Concept: Recipe Cards
A Recipe Card is a standardized, human-readable operational specification designed to govern autonomous super-agents (such as gpt-6-astra) executing complex, multi-day mandates.
Pioneered by nate-b-jones as a successor to traditional “master prompts,” Recipe Cards provide the structured bridge between human intent and autonomous multi-agent execution, eliminating prompt paralysis while enforcing strict boundaries.
The Core Problem with Master Prompts
Traditional prompting assumes a human can write a single, perfectly structured instruction set that anticipates every operational hurdle. For real-world knowledge work and administrative projects spanning 20+ hours (e.g. household moves, multi-jurisdiction tax filings, complex vendor onboarding), master prompts break down:
- They become too unwieldy for humans to draft and maintain.
- They fail to account for dynamic upstream dependencies and third-party blocking states.
- They lack explicit governance over what actions an agent may take silently versus what requires human sign-off.
Anatomical Components of a Recipe Card
A standard Recipe Card contains six structural sections:
- Job Objective & Scope: Clearly names the end-state deliverable and delineates which sub-components the agent manages end-to-end.
- Intake Questions: The precise questions the manager-loops supervisor agent must ask the human during onboarding to uncover hidden constraints and preferences.
- Required Information & Access: The credentials, documents, email inboxes, or cloud accounts needed to execute the work.
- Autonomous Execution Zone: Actions the agent is fully empowered to execute autonomously without notifying or interrupting the user (e.g., cross-referencing maps, calling public APIs, downloading PDF application forms, comparing pricing tiers).
- Human Approval Gates & Escalation: Clear demarcation of irreversible decisions requiring explicit human authorization (e.g., submitting credit card information, paying fees, signing contracts, choosing between shortlisted finalists).
- Error Recovery & Resilience Protocols: Clear contingency procedures defining how the agent should handle blocked portals, missing records, or contradictory external data without abandoning unblocked parallel tracks.