Skip to main content

Agent Teams

Available in Rigour v2.14+ | Requires Opus 4.6/GPT-5.3+ frontier models

Agent Teams enable Rigour to supervise multiple AI agents working together on the same codebase—a capability unlocked by frontier models like Claude Opus 4.6 (agent teams) and GPT-5.3-Codex (coworking mode).

Why Agent Team Governance?​

Traditional Rigour supervision assumes a single agent making changes. With frontier models now supporting:

  • Agent Teams (Opus 4.6): Multiple agents collaborating with specialized roles
  • Coworking Mode (GPT-5.3): Real-time steering during long-running tasks

New failure modes emerge:

  • Cross-agent pattern conflicts
  • Task ownership violations
  • Handoff context loss
  • Duplicate work detection failures

Configuration​

# rigour.yml
gates:
  agent_team:
    enabled: true
    max_concurrent_agents: 3
    cross_agent_pattern_check: true
    handoff_verification: true
    task_ownership: strict  # or "collaborative"
OptionDefaultDescription
enabledfalseEnable multi-agent governance
max_concurrent_agents3Max agents in a session
cross_agent_pattern_checktrueDetect conflicting patterns
handoff_verificationtrueVerify context at handoffs
task_ownershipstrictstrict = 1 agent per file, collaborative = allow overlap

MCP Integration​

Register agents via MCP:

// Agent A registers
await mcp.call("rigour_agent_register", {
  cwd: "/project",
  agent_id: "agent-a",
  task_scope: ["src/api/**"]
});

// Agent B registers
await mcp.call("rigour_agent_register", {
  cwd: "/project",
  agent_id: "agent-b", 
  task_scope: ["src/ui/**"]
});

Task Ownership Models​

Strict Mode (Default)​

Each agent owns specific files/directories. Rigour blocks:

  • Any agent editing files outside their scope
  • Multiple agents claiming the same scope

Best for: Parallel development, clear separation of concerns

Collaborative Mode​

Agents can overlap with conflict resolution:

  • Rigour detects conflicting changes
  • Prompts for human arbitration via Studio

Best for: Pair programming, complex refactoring

Cross-Agent Pattern Detection​

Rigour detects when agents introduce conflicting patterns:

⛔ AGENT TEAM CONFLICT DETECTED

Agent A (src/api/users.ts):
  - Using: camelCase naming
  - Pattern: async/await

Agent B (src/api/orders.ts):
  - Using: snake_case naming  ← CONFLICT
  - Pattern: async/await      ✓ OK

See Also​