Menu
Knowledge for launching, managing, and monitoring cowork sessions that orchestrate multiple plugin agents in parallel
Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Baseado na classificação ocupacional SOC
Select and coordinate multi-agent teams (topology kits, role-based squads, lifecycle, worktree isolation). Use this skill whenever launching parallel agents, designing a review board, running a debug council, scheduling an orchestrator-workers team, configuring agent tool restrictions, or deciding between solo and team execution. Triggers on: "launch a team", "parallel agents", "review board", "debug council", "architect-implementer-reviewer", "swarm", "multi-agent", "subagents for X", "team topology", "agent lifecycle".
Select and wire an agentic design pattern (reflection, prompt chaining, routing, parallelization, orchestrator-workers, evaluator-optimizer, ReAct, blackboard) into the 5-layer Claude Code stack. Use this skill whenever deciding how to structure a multi-step task, whether to spawn subagents, how to run parallel review, or when to use which pattern. Triggers on: "which pattern", "orchestrate", "parallel review", "self-review", "chain of thought", "eval-optimizer loop", "blackboard", "ReAct", "how to decompose this task".
Auto mode permission handling — classifier-based approvals, PermissionDenied hook, defer permissionDecision, and autonomy profiles for hands-off Claude Code usage
Configure Claude Code's autonomous operating mode — profile selection (conservative, balanced, aggressive, unattended-review), permission blocks, and the three gate agents (planner, verifier, reviewer). Use this skill whenever enabling autonomous mode, switching profiles, tightening permissions for production branches, or setting up unattended execution. Triggers on: "autonomy", "unattended mode", "auto-approve", "permission mode", "autonomy profile", "gates", "/cc-autonomy", "planner verifier reviewer", "let claude run on its own".
Manage Claude Code's context window — token arithmetic, /compact strategy, anchor preservation, progressive loading, session analytics. Use this skill whenever a session gets long, context approaches limits, after /compact, when deciding what to load into CLAUDE.md vs leave in references, or when analyzing session cost/token usage. Triggers on: "context full", "compact", "too many tokens", "budget", "session analytics", "save tokens", "context window", "/compact strategy".
Evidence-driven deep analysis for hard coding problems — architecture decisions, root-cause investigation, high-stakes refactor planning, performance bottleneck isolation. Use this skill whenever the user asks for "the best approach", a "deep analysis", "root cause", "principal engineer review", or runs /cc-intel. Also triggers on hard debugging questions, major architectural choices, tricky performance problems, or any task where a hypothesis tree and evidence table matter more than a fast answer.
| name | cowork-sessions |
| description | Knowledge for launching, managing, and monitoring cowork sessions that orchestrate multiple plugin agents in parallel |
| allowed-tools | ["Read","Bash","Task","Glob","Grep"] |
| triggers | ["launch cowork session","start session","run cowork","parallel agents","session management","cowork orchestration"] |
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 Agent 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 |