one-shot-subagent-prompts
One-Shot Subagent Task Specification Protocol
Overview
When a primary orchestrator agent delegates work to a child subagent (e.g., spawning an autonomous research subagent or browser subagent), providing a vague, underspecified prompt ("Look into why database connections are failing") triggers a high-cost Clarification Ping-Pong Loop.
The child subagent spends 4 to 8 turns asking for workspace paths, exploring irrelevant folders, asking for permission, and returning conversational status updates - burning 20,000+ tokens across both agent transcripts.
The One-Shot Subagent Task Specification Protocol packages Context Scope, Deterministic Actions, Hard Boundaries, and a Return Contract into the initial invocation prompt, enabling the subagent to complete the task autonomously in a single execution trajectory.