Choose the Right Dispatch: Spawn, Fork, or Direct

PROJECT CONTEXT:
My agent either does everything in one bloated context or spawns sub-agents indiscriminately, paying full prompt cost for trivial lookups. Claude Code distinguishes spawn (fresh context, specialist definition), fork (inherited context, shared cache), and direct execution - each with distinct costs and fit.

GOAL:
Implement spawn and fork dispatch paths with correct context handling, and encode a decision policy for when each strategy (including direct) applies.

REQUIREMENTS:
- A spawn path: named agent definitions with fresh context, own tool pool, own permission mode and model
- A fork path: child inherits the parent conversation and system prompt for cache sharing
- A direct-execution default for tasks below dispatch overhead
- Recursion guards: forked children must not fork again
- A written decision matrix the orchestrator prompt actually encodes

CONSTRAINTS:
- A spawned agent knows nothing - its dispatch prompt must carry complete background
- Fork children must run under strict behavior rules (no chatting, bounded reports)

DEFINITION OF DONE:
- Spawn produces a sub-agent with zero parent history and an independent tool pool
- Fork produces a child whose request prefix is byte-identical to the parent's
- A forked child attempting to fork is refused by a guard, not by luck
- The decision matrix is followed in transcripts: specialists spawned, research forked, trivia direct

COMMON FAILURES TO AVOID:
- Spawning with a one-line prompt and wondering why the sub-agent flails without context
- Forking for everything, including tasks needing a specialist definition the fork lacks
- Unbounded recursive forking melting the budget
- Dispatching a sub-agent (with full startup cost) for a two-step task the parent could do inline