project-knowledge-extraction

Extracting durable project knowledge involves scanning heterogeneous repository evidence like code, docs, and tests to identify stable facts, vocabulary, and boundaries. This process classifies findings into ADRs, context docs, glossaries, and runbooks, preserving decision history for future agents.

ADR mining is the extraction of architecture decision records from repository evidence to preserve decision history. It helps knowledge-capture workflows by turning scattered project context into grounded, reusable artifacts that prevent teams from repeatedly rediscovering the same project facts.

Grounding agent context artifacts requires attaching truth sources and verification metadata during extraction. By classifying evidence into context docs and runbooks while tracking drift-aware freshness metadata, you ensure future agents can check whether the extracted project knowledge remains valid.

Code archaeology can build a knowledge graph by scanning issues, incidents, reports, tests, and scripts to find relevant material. It extracts recurring failure patterns and stable facts, then links these new artifacts into the context graph for future agent reuse.

The best way to maintain project vocabulary and boundaries is extracting durable facts from repository evidence and classifying them into glossaries and context docs. This preserves stable vocabulary and structural boundaries, directly supporting maintenance workflows requiring reliable agent context.

Evidence grounding is not suitable when a repository lacks sufficient heterogeneous evidence like docs, tests, or incidents to inventory. Without source material to attach truth sources and verification metadata, the extracted knowledge cannot be reliably classified or checked for drift.