Team Agent Settings workflow
Shared Claude Code settings for a team: hooks against dangerous commands, a check before finishing and skills from spec to review
- 92 ★ on the repository
- Last commit
Broad access
Why this level
- Hooks run shell commands on every agent action, and the install copies files over `~/.claude` and `~/.codex`.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add workflows/team-agent-settingsDetects 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 run shell commands on every agent action, and the install copies files over `~/.claude` and `~/.codex`.
This is third-party code. Review the repository files before installing.
What it does
Team Agent Settings is a ready `~/.claude` folder for a team that works in Claude Code and calls Codex from it as an executor. The author describes a workflow: rate the weight of a task first (levels E, 0, 1, 2, R) and record it with `harness-log triage`, research with `research`, `how` and `why`, design with `spec` using the verification marks `[e2e]`, `[api]`, `[test]`, `[check]`, `[human]`, then implement and hand the diff to the `code-reviewer` skill. A human does the merge. Hooks stop `rm -rf` and similar commands, direct commits to main, chains of several git commands, a package manager that contradicts the lock file and edits to `.env`. A Stop hook runs `make check` up to three times, checks the evidence file and sends the agent back to work if the check fails. Another hook asks to rewrite an English reply in Japanese. The set includes the skills `playbook`, `research`, `spec`, `spike-rebuild`, `codex`, `code-reviewer`, plus manual `/harness-init`, `/create-e2e-ci` and `/create-preview-env`. For Codex there is a minimal pair of `AGENTS.md` and `config.toml` with a hook that blocks `sudo`, environment variable dumps and `.env` reads.
Who it is for. Team leads and development teams who want shared rules, hooks and review order for everyone who lets Claude Code into working repositories.
Good fit when
- You want to forbid `rm -rf`, commits to main and `.env` edits at hook level instead of asking in a prompt
- A task is not done until `make check` passes and an evidence file appears in `.tmp/evidence`
- The team wants one scheme for bugs, features, refactoring and research: `playbook` gives step-by-step procedures
- Implementation is handed to Codex while diff and test verification stay with Claude Code
- A new repository needs `make check`, CI and CLAUDE.md: the manual `/harness-init` skill handles it
Not a fit when
- You work alone in a personal setup where shared rules and mandatory checks only get in the way
- The project has no `make` and nothing to run as the `make check` target
- You need rules in a language other than Japanese: skill texts, prompts and the reply language check are written for a Japanese team
Example request
Use playbook to fix this bug: rate the task level, record it with harness-log, prepare a spec and hand the result to code-reviewer.Limitations
Documentation and many prompts are written in Japanese. The set pulls external tools: `uv`, Python 3, Node.js 24, `jq`, `make`, `shasum`, `git`, `gh`, `codex` and the 1Password CLI. Files are copied over `~/.claude` and `~/.codex`, so the author advises a backup first and merging `settings.json` and `config.toml` by hand. `rsync` does not delete old files on update. The Codex hook only matches the command string, so it does not stop `cat .env`. The Stop hook runs `make check` on every agent stop, and the reply language check can be irritating in English-language projects.
How to disable. Restore the backup of `~/.claude` and `~/.codex` made before installing, or delete the copied files and the `hooks` entries from `settings.json`. Old files left after an update are removed by hand.
FAQ
How do I install the settings?
Clone the repository, back up `~/.claude` and run `rsync -a --exclude settings.json claude/ ~/.claude/`. `settings.json` is excluded so personal settings stay intact: merge the shared parts with `jq`.
What does the Stop hook do?
It runs `make check` and asks the agent to fix failures, up to three attempts. A second hook checks that the evidence file in `.tmp/evidence` is fresh when behavior changed.
What is the `codex` skill for?
It hands work to Codex CLI with an explicit model and reasoning level. The author suggests asking in words like "implement with sol" while Claude Code verifies the diff and test results itself.
Which task levels exist?
E, 0, 1, 2 and R. Migrations, authorization, public APIs, data deletion, billing and production releases count as level 2 and above.
Does the set work with Cursor or other agents?
No. Hooks and skills are written for Claude Code, while Codex joins as an executor and receives minimal rules and a protective hook.
Discussion
No questions or reviews yet.
Related
WorkflowsOne-command Claude Code environment installer: a wizard sets up agents, rules, commands, skills and hooks in three profiles
Claude Code Blueprint: 12 agents, 19 skills, 16 hooks and 6 CLAUDE.md rules, copy what you need or install a preset
A ready-made Claude Code setup: agents, slash commands, skills, safety hooks, rules and MCP connections in one install
A CLI, hooks and skills for long Claude, Codex and Pi work: the issue and PR own the task and review gets a PR with checks
A Claude Code working configuration: a planner with per-milestone review, sub-agents and skills for analysis, refactoring and docs sync
A starter kit for Claude Code: a CLAUDE.md memory-bank template with ready hooks, subagents and skills you copy into your project
README in short
The README describes the workflow from weight rating to review, the `claude/` and `codex/` folder layout, required tools, the copy procedure and tables of hooks and skills.