CORE Cookbook

The foundational recipe that establishes CRAFT session context, loads project files, detects master templates, and activates all framework protocols. This recipe ensures proper initialization of every CRAFT interaction, setting up the communication system, loading handoffs, and preparing the AI assistant for project-specific work.

The Single Recipe Runner safely imports and executes external CRAFT recipes from files or URLs through comprehensive security validation (code injection detection, file system protection, network access control), structure verification (format compliance, parameter validation), parameter collection, and session-only execution – enabling secure experimentation with community recipes and one-time specialized tasks without permanently modifying cookbooks or risking framework integrity.

Continuously monitors AI comment usage to ensure proper acknowledgment of user directives, appropriate warning markers, required question flags, and status updates. Operates silently by default to self-correct communication issues without disrupting conversation flow, maintaining CRAFT’s bidirectional comment system standards.

Converts lengthy AI platform usage policies (20+ pages) into concise, scannable summaries with actionable categories, trigger patterns, and practical examples. Creates Python-compatible digests that integrate directly into CRAFT framework’s policy awareness system, making compliance practical rather than burdensome.

The Prompt Framework Intelligence Assistant embeds expertise on 20+ proven prompt engineering frameworks (Chain of Thought, Tree of Thoughts, ReAct, Few-Shot, etc.) to analyze prompts and suggest optimal structures for dramatically improved clarity and AI response quality. It matches task types to frameworks, explains why specific patterns help, offers restructuring, and tracks effectiveness – making expert-level prompt engineering accessible within CRAFT.

This recipe is required to make CRAFT function effectively as a cumulative intelligence system and will be used by the AI when concluding completed projects, identifying valuable mid-project discoveries worth preserving, preparing for project archival, or explicitly requested for knowledge extraction. The recipe ensures individual project learnings become organizational assets available to all future CRAFT implementations.

The Expectation Setter recipe analyzes incoming tasks, communicates scope, complexity, deliverables, and limitations upfront before work begins. It automatically initializes progress tracking for complex tasks, preventing miscommunication and building trust through transparency. This recipe transforms ambiguous requests into clear, shared understanding between user and AI.

Real-time tracking and proactive management of token consumption during AI chat sessions. Provides continuous usage updates, threshold warnings, and automatic preservation strategies to prevent work loss from unexpected session limits.

Automatically detects technical errors and translates them into plain-language explanations adapted to user expertise level. Categorizes severity, provides context-aware recovery guidance, and offers prioritized action steps. Transforms confusing error codes into constructive problem-solving opportunities.

Automatically detects vague, ambiguous, or incomplete instructions in user requests and proactively asks clarifying questions before proceeding, ensuring AI responses match user intentions.

Automatically validates factual claims in AI responses by checking against available sources, assigning confidence levels, flagging unsourced assertions, and generating structured Deep Research prompts when verification is needed. This recipe ensures accuracy, prevents hallucinations, and creates clear pathways to deeper research.