A-Team — Role-Based Project Team Coordination

Overview

Inspired by a.team: don't send solo agents — form a project team with specific roles. A PM coordinates, engineers execute, specialists advise, reviewers gate quality. The team has continuity across the project lifecycle.

Roles

Coordination Flow

Phase 1: PM Breakdown
  ┌─────────┐
  │   PM    │  reads brief → task list + role assignments
  └────┬────┘
       │
Phase 2: Parallel Execution
  ┌────┴────┬────────────┐
  │ Engineer │ Specialist │  work assigned tasks in parallel
  └────┬────┴─────┬──────┘
       │          │
Phase 3: Review Gate
  ┌────┴──────────┴──┐
  │     Reviewer     │  checks all output, flags issues
  └────────┬─────────┘
           │
Phase 4: PM Synthesis
  ┌────────┴─────────┐
  │   PM (final)     │  integrates, writes summary/ship report
  └──────────────────┘

Each phase runs explicitly via --phase 1|2|3|4 so you see output between phases.

Quick Start

# Phase 1: PM breaks down the project
nu ~/.claude/skills/a-team/a-team.nu --phase 1 \
  --brief "Deploy gastown rate limiter to Pi fleet"

# Phase 2: Engineer + Specialist execute in parallel panes
nu ~/.claude/skills/a-team/a-team.nu --phase 2 \
  --brief "Deploy gastown rate limiter to Pi fleet" \
  --specialist "ansible"

# Phase 3: Reviewer checks output
nu ~/.claude/skills/a-team/a-team.nu --phase 3 \
  --brief "Deploy gastown rate limiter to Pi fleet"

# Dry run any phase
nu ~/.claude/skills/a-team/a-team.nu --dry-run --phase 1 \
  --brief "Deploy gastown rate limiter to Pi fleet"

When to Use: a-team vs tiger-team vs zellij-team

Agent Flags

All agents run with --dangerously-skip-permissions --permission-mode bypassPermissions by default. Override with --mode default for interactive permission prompts.

Output Convention

Each phase writes output to /tmp/a-team-<brief-hash>/:

• phase1-pm-plan.md — PM's task breakdown
• phase2-engineer.md — Engineer's output
• phase2-specialist.md — Specialist's output
• phase3-review.md — Reviewer's findings
• phase4-summary.md — PM's final synthesis

Verification contract (auto-injected by proveryay)

Every agent dispatched through this skill receives the following appended to its prompt automatically:

<see ~/.claude/skills/proveryay/CLAIMS-TEMPLATE.md — injected verbatim>

This means every agent MUST emit a <claims> block when it makes any deployment/state-change assertion. The proveryay SubagentStop hook will verify each claim in parallel and inject structured verdicts into the dispatcher's next-turn context.

If the agent's work is purely planning / research / narrative (no state changes), the agent may omit the <claims> block. Proveryay treats absence as silent (no noise).

Invocation example with template injection:

# Prompt injection: template auto-appended (idempotent)
let tpl = (open --raw ~/.claude/skills/proveryay/CLAIMS-TEMPLATE.md)
let prompt = if ($prompt | str contains "<claims>") { $prompt } else { $"($prompt)\n\n($tpl)" }

See:

• ~/.claude/skills/proveryay/SKILL.md — hook behavior
• ~/.claude/skills/proveryay/CLAIMS-TEMPLATE.md — canonical format
• ~/.claude/skills/fleet-eval/SKILL.md — probe taxonomy