Skip to content

agent-team-orchestration.md

Latest commit

 

History

History
125 lines (96 loc) · 3.32 KB

agent-team-orchestration.md

File metadata and controls

125 lines (96 loc) · 3.32 KB
Raw

Agent & Team Orchestration

ClawCode provides a rich multi-agent system, from simple subagents to full R&D team orchestration.

Subagent System

The main agent can spawn subagents with isolated context, custom prompts, and tool allowlists.

Built-in Subagent Roles

Agent ID Purpose Tool Access
explore Read-only exploration Read, Glob, Grep, ...
plan Research for planning Read-only tools
code-review Review-focused Read-only tools
general-purpose Full tool surface All (minus delegate tools)

Invoking a Subagent

{
  "agent": "plan",
  "task": "Map how authentication is implemented; list key files."
}

Aliases: subagent_type ↔ agent, prompt ↔ task.

ClawTeam: Multi-Role Orchestration

/clawteam orchestrates a virtual R&D team with 14+ professional roles.

Role Registry

Role ID Responsibility
product-manager Priorities, roadmap, acceptance criteria
business-analyst Process, rules, edge cases
system-architect Architecture, tech choices, NFRs
ui-ux-designer Information architecture, UX constraints
dev-manager Rhythm, risks, milestones
team-lead Technical decisions, quality bar
rnd-backend Services, APIs, data layer
rnd-frontend UI components, state, integration
rnd-mobile Mobile/cross-platform development
devops CI/CD, pipelines, environments
qa Test strategy, gates, regression
sre Availability, SLOs, runbooks
project-manager Scope, schedule, change control
scrum-master Iteration rhythm, blockers

Usage

/clawteam <your ask>
/clawteam:<role> <specific task>
/clawteam --deep_loop <complex task>

Deep Loop: Convergent Iteration

/clawteam --deep_loop runs multiple converging rounds:

  1. Structured contract per round (goals, handoffs, gaps)
  2. Parse DEEP_LOOP_WRITEBACK_JSON for automated write-back
  3. Tunable convergence thresholds, max iterations
  4. Consistency checks across rounds

Convergence settings in .clawcode.json:

{
  "closed_loop": {
    "clawteam_deeploop_enabled": true,
    "clawteam_deeploop_max_iters": 100,
    "clawteam_deeploop_convergence_rounds": 2,
    "clawteam_deeploop_handoff_target": 0.85
  }
}

Custom Agent Roles

Define custom roles as Markdown files with YAML frontmatter:

---
name: api-guardian
description: Reviews public HTTP API changes only.
tools:
  - Read
  - Glob
  - Grep
  - diagnostics
maxTurns: 24
---

You only analyze API routes and OpenAPI/contract files.
Report breaking changes as a bullet list.

Discovery Paths

Scope Path
User-wide ~/.claude/agents/*.md
Project .claw/agents/*.md, .clawcode/agents/*.md, .claude/agents/*.md

Plan Mode

In plan mode (/plan), only these subagents are allowed:

  • plan
  • explore
  • code-review

All tools are restricted to read-only policy.

Multi-Model Workflows

Command Focus
/multi-plan Collaborative planning (plan-only)
/multi-execute Collaborative execution with traceable artifacts
/multi-backend Backend-focused workflow
/multi-frontend Frontend-focused workflow
/multi-workflow Full-stack workflow (backend + UI)