A skill is a SKILL.md file that gets installed into your AI agent. It teaches the agent how to use Box0: when to delegate, what commands to run, and how to write good prompts.
npx skills add risingwavelabs/skills --skill b0This uses npx skills to install the Box0 skill. You only need to do this once per machine.
Claude Code: writes ~/.claude/skills/b0/SKILL.md with YAML frontmatter and markdown instructions. Claude Code automatically discovers skill files in this directory and loads them when relevant.
The frontmatter includes:
---
name: b0
description: |
Delegate tasks to AI agents via Box0. Use when the user asks to
review code, check security, run tests, compare tools, get multiple
perspectives, research a topic, analyze data, write docs, or any
task that could benefit from specialized or parallel execution.
Also use when the user mentions agent names or says "ask", "delegate",
"get opinions from", or "have someone".
allowed-tools:
- Bash
---Claude uses the description field to decide when to auto-load the skill. allowed-tools: Bash lets the agent run b0 commands without permission prompts.
Codex: writes instructions to ~/.codex/AGENTS.md. Codex reads AGENTS.md on startup as custom instructions.
The skill body teaches the agent:
- Discover agents: run
b0 agent ls, match agents to the task by description - Write detailed prompts: gather context first (read files, run
git diff), include specifics, state the deliverable. Never just forward the user's words. - Parallel execution: delegate to multiple agents, then
b0 wait --all - Pipe large content:
git diff | b0 delegate reviewer "Review this diff." - Handle questions: answer with
b0 reply, thenb0 waitagain - Multi-turn: use
--threadto continue conversations - Cron jobs: schedule recurring tasks with
b0 cron add - Proactive status: run
b0 statusbefore responding to check for completed results - Error handling: retry, try a different agent, handle it yourself, or report to user
The full skill content is in SKILL.md at the repository root.