anywhere-agents
One config for every agent — Claude Code, Codex, and whatever comes next
How it works Start with effective defaults. Add packs — small bundles of rules, skills, or permissions — as you need them.
Config sources
agent-style
Rule pack · writing defaults + banned-word guard
ships by default
aa-core-skills
Skill pack · review, router, figures, README polish
ships by default
Your pack selections
Which packs to add or remove, per project
agent-config.yaml · agent-config.local.yaml
AGENT_CONFIG_PACKS env
Your user-level picks
CLI writes today · bootstrap reads in v0.4.x
v0.4.x next
anywhere-agents composer
bootstrap
One source of truth
Every time you bootstrap a project, the composer reads your picks, resolves versions, and writes each piece to where the agent expects it.
1Read shipped defaults + your config
2Compose rules, skills, hooks, permissions
3Write to the right place on disk
Staged writes today · lock-recovery wiring in v0.4.x · CLI anywhere-agents pack add | remove | list
What shows up on disk
Rules for every agent
Writing defaults and project conventions, merged
→ AGENTS.md (+ CLAUDE.md, agents/codex.md)
Skills
/implement-review, /readme-polish, and friends
→ .claude/skills/
Slash commands
Shortcuts the agent sees when you type /
→ .claude/commands/
Safety hooks
Guard against git push, unsafe edits, banned words
→ ~/.claude/hooks/
Permissions & env
Shared allow/deny rules, effort level, hook wiring
→ ~/.claude/settings.json
Agents
Claude Code
Reads rules + runs hooks
Codex
Reads rules from agents/codex.md
…and others
Any agent that reads AGENTS.md
pip install anywhere-agents · github.com/yzhao062/anywhere-agents · macOS · Linux · Windows