PRAJNA Protocol — AI Session Knowledge Lifecycle Management for Multi-Service Software Systems

We disclose PRAJNA (प्रज्ञा), a protocol for capturing, structuring, and persisting tacit knowledge held by AI coding assistants during active software development sessions. Key novelties: (1) Pre-build knowledge seeding — AI assistant externalises domain understanding into structured canonical documents (brainstorm, project, SHASTRA rules, YUKTI reasoning, VIVEKA inferences, task list) before any code is written, inverting the traditional code-then-document flow; (2) Progressive knowledge enrichment — every subsequent AI session appends a dated stratum of new knowledge to existing documents, never replacing prior layers, creating a compounding knowledge archaeology; (3) Autonomous gap-filling daemon — a structural extractor that runs daily, produces provisional knowledge documents from source code archaeology, and injects machine-readable knowledge quality signals into service registries; (4) Machine-readable knowledge confidence encoding — capability declarations accompanied by trust_mask (32-bit bitmask), lifecycle status, and capability audit counts enabling downstream systems to query both capability and confidence without natural language interpretation. The three-layer SHASTRA/YUKTI/VIVEKA schema (domain rules / reasoning patterns / pre-computed inferences) is domain-agnostic and AI-native, designed to receive the three types of knowledge an AI session naturally produces. Reference implementation: ANKR Labs 222-service platform, 2026.