Manus-style context engineering for Claude Code Agent Teams.
Coordinate multiple Claude Code instances with shared planning files, structured task assignment, and persistent working memory. The first skill to apply proven context engineering methodology to multi-agent workflows.
Based on planning-with-files methodology
Click to expand - You MUST enable Agent Teams before using this skill
Without Agent Teams enabled, this skill will fall back to using Task subagents. While subagents work, they cannot use the native Agent Teams features:
- ❌ No peer-to-peer messaging between teammates
- ❌ No Teammate, SendMessage, TaskCreate tools
- ❌ Limited to Task tool (standard subagents)
Option 1: Settings.json (Recommended)
Add this to ~/.claude/settings.json:
{
"env": {
"CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
}
}Option 2: Environment Variable
Add to your shell profile (~/.bashrc, ~/.zshrc, etc.):
export CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1After enabling, restart Claude Code and check that the skill uses "Agent Teams" mode instead of "Task subagents" mode in the team plan.
Claude Code's Agent Teams feature lets you coordinate multiple Claude instances working in parallel. But without proper coordination:
- Teammates forget the overall goal
- Findings get siloed in individual contexts
- Work gets duplicated or conflicts
- Token costs skyrocket with no benefit
Planning with Teams solves this by applying Manus principles to multi-agent coordination:
Single Agent: Context Window = RAM (volatile)
Filesystem = Disk (persistent)
Agent Team: Each Agent = Separate RAM (isolated)
Shared Files = Shared Disk (accessible to ALL)
→ Shared planning files become the team's "collective memory"
- 100% Native Agent Teams — Uses Anthropic's built-in Teammate, SendMessage, TaskCreate tools
- Shared Planning Files —
team_plan.md,team_findings.md,team_progress.md - Manus Methodology — Re-read before decide, log errors, 3-Strike Protocol
- Clear Ownership — File ownership prevents conflicts
- Hooks Integration — Automatic status checks on session end
- Cross-Platform Scripts — Bash, PowerShell, and Python utilities
# In your shell
export CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1
# Or in settings.json
{
"env": {
"CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
}
}# Add marketplace
/plugin marketplace add OthmanAdi/planning-with-teams
# Install plugin
/plugin install planning-with-teamsOr clone manually:
git clone https://github.com/OthmanAdi/planning-with-teams.git ~/.claude/plugins/planning-with-teams/team
Or describe your task:
Create an agent team to review PR #142 for security, performance, and test coverage.
Use planning-with-teams for coordination.
┌─────────────────────────────────────────────────────────┐
│ TEAM LEAD │
│ └─ Creates team_plan.md │
│ └─ Assigns phases to teammates │
│ └─ Synthesizes team_findings.md │
├─────────────────────────────────────────────────────────┤
│ TEAMMATE 1 TEAMMATE 2 TEAMMATE 3 │
│ (Phase 1) (Phase 2) (Phase 3) │
│ │ │ │ │
│ └─────────────────┴─────────────────┘ │
│ │ │
│ SHARED PLANNING FILES │
│ team_plan.md | team_findings.md | team_progress │
└─────────────────────────────────────────────────────────┘
| File | Purpose | Who Updates |
|---|---|---|
team_plan.md |
Shared roadmap, phases, status | Lead + all teammates |
team_findings.md |
Discoveries, errors, decisions | All teammates |
team_progress.md |
Session log, activity tracking | All teammates |
- Read team_plan.md before major decisions — Re-orients to team goal
- Write findings immediately — Don't wait, context is volatile
- Apply 3-Strike Error Protocol — Log failures, don't repeat
- Message lead when phase complete — Coordination
| Command | Autocomplete | Description |
|---|---|---|
/planning-with-teams:plan |
Type /plan |
Start team planning (v1.1.0+) |
/planning-with-teams:status |
Type /status |
Show team progress at a glance (v1.1.0+) |
/planning-with-teams:team |
Type /team |
Guided team setup |
/planning-with-teams:spawn-team |
Type /spawn |
Quick spawn team for a task |
/planning-with-teams:team-status |
— | Get comprehensive status report |
New in v1.1.0:
/planand/statuscommands for consistency with planning-with-files.
Use teams for:
- Parallel code review (security + performance + tests)
- Research with competing hypotheses
- Feature development (frontend + backend + tests)
- Large refactoring (multiple modules)
- Debugging with multiple theories
Don't use teams for:
- Simple single-file edits
- Sequential dependent work
- Tasks under 5 tool calls
- Same-file modifications (conflict risk)
Create an agent team to review PR #142:
Teammate 1 (security-reviewer):
- Focus on security vulnerabilities
- Write to team_findings.md under "## Security Findings"
Teammate 2 (perf-reviewer):
- Focus on performance impact
- Write to team_findings.md under "## Performance Findings"
Teammate 3 (test-reviewer):
- Focus on test coverage
- Write to team_findings.md under "## Test Coverage"
Create an agent team to investigate the connection dropout bug:
Spawn 4 investigators with different hypotheses:
1. WebSocket handling issue
2. Timeout/keepalive misconfiguration
3. State management corruption
4. Memory leak
Have them document evidence FOR and AGAINST in team_findings.md,
challenge each other's findings, and converge on root cause.
See examples.md for more.
| Mode | Cost | When to Use |
|---|---|---|
| Single agent | 1x | Simple tasks |
| Subagents | 1.5-2x | Focused delegation |
| Agent Teams | 3-5x | Complex parallel work |
Agent Teams are expensive. Use them when parallel exploration genuinely adds value.
- Claude Code 2.1.32+
CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1- For split-pane mode: tmux or iTerm2
- SKILL.md — Full methodology
- reference.md — Manus principles adapted for teams
- examples.md — Real use cases
- planning-with-files — Single-agent planning (the foundation)
- Claude Code Agent Teams Docs — Official documentation
MIT License - see LICENSE
Ahmad Othman Ammar Adi
- GitHub: @OthmanAdi
- Website: othmanadi.com