b0（乙 0）

Box0 (b0) Multi-Agent Platform

Run AI agents in parallel on one machine or many. Delegate tasks, collect results, schedule cron jobs.

Setup

Step 1: Check if Box0 is installed

b0 --version

If this succeeds, skip to Step 3.

Step 2: Install

npm install -g @box0/cli@latest

If npm is not available, build from source:

git clone https://github.com/risingwavelabs/box0.git
cd box0 && cargo build --release
export PATH="$PWD/target/release:$PATH"

Step 3: Check if server is running

b0 status

If this shows "Status: connected", skip to Step 5.

Step 4: Connect to a server

Option A: Start a local server (self-hosted)

Run in a separate terminal or background process:

b0 server

On first start, Box0 creates an admin account and auto-configures ~/.b0/config.toml.

Option B: Connect to a remote server (cloud/team)

If the user already has a remote Box0 server, log in instead of starting a local one:

b0 login <server-url> --key <api-key>

The user must provide the server URL and API key. After login, continue to Step 5.

Step 5: Register this machine (remote servers only)

If b0 status shows a remote server (not localhost or 127.0.0.1), check if this machine is registered:

b0 machine ls

If no machines are listed, or if agent creation later fails with "no local machine", register this machine. Read server_url and api_key from ~/.b0/config.toml, then run:

b0 machine join <server-url> --name <hostname> --key <api-key>

Replace <server-url> with the server URL, <hostname> with this machine's hostname, and <api-key> with the API key from the config file.

If the server is local (localhost or 127.0.0.1), skip this step - the local machine is registered automatically.

Step 6: Install the skill

npx skills add risingwavelabs/skills --skill b0

Step 7: Verify

b0 agent ls

This should run without errors. Setup is complete.

Tell the user: "Box0 is installed and ready. You can now delegate tasks to agents."

When to use

When the user's request could benefit from specialized agents or parallel execution, delegate.

Choosing an agent

Always use temp agents unless the user explicitly names an existing agent. b0 agent temp "<task>" is the default for everything. No setup, no cleanup, no b0 agent add. Even if the user says "find 3 agents" or "use multiple agents", create 3 temp agents with b0 agent temp.

Only use b0 agent add when:

• The user explicitly says "create a permanent agent" or "add an agent that I can reuse"
• Never for one-off tasks, debates, research, reviews, or any task that will be done today

Only use b0 delegate <name> when:

• b0 agent ls shows an existing agent that matches the task
• The user mentions an agent by name ("ask the reviewer")

Commands

b0 agent ls                                           # list available agents
b0 delegate <agent> "<detailed task prompt>"          # send task (non-blocking)
b0 delegate --thread <id> <agent> "<follow-up>"       # continue conversation
b0 wait                                                # wait for next completed result
b0 wait --all                                          # wait for all pending results
b0 wait --timeout 0                                    # non-blocking check for completed results
b0 reply <thread-id> "<answer>"                        # answer an agent's question
b0 status                                              # check pending tasks
b0 agent temp "<task>"                                 # one-off task, no named agent
b0 agent add <name> --instructions "..."               # create a named agent
b0 agent remove <name>                                 # delete an agent
b0 cron add --every <interval> "<task>"                # schedule recurring task (auto-creates temp agent)
b0 cron add --agent <name> --every <interval> "<task>" # schedule with existing agent
b0 cron ls                                             # list scheduled tasks
b0 cron remove <id>                                    # remove a scheduled task

How to write delegation prompts

This is critical. Do NOT forward the user's words. Compose a complete, actionable prompt.

Bad:

b0 delegate reviewer "review this PR"

Good:

b0 delegate reviewer "Review the changes on branch feature-timeout in this repo.
The PR adds timeout handling to src/handler.rs.
Focus on correctness, edge cases, and error handling.
Cite line numbers for any issues found."

Steps:

1. Gather context first - read relevant files, run git diff, check the branch
2. Include specifics - file paths, line numbers, branch names, what changed and why
3. State the deliverable - what the agent should produce (a list of issues, a summary, a fix)

For large content (diffs, file contents), pipe via stdin:

git diff main..HEAD | b0 delegate reviewer "Review the following diff. Focus on correctness."

Concurrent tasks

Delegate to multiple agents, then collect all results:

b0 delegate reviewer "Review the changes on branch feature-timeout..."
b0 delegate security "Check src/handler.rs for OWASP top 10 vulnerabilities..."
b0 delegate doc-writer "Update README to reflect the new timeout config option..."
b0 wait --all

All three run in parallel. b0 wait --all blocks until all complete.

Handling agent questions

During b0 wait, an agent may ask a question:

reviewer asks (thread thread-abc): "Is the timeout change on line 42 intentional?"
  -> Use: b0 reply thread-abc "<your answer>"

Answer with b0 reply, then run b0 wait again to continue collecting results.

Proactive status checks

Before responding to a new user message, run b0 status to check if any previously delegated tasks have completed. Report results to the user if any are ready.

Error handling

If an agent fails, b0 wait reports it. Decide whether to:

• Retry with a clearer prompt
• Try a different agent
• Handle the task yourself
• Report the failure to the user

Multi-turn conversations

To continue a conversation with an agent, pass the thread ID from the first round:

b0 delegate --thread <thread-id> <agent> "<follow-up>"
b0 wait

The agent remembers all previous turns.

Troubleshooting