From agent-core
Guidelines for when to delegate tasks to sub-agents vs handle directly. Use when deciding whether to use Explore, Plan, or other specialized agents for a task.
How this skill is triggered — by the user, by Claude, or both
Slash command
/agent-core:delegation-triggersThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
```
Is this task...
│
├─ Multi-file, independent units? ─► Parallel Orchestration
│ (see complex-orchestrator)
│
├─ Deep analysis / reasoning? ─────► Advanced General Purpose (Opus)
│ ⚠️ High cost - use sparingly
│
├─ Exploration / Research? ────────► Explore Agent
│
├─ Planning / Design? ─────────────► Plan Agent
│
├─ Simple & Direct? ───────────────► Execute Directly
│
└─ Unclear scope? ─────────────────► Decompose First
| Signal | Agent | Trigger Phrases |
|---|---|---|
| Multi-file | Parallel (complex-orchestrator) | "all files", "across codebase", "every X" |
| Deep analysis | Advanced (Opus) | "thoroughly", "しっかり", "implications" |
| Find/Understand | Explore | "where is", "how does", "find all" |
| Design/Architect | Plan | "how should I", "design", "best approach" |
| Simple edit | Direct | "fix", "add field", "update text" |
Launch in single message for parallel:
Task({ description: "A", ... })
Task({ description: "B", ... }) // runs parallel with A
For dependencies, use waves:
Wave 1: shared dependency
Wave 2: parallel consumers (after Wave 1)
For detailed examples and anti-patterns, see reference.md
npx claudepluginhub xmgrex/ccx-arsenal --plugin agent-coreUse when deciding whether to spawn subagents — explicit rules for when delegation pays off and when direct work is faster. Opus 4.8 under-delegates by default; these rules correct that.
Patterns and principles for orchestrating parallel subagent execution: work decomposition (fan-out/fan-in, map-reduce), isolation, result synthesis, and failure handling. Use when a task splits into independent subtasks.
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.