team-mode

Team Mode

Use this skill when the user explicitly wants a team, squad, multiple agents, or a staged multi-step workflow.

This skill adapts the spirit of OMC team orchestration to Codex:

• no custom runtime
• no fake task system
• real Codex subagents with clear ownership and staged handoffs

Default Squad

Default to a 3-agent structure when the task is large enough to benefit:

• explorer: maps the codebase, conventions, and likely risks
• worker: owns the main implementation slice
• reviewer: runs a second-pass verification or review

The main agent remains the orchestrator and integrator.

Drop to 2 agents for small tasks. Go beyond 3 only when the work can be cleanly partitioned into disjoint slices.

Pipeline

Run work in this order:

plan -> execute -> verify -> fix -> verify

Repeat the execute/fix/verify loop until:

• acceptance checks pass, or
• the work is blocked by a real external dependency, or
• repeated failures show a fundamental issue that should be surfaced

Stage Rules

1. Plan

• Read the critical path locally first.
• If the codebase is unclear, spawn an explorer.
• If the task is large or ambiguous, spawn one default subagent as a planner or architect reviewer.
• Produce a short execution plan with explicit ownership boundaries.

2. Execute

• Keep shared glue or the most blocking slice local.
• Spawn worker agents only for disjoint file or module scopes.
• Give every worker a bounded deliverable and ownership area.

3. Verify

• Review locally first.
• Prefer a dedicated reviewer pass for medium or large tasks.
• Spawn an explorer or default subagent for a second-pass check when risk or scope justifies it.
• Run tests, typecheck, lint, or build commands where available.

4. Fix

• Convert failed checks or review findings into explicit follow-up tasks.
• Fix locally or delegate another bounded patch.
• Re-run verification after changes.

Worker Prompt Expectations

Every worker prompt should say: