ops agent
Multi-agent orchestration. Broadcasts tasks to project master agents (Claude Code instances) in parallel, chains agents sequentially, and manages saved chain configs. Each project's CLAUDE.md makes a spawned agent an instant expert on that project.
ops agent broadcast uses wg-orchestration as the orchestrator agent, which fans out to every other project agent simultaneously. Each spawned worker inherits its project's CLAUDE.md context, so it already knows the codebase before the prompt arrives.
Subcommands
| Subcommand | Description |
|---|---|
broadcast | Broadcast a task to all master agents in parallel |
chain | Chain agents sequentially — each output feeds the next input |
run | Run a saved chain by name |
list-chains | List available chain configs |
list-masters | List all registered master agents |
add-claude-md | Create or update CLAUDE.md in a project |
history | Show recent agent run history |
Capabilities at a Glance
Fan-Out Broadcast
Send the same task to every project agent in parallel — perfect for portfolio-wide audits, sweeps, and summaries.
Sequential Chains
Compose multiple agents into a pipeline where each agent's output becomes the next agent's input.
Project-Native Context
Workers inherit their own project's CLAUDE.md, so they answer like an expert in that codebase — not a generic LLM.
ops agent broadcast
Broadcast a task to every master agent in parallel and collect their responses.
Flags
| Flag | Type | Description | Default |
|---|---|---|---|
--prompt-p | TEXT | Task to send to all master agents | — |
--file-f | PATH | Read task from a file (use instead of --prompt for long prompts) | — |
--project | TEXT | Target specific project IDs (repeatable; default: all) | — |
--exclude-x | TEXT | Exclude specific project IDs (repeatable) | — |
--timeout-t | INTEGER | Per-agent timeout in seconds | 300 |
--workers-w | INTEGER | Max parallel agents | 4 |
--output-dir-o | PATH | Save each agent's output to this directory (one file per project) | — |
--json | flag | Output results as JSON | — |
--dry-run | flag | Show which agents would be contacted without running them | — |
Examples
For multi-paragraph prompts or anything you want to version, write the prompt to a file and pass it with --file. This also keeps shell quoting from mangling the input.
Each broadcast spawns one Claude Code instance per project. Use --workers to cap concurrency and --project to narrow scope when running expensive prompts.
ops agent chain
Chain agents sequentially. Each agent receives the previous agent's output as its input.
ops agent run
Run a saved chain by name. Equivalent to ops agent chain but without typing the full config path.
ops agent list-chains
List every available chain config, with each chain's description and the agents it invokes.
ops agent list-masters
List every registered master agent and whether it is ready to receive work.
A master is considered ready when its project directory exists, its CLAUDE.md is present, and the Claude Code CLI can launch in that directory.
ops agent add-claude-md
Create or update a project's CLAUDE.md, then commit and push the change. Use this when bootstrapping a new project agent or refreshing context after architectural changes.
ops agent history
Show recent agent run history — broadcasts, chains, and one-off runs.