Personal knowledge base engine: PDF/Word/EPUB → LLM entity relationship graph + hybrid retrieval (vector + BM25 + RRF fusion) + 0.92 semantic cache + Obsidian entity-isolated export + MCP (direct connection to Claude/Cursor). Intelligent PDF classifier (automatically uses MinerU when formula density ≥2%), lazy initialization, and production-grade asynchronous robustness. Contributions of examples and benchmarks are welcome.
11 files for Claude Code, OpenCode and Codex: opsx-apply, opsx-archive, opsx-explore, opsx-propose and 7 more — 1,225 tokens loaded in every session.
AGENTS.md A 1,225 tok .opencode/skills/openspec-apply-change/SKILL.md A 31 tok .opencode/skills/openspec-archive-change/SKILL.md A 31 tok .opencode/skills/openspec-explore/SKILL.md A 39 tok .opencode/skills/openspec-propose/SKILL.md A 47 tok .opencode/skills/openspec-sync-specs/SKILL.md A 38 tok .opencode/commands/opsx-apply.md A 10 tok .opencode/commands/opsx-archive.md A 8 tok .opencode/commands/opsx-explore.md A 13 tok .opencode/commands/opsx-propose.md A 15 tok .opencode/commands/opsx-sync.md A 9 tok These files are q1ngn1ng-web/kb-rag's own configuration — they tell Claude Code, OpenCode and Codex how to work on this repository, so they are not mods to install elsewhere. Copy one as a starting point and replace the parts that are about this project.