Stats
Actions
Tags
Help us improve
Share bugs, ideas, or general feedback.
From cowork-marketplace
Guides launching, managing, and monitoring cowork sessions that orchestrate multiple plugin agents in parallel. Covers lifecycle stages, coordination patterns, resource tracking, error handling, and controls.
npx claudepluginhub markus41/claude --plugin cowork-marketplaceHow this skill is triggered — by the user, by Claude, or both
Slash command
/cowork-marketplace:cowork-sessionsThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
Domain knowledge for launching and managing cowork sessions that coordinate multiple plugin agents.
Automates multi-agent execution with task decomposition, parallel dispatch, memory coordination, and verification loops for complex features.
Orchestrates session plan execution in waves with role-based dispatch, inter-wave quality gates, parallel subagents, and progress tracking. Triggered by /go. Handles plan adaptation and recovery from agent failures.
Spawns and coordinates whip agent sessions for tasks, dispatching single agents or teams with backend, scope, and ownership. Manages git worktrees, workspaces, and execution models.
Share bugs, ideas, or general feedback.
Domain knowledge for launching and managing cowork sessions that coordinate multiple plugin agents.
1. INITIALIZING
- Validate item is installed
- Check plugin dependencies
- Parse task description
2. PLANNING
- Break task into subtasks
- Match subtasks to available agents
- Determine execution order and parallelism
3. RUNNING
- Dispatch agents with their subtasks
- Monitor progress (polling every 3-5 seconds)
- Collect intermediate outputs
4. COMPLETING
- Merge outputs from all agents
- Generate session summary
- Record metrics (tokens, cost, duration)
5. COMPLETED / FAILED
- Present final outputs
- Archive session for history
Sessions use the Task tool to spawn sub-agents. Key patterns:
For dependent tasks (e.g., "generate code then write tests"):
Agent 1 completes → output feeds Agent 2 → Agent 2 completes
For independent tasks (e.g., "review security AND check performance"):
Agent 1 ─────────────→ output ─┐
Agent 2 ─────────────→ output ─┤─→ merge
Agent 3 ─────────────→ output ─┘
Common pattern for comprehensive analysis:
Coordinator decomposes task
├──→ Specialist Agent A
├──→ Specialist Agent B
└──→ Specialist Agent C
↓
Coordinator merges results
Each marketplace item defines defaults:
maxParallelAgents - How many agents run simultaneously (1-10)estimatedDuration - Expected session lengthavgSessionMinutes - Historical averagecompletionRate - Success rate (0.0-1.0)Sessions track:
If an agent fails during a session:
| Action | When to Use |
|---|---|
| Pause | Long-running session, need to step away |
| Resume | Continue a paused session from where it stopped |
| Cancel | Task is no longer needed, stop all agents |