bootstrap`
- You plan PowerPoint figure export on Linux: editable-figure needs desktop PowerPoint and its export script is Windows-only
Example request
Install anywhere-agents in this project and show which packs and hooks it addedLimitations
The Python package requires Python 3.9 or later and provides package-management commands. The npm variant performs bootstrap setup. Running it changes project instructions and global Claude Code settings, including hooks, so back up the current configuration first. Locked update policy and pinned versions limit automatic rule changes.
Review through /vet supports a manual handoff or automatic launch of a second CLI. The opt-in automatic path receives broad permissions to execute commands, access the network and write files. Restricting it to review is a textual instruction. Use a separate project copy and agree on sending the diff to the second model provider.
How to disable. Run pipx uninstall anywhere-agents or npm uninstall -g anywhere-agents. The guard.py and session_bootstrap.py hooks live in ~/.claude/hooks/ and shared keys were merged into ~/.claude/settings.json: remove them there. In a repo delete AGENTS.md, CLAUDE.md and .agent-config/ if unneeded.
FAQ
How do I install anywhere-agents?
Run pipx install anywhere-agents or uv tool install anywhere-agents, then run anywhere-agents in the project root. You can also skip the install with pipx run anywhere-agents or npx anywhere-agents. The author also suggests simply asking your agent: "Install anywhere-agents in this project".
Why pipx and not pip?
Plain pip installs into the active project environment or hits the externally-managed-environment error on modern Ubuntu, Debian and Homebrew Python. Pipx gives the CLI its own environment and exposes the command globally.
What does /vet do?
It is a short alias for implement-review. It hands your staged diff to a second reviewer from another model family. Codex runs without arguments, /vet agy selects Gemini, and Copilot and headless Claude Code are also available.
What does the guard hook do with git commands?
The hook checks a set of Git write commands and requests approval while allowing status and history reads. Its filter does not guarantee protection against every way to change a repository; the agent process permissions still matter.
How do I add my own rule pack?
Run anywhere-agents pack add with a repository URL, for example --ref v0.1.0 to pin a version. The agent-pack repository serves as the blueprint: fork it, replace the content and tag a release. Private repos use the chain SSH agent, gh CLI, GITHUB_TOKEN.
Discussion
No questions or reviews yet.
Related
CLI toolsA plugin and kit for Claude Code and Codex: agent working principles, safety hooks, skills and CLAUDE.md templates in one repository
A skill for messaging between Claude Code, Codex, Gemini CLI and other CLI agents through a shared local SQLite database, written in bash
Agentic Substrate
Agentic Substrate (claude-user-memory)
A short reversible rule set for an agent: research the unclear, plan proportionally, verify the outcome and keep checked knowledge
oah links 80-plus skills, five roles and /commit, /review, /test-tdd commands into Claude Code, Cursor, Codex and other agent folders
A CLI for AI agents sharing a repository: a task board, a spec and acceptance by a second agent against a specific commit
A bridge between local CLI agents: Codex, Cursor or Claude Code delegate work to Antigravity, Grok, Kimi and seven more.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| anywhere-agents (this card) | no | 261 | |
| ZCF | no | 6,071 | sets up Claude Code and Codex in one run, without shared rule packs and a git confirmation hook |
| Agent Scripts | no | 7,463 | one author's personal set of shared rules and skills |
README in short
The collection uses Apache 2.0. Its optional editable-figure skill requires desktop PowerPoint; the export script targets Windows, independently of the main CLI’s platform support.