Long-Horizon “Super Prompts”
Research, inventory, and ready-to-deploy prompts that keep AI agents working nonstop until the whole project is done — built on the pseudo-formal task-brief discipline: an exact success predicate, a non-counting list, an artifact-predicate return condition, and a verification gate paired with every persistence instruction.
Deep Research
Why agents quit or fake done; the 10-block brief anatomy; the CDC exemplar; vendor doctrine; four copy-paste templates (master brief, orchestrator, worker spec, session ritual).
The Full Report
The 12-part treatment: diagnosis, the persistence-vs-hacking tradeoff, the verification bottleneck, orchestration prompts, what must live outside the prompt, failure taxonomy, the ten rules.
Prompt Inventory
Every already-written long-horizon super prompt we could find — OpenAI/Anthropic persistence blocks, Ralph loop, /goal, GSD, leaked agent loops, and the marketing/business sets — graded against the bar.
The Kit
Nine ready-to-deploy files: master brief template, four prompts, two harness specs (loop + progress ledger), a pre-launch rubric, and the red-team method with 13 patched exploits.
Never add a persistence instruction without a matching verification gate.