agent-team-workflows

What This Skill Does

The Agent Team Workflows skill is a robust orchestration framework designed to supercharge your productivity by deploying specialized, multi-agent teams within the OpenClaw/Claude Code ecosystem. Instead of a single AI assistant tackling a complex task, this skill manages a structured hierarchy consisting of one Lead and four specialized Teammates (Architect, Developer, Tester, and Reviewer). It excels at decomposing massive, multi-faceted projects into manageable, high-quality deliverables by applying specific control-flow patterns, such as sequential pipelines, parallel-merge, iterative-review, and fan-out/fan-in map-reduce strategies.

Installation

To install this skill, run the following command in your terminal: clawhub install openclaw/skills/skills/doanbactam/claude-agent-team-workflows

Ensure you have enabled experimental agent teams in your local configuration by adding the following to your ~/.claude/settings.json file:

{
  "env": {
    "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
  }
}

Use Cases

• Software Engineering: Building complex features where architectural design, implementation, unit testing, and rigorous code reviews must occur in sync.
• Content Creation: Managing long-form writing projects involving research, drafting, rigorous fact-checking, and editorial oversight.
• Business Strategy: Synthesizing complex market data by splitting tasks into strategy planning, business analysis, financial modeling, and risk assessment.
• Data Science: Coordinating the lifecycle of a data project, from initial hypothesis and data cleaning to statistical validation and peer-reviewed modeling.

Example Prompts

1. "Run a parallel-merge workflow to research the impact of AI on green energy, assigning the architect to map sources, and the other teammates to summarize, verify data, and critique the final impact report."
2. "Start an iterative-review workflow to draft a technical proposal for our new cloud architecture. Make sure the builder and critic loop at least twice before moving to the final validator gate."
3. "Coordinate a fan-out-fan-in workflow to audit our codebase security. Use the planner to identify top risk areas, then have the team scan separate modules before merging findings into a final report."

Tips & Limitations

• Role Mapping: Remember that your domain context matters. Use the domain-specific Role Cards to ensure teammates act according to the relevant context (e.g., in research, the 'Developer' role acts as a Researcher).
• Iteration Control: For iterative-review patterns, the default maximum is 2 rounds of critique to manage token usage and latency. You must provide manual approval for deeper, multi-round loops.
• Token Costs: Managing 5 agents simultaneously increases token consumption significantly compared to single-agent interactions. Monitor your usage during complex fan-out operations.
• Pre-requisites: This skill requires an active internet connection to coordinate agent teams and necessitates that the experimental feature flag is set correctly in your environment settings.