Jayalexandermg Skilljacked AI Agent Project Context Documentation

AI Agent Project Context Documentation

Generate comprehensive project documentation to eliminate AI agent errors through proper context provision.

Quick Start

When starting a new project, immediately create four core documents: PRD, architecture.md, decision.md, and feature.json using a single Claude prompt containing all project information.

Core Workflow

Trigger: Starting new project OR AI agents producing context-related errors

1. Gather Complete Project Information

   • Collect all project requirements, scope, and technical details
   • Identify frameworks and libraries to be used
   • Document project constraints and goals

2. Create Master Prompt

   • Structure all collected information into comprehensive prompt
   • Include specific instruction to generate four distinct documents
   • Request token-efficient formatting for feature documentation

3. Generate Four Core Documents:

   • PRD: Project requirements and scope
   • architecture.md: Data formatting, file structure, APIs, architecture details
   • decision.md: All architectural and implementation decisions for future reference
   • feature.json: All features in JSON format with completion criteria and tracking

4. Validate Documentation Set

   • Ensure PRD covers complete project scope
   • Verify architecture.md includes all technical specifications
   • Confirm decision.md captures rationale for future agents
   • Test feature.json format for token efficiency

Techniques

Feature.json Structure:

• Use token-efficient JSON format
• Include completion criteria for each feature
• Add "passes" key for implementation tracking
• Structure: feature details + completion criteria + status tracking

Context Optimization:

• Break large projects into documented subparts
• Document framework/library specifics agents will encounter
• Create decision log for consistent future agent behavior

Anti-Patterns

NEVER skip documentation creation when starting projects - context errors compound exponentially.

NEVER create incomplete PRDs - agents need full scope understanding to avoid scope creep.

NEVER omit decision documentation - future agents will remake the same decisions inconsistently.

NEVER use verbose feature documentation - token efficiency is critical for agent processing.

Edge Cases & Error Handling

Complex Multi-Module Projects: Break PRD into module-specific sections while maintaining overall coherence.

Evolving Requirements: Update all four documents simultaneously to maintain context consistency.

Legacy Project Documentation: Generate documentation retroactively by analyzing existing codebase with Claude before making changes.

Bundled Resources Plan

• templates/prd-template.md - Standard PRD structure and required sections
• templates/architecture-template.md - Architecture documentation format
• templates/feature-schema.json - JSON schema for feature documentation
• prompts/documentation-generator.txt - Master prompt for generating all four documents

Source: jayalexandermg/SkillJacked — distributed by TomeVault.