Guide · Agent skill
How to use the Coding Agent Delegation agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The coding-agent skill teaches an AI agent to hand larger coding jobs to another coding tool, such as OpenAI Codex, Claude Code, Pi or OpenCode, instead of editing code itself. It is aimed at developers who already use one of those command-line tools and want their agent to start, monitor and stop them. After this guide you will know how to install the skill, how to phrase requests that trigger it, and which safety settings to check before running it.
Coding Agent Delegationspinabot/brigade · MIT Download the skill Quick start
- 1Install a coding tool. Install at least one of claude, codex, opencode or pi, and sign in with an account or API credentials for it.
- 2Install the skill. Download the ZIP from this page and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder, or upload it under Settings → Capabilities → Skills.
- 3Pick a project folder. Choose a folder that contains only the code the task needs, and never use the ~/brigade workspace.
- 4Ask for a delegated job. Describe a feature, PR review or refactor and name the project folder.
- 5Run in the background. For longer work, start the tool with background mode and the correct pty setting for that tool.
- 6Monitor and answer prompts. Use poll or log to check progress, and submit or write to answer questions the agent asks.
- 7Review and clean up. Check the result yourself, post PR comments with gh if needed, and delete any temporary clone.
What does the coding-agent skill make the agent do?
The skill tells the agent to delegate coding work to Codex, Claude Code, Pi or OpenCode. The agent starts one of these tools in a chosen project folder, optionally in the background, then checks its progress, answers its prompts, or stops it.
The skill is for bigger jobs: building new features or apps, reviewing pull requests, refactoring large codebases, and iterative coding that needs file exploration. It says not to use it for simple one-liner fixes, for reading code, or for any work in the ~/brigade workspace.
- Building or creating new features or apps
- Reviewing PRs, with the agent working in a temporary directory
- Refactoring large codebases
- Iterative coding that needs file exploration
How do you install the coding-agent skill?
Download the ZIP from this page and unzip it into your personal skills folder, ~/.claude/skills/, or into a project's .claude/skills/ folder. Alternatively, in the Claude apps, open Settings, then Capabilities, then Skills, and upload the skill there.
The skill itself ships as a single file, SKILL.md. The file's metadata lists two npm installs you may need, depending on which tool you use: the Claude Code CLI (@anthropic-ai/claude-code) and the Codex CLI (@openai/codex).
- Personal install: unzip into ~/.claude/skills/
- Project install: unzip into the project's .claude/skills/
- Claude apps: Settings → Capabilities → Skills, then upload
Which coding tools do you need installed?
The skill needs at least one of the claude, codex, opencode or pi command-line tools on your machine. You also need an account or API credentials for whichever coding agent you pick; the source does not specify what those are. For pull request workflows you also need git and the gh GitHub CLI.
To install the Claude Code or Codex CLIs, the skill metadata points to npm packages. Pi's install line in the source is npm install -g @earendil-works/pi-coding-agent.
- One or more of: claude, codex, opencode, pi
- npm (for the Claude Code or Codex CLIs)
- Account or API credentials for your chosen agent
- git and gh for PR review
npm install -g @earendil-works/pi-coding-agentHow do you trigger the skill with a natural request?
You do not need special syntax. Ask the agent for a larger coding job and make clear which project it applies to. The skill's description lists the situations it covers, so requests that match them are the ones that should route to it.
Requests that fit: asking for a new feature or app, asking for a review of a pull request, asking for a large refactor, or asking for iterative work across many files. Requests that do not fit: a one-line typo fix, or a question about what a function does.
- "Build a dark mode toggle in ~/Projects/myproject."
- "Review PR #86 in a temporary clone."
- "Refactor the auth module across the codebase."
How do you run a coding tool safely in the background?
The skill's core pattern is to set a working directory, run the tool in background mode, then poll it. Codex, Pi and OpenCode need a pseudo-terminal (pty:true). Claude Code should not use a PTY; the skill says to use --print with --permission-mode bypassPermissions instead.
Codex refuses to run outside a trusted git directory, so for scratch work the skill creates a temporary git repo first. Monitoring uses process actions such as list, poll, log, submit and kill.
- Set workdir so the agent sees only that folder
- Use background:true for longer tasks
- Check progress with process action:log or action:poll
- Answer prompts with process action:submit, or stop with action:kill
bash pty:true workdir:~/project background:true command:"codex exec --full-auto 'Build a snake game'"What are safe ways to review pull requests?
The skill warns never to review PRs in Brigade's own project folder. Instead, clone the repo to a temporary folder or use a git worktree. The source also shows a batch pattern that fetches PR refs and runs one Codex review per PR in the background.
When finished, the skill suggests cleaning up the temporary clone. Results can be posted back with the gh CLI.
- Clone to a temp folder with mktemp -d, or create a git worktree
- Fetch PR refs before a batch review
- Post a review with gh pr comment
git worktree add /tmp/pr-130-review pr-130-branchWhat safety settings should you check first?
This skill runs scripts and uses the network. Several recommended modes skip safeguards. The source recommends --yolo for Codex, which disables the sandbox and approvals, and bypassPermissions for Claude Code, which skips permission prompts. Under these modes code can change or run without confirmation.
Agents may also read unrelated files in the directory they are given, so do not point the skill at a sensitive or personal folder. Prefer --full-auto for Codex, which keeps a sandbox, when you can.
- --yolo (Codex): no sandbox, no approvals
- --full-auto (Codex): sandboxed, auto-approves inside the workspace
- bypassPermissions (Claude Code): skips permission prompts
- Avoid sensitive or personal folders
What people use it for
- Starting a new feature in a project folder with Codex in the background
- Reviewing several open pull requests in parallel, each in its own temporary checkout
- Running a large refactor with Claude Code in print mode
- Using a Pi session for a one-off summary of a src/ folder
Frequently asked questions
How do I install the coding-agent skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder. In the Claude apps you can instead upload it under Settings → Capabilities → Skills.
What do I need to use the coding-agent skill?
You need at least one of the claude, codex, opencode or pi command-line tools, plus an account or API credentials for that agent. For pull request reviews you also need git and the gh GitHub CLI. The skill metadata also references npm for installing the Claude Code and Codex CLIs.
Is the coding-agent skill free?
The skill is published under the MIT licence. The source does not state prices for the coding tools or their accounts, so check those providers' terms directly.
Why does Claude Code not use a PTY?
The skill says that Claude Code with --dangerously-skip-permissions can exit after the confirmation dialog when run with PTY. It recommends --print with --permission-mode bypassPermissions instead, which keeps full tool access without interactive confirmation.
Why does Codex fail with a message about git?
Codex refuses to run outside a trusted git directory. The skill's fix for scratch work is to create a temporary repo with git init before running codex exec.
Is it safe to let the agent review PRs in my main project?
The skill says never to review PRs in Brigade's own project folder. Use a temporary clone or a git worktree so your main checkout stays intact.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
This skill suits developers who already use Codex, Claude Code, Pi or OpenCode and want to hand off feature builds, refactors or batch PR reviews from a shell-based agent. It is not worth it for one-line fixes, and it is a poor fit if you cannot accept tools running with permissions skipped or if the target folder holds sensitive files. In those cases, edit directly or use the coding tool in a supervised session instead.
Download the skill