@yohakuforce/core

Salesforce metadata → SQLite knowledge graph → deterministic Markdown.

Core is the deterministic foundation layer of the 余白フォース / yohakuforce suite. It turns Salesforce DX metadata into a queryable SQLite knowledge graph, then renders documentation deterministically — the same input always produces the same output, and every fact carries its provenance. Core never calls an LLM; it is the stable scaffold that AI tools build on top of.

Why deterministic?

AI-generated docs drift: re-run them and you get different prose, sometimes different facts. Core separates the two concerns — it owns the deterministic half (extract → graph → render), and leaves the interpretive half (explanations, summaries) to AI tools that consume the graph. That makes the output auditable and reproducible.

Install

npm install -g @yohakuforce/core
yohaku --help

Quick start

# 1. Initialise the scaffold in your Salesforce DX project
yohaku init --bootstrap --profile standard

# 2. Build the knowledge graph from force-app/ metadata
yohaku graph build --incremental

# 3. Inspect the graph
yohaku graph schema --tables
yohaku graph query "SELECT name FROM sobject LIMIT 10"

# 4. Render documentation (deterministic Markdown)
yohaku render all            # everything
yohaku render system-index   # project overview
yohaku render objects        # per-SObject (fields / validation rules / dependencies)
yohaku render flows          # per-Flow
yohaku render apex           # per-ApexClass

Pipeline

Salesforce metadata  →  SQLite knowledge graph  →  deterministic render  →  Markdown docs
   (SFDX / DX-MCP)        (yohaku graph build)        (yohaku render)

Optional: Context-Hub context injection

Core can pull abstracted project context from Context-Hub to enrich explain / change-summary output. It is opt-in (default: none), configured in .yohaku/config.json, and only abstracted context is used — customer names, PII and secrets are never written into generated artifacts.

yohaku context --kind explain --fqn Account

Design principles

1. Three-layer separation — deterministic processing lives here; AI inference does not.
2. Deterministic I/O — same input → same output, verified by golden tests.
3. Read force-app/, never write it — the source of truth is never mutated.
4. Graph is CLI-only — no direct INSERTs from AI or hooks.

The suite

Core is one of three layers (Core / Context-Hub / AI Manager). See the official help site at yohakuforce.github.io/docs for how they fit together.

License

Apache-2.0