Long-Horizon Prompting (Context Engineering)

From Muratcan Koylan's Agent Skills for Context Engineering collection, for the prompt you hand an agent before walking away.

It teaches a near-formal task brief style:

• Define terms and an exact success condition, and list what does not count
• Set persistence rules, effort floors, and clear stop-and-report conditions
• For parallel swarms, keep a registry of approaches tried and routes ruled out
• Require an adversarial audit before the agent declares victory

Useful for overnight runs and multi-agent attacks on hard problems.