meta-kb

Show pending changes and LLM-powered impact analysis without compiling. Detects new/modified/deleted sources via content hashing, maps them to dirty buckets, reads changed sources to assess impact and detect contradictions with existing claims. Use to preview what incremental compilation would do before running it.

Incrementally recompiles changed parts of the wiki after new sources are ingested. The skill handles change detection, LLM-powered impact analysis, and user confirmation. All compilation is delegated to the script pipeline (bun run compile --incremental), which handles entity extraction, resolution with anchoring, synthesis with evidence registry, reference cards, claims, self-eval, and state saving.

Extracts atomic verifiable claims from synthesis articles and runs self-evaluation by verifying a sample against raw sources. Outputs build/claims.json and build/eval-report.json. Use when extracting claims, running self-eval, or verifying wiki accuracy.

Writes field-map.md, the flagship overview article that connects all taxonomy areas into one system. A 3000-5000 word systems map showing integration points, paradigm routing logic, and a concrete practitioner flow. Use when writing the field map or systems overview.

Compiles the wiki from raw markdown sources. Read `config/domain.ts` for the domain topic and taxonomy. Orchestrates a multi-phase pipeline using subagents: read sources, write synthesis articles, write reference cards, generate field map and indexes, extract claims, and run self-eval. Use when asked to compile, build, or generate the wiki from raw sources.

Deep-researches a project or paper for the knowledge base. Clones repos to read source code, fetches full paper text, and produces structured analysis files in raw/deep/. Use when asked to deep-research, deep-dive, analyze source code, or enrich a source with implementation details.

Writes a single synthesis article (3000-5000 words) for one taxonomy bucket. Produces landscape analysis with approach categories, convergence, failure modes, and selection guide. Use when writing a synthesis article for a bucket.

Generates the agent-optimized ROOT.md topic index, project/topic indexes, missing coverage list, landscape comparison table, and wiki README. Use when generating indexes, the ROOT.md entry point, or the comparison table.

Writes reference cards for projects, concepts, or people in the wiki. Projects get 800-1500 words with architecture, benchmarks, limitations, and alternatives. Concepts get 1500-2500 words. People get 150-250 words. Use when writing wiki reference cards for specific entities.