`README.md`

54 lines
# knowledge-layer-skill

A Claude Code **skill** that builds and maintains an **AI Knowledge Layer** — a
plain-directory, LLM-compiled wiki (the Karpathy / "gBrain" pattern) that agents
read *first* before any task. No RAG, no vector database: the LLM reads curated
sources once and compiles them into typed, cross-referenced, contradiction-aware
entries. The **structure is the index**.

> Pattern credit: Andrej Karpathy's LLM knowledge-base workflow, and Shann
> Holmberg ([@shannholmberg](https://x.com/shannholmberg)), who popularized the
> "gBrain" / AI-Knowledge-Layer framing.

## Layout

```
knowledge-layer-skill/          ← this git repo
└── knowledge-layer/            ← the skill package (symlinked into ~/.claude/skills/)
    ├── SKILL.md                ← metadata + workflow
    ├── scripts/                ← stdlib-only Python 3, $0 to run
    │   ├── kl_init.py          ← scaffold a new knowledge layer
    │   ├── kl_scan.py          ← incremental compile work-list (content-hash diff)
    │   ├── kl_query.py         ← read-first retrieval (keyword/frontmatter, no vectors)
    │   └── kl_status.py        ← health report (drift / contradictions / orphans)
    ├── references/
    │   ├── compile-protocol.md ← the exact LLM compile procedure
    │   └── wiki-principles.md   ← "a layer you keep vs. one you abandon"
    └── assets/template/         ← seed docs copied into each new layer
```

## Quick start

```bash
cd knowledge-layer
python3 scripts/kl_init.py mybrain           # scaffold
#  … drop sources into mybrain/sources/ …
python3 scripts/kl_scan.py mybrain           # what changed?
#  … compile per references/compile-protocol.md, then:
python3 scripts/kl_scan.py mybrain --commit
python3 scripts/kl_query.py "some question" mybrain --print
python3 scripts/kl_status.py mybrain         # health
```

## Install as a skill

The package is symlinked into `~/.claude/skills/knowledge-layer` so Claude Code
discovers it. To install elsewhere, symlink or copy the `knowledge-layer/`
folder into any `skills/` directory, or run
`skill-creator/scripts/package_skill.py knowledge-layer` to produce a
distributable `.zip`.

## License

MIT — see `LICENSE`.