subagent-prompt-engineering-pattern

What problem does it solve?

It solves unreliable multi-agent subtask execution where parallel edits and git races can cause conflicts, stalls, or unsafe behaviors like force pushes.

Core Features & Use Cases

• 7-Section Subagent Prompt Template: Provides a repeatable structure (context, mandatory race-prevention section, explicit outputs, step-by-step tasks, constraints, and a short report) that improves success rate.
• Race-Ready Git Workflow Instructions: Forces a standardized recovery strategy (grep verify → re-read/edit retry → pull --rebase → retry → force push forbidden).
• Atomic Commit Guardrails: Requires explicit file staging and a retryable push procedure to prevent cross-agent interference and data loss.
• Operationally Validated at Scale: Tuned against gem-llm runs (~60h, ~530 subagent dispatches) with recorded zero force pushes and successful automatic race recovery.

Quick Start

Tell your AI to draft and dispatch a Claude Code Agent subtask using this skill’s 7-section prompt structure and mandatory race/atomic commit sections, with a 200–400 word completion report.