Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add instructions/kimsanguine/llm-brain/claude-mdgit clone --depth 1 https://github.com/kimsanguine/llm-brainWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.03289 | $0.03289 |
| Opus 5 | $0.01644 | $0.01644 |
| Sonnet 5 | $0.00658 | $0.00658 |
| Haiku 4.5 | $0.00329 | $0.00329 |
Grade A, and why
llm-brain CLAUDE.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 173 lines — stays where its author put it; the contents beside it link to each section on GitHub.
llm-brain — Claude Code 운영 지침
Claude는 이 시스템의 컴파일러다. raw/ 소스를 읽어 wiki/를 생성·갱신하고, 사용자 질문에 wiki 기반으로 답한다.
가드레일 (절대 위반 금지)
raw/출처 없이wiki/신규 생성·사실 수정 금지- query 응답 중
wiki/편집 금지 - 학습 데이터만으로
wiki/작성 금지 — 반드시raw/근거 필요 raw/파일 수정 금지 — 읽기 전용
프라이버시 경계 (Agent Memory OS)
episodes/— 운영 맥락(user_goal·inputs·outputsverbatim)을 담는 append-only 원장. ingest·express·query·curate 4 경로가 자동 기록(fail-soft). gitignored (스키마 예시 1개만examples/에 커밋) — one-way door 누출 방지.index.md— wiki 목차(gitignoredwiki/의 파생물).business/민감 제목 노출 방지로 gitignored·git 추적 제외(결정 2026-06-28; 과거 public 커밋 history scrub은 별도 사람 판단).procedures/— git-tracked이되 OKF export 제외(schema/okf_export.yaml의exclude_paths).wiki/memory_health_report.md— okfMETA_FILES에 등재돼 공개 OKF 번들에서 봉인(미포함).
명령어
플러그인(
commands/)으로 제공된다. 설치 시 커맨드는/llm-brain:ingest처럼 네임스페이스가 붙는다 — 아래/ingest·/okf등은/llm-brain:접두로 읽는다. 자연어("ingest 해줘", "okf 해줘")로도 호출된다.
ingest
"ingest 해줘"
"/ingest https://url [--resonance high|medium|low]"
"/ingest ~/path/to/file.pdf [--resonance high]"
"/ingest '텍스트 내용' [--resonance medium]"
scripts/ingest.py 실행 → schema/ingest.md 규칙 적용 → wiki/ 생성·갱신 → index.md 업데이트
에피소드 자동기록: raw 저장 직후
episodes/YYYY-MM.jsonl에 1줄 append (statuspending_wiki_compilation, fail-soft — 실패해도 ingest 경로 불간섭).
curate
"curate --distill" # distill_level 점진 압축
"curate --lifecycle" # TTL 초과 페이지 archive 후보
"curate --all" # 전체 실행 (audit + distill + lifecycle)
scripts/curate.py 실행 → schema/curate.md 규칙 적용
export-graph (wikilink 그래프 export)
uv run python scripts/export_graph.py
[[wikilink]] 파싱 → wiki/graph.json 생성 (D3 force-graph 형식).
mini-graph는 wiki_app /api/page/{slug}/graph 엔드포인트로 조회.
okf (wiki → OKF v0.1 호환 번들 export)
"okf 해줘" # /llm-brain:okf — 먼저 dry-run 검토 후 okf/ 번들 생성
"/llm-brain:okf --dry-run" # export 대상·제외·통계만 (파일 미작성, 보안 검토용)
"/llm-brain:okf --strip-internal" # 외부 공유본 (x-llmbrain-* 제거)
okf 커맨드(commands/okf.md)가 scripts/okf_export.py를 실행해 wiki/를
OKF v0.1(Google Open Knowledge Format) 호환 번들 okf/로 투영한다 (동료·외부 에이전트·habix
제품이 번역 없이 소비). frontmatter는 OKF 예약 6필드로 매핑, 내부 필드는 x-llmbrain-*로 보존,
[[wikilink]]는 /-절대경로 마크다운 링크로 변환. 변환 규칙: schema/okf.md.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- yesterday First seen · 173 lines · 3,289 tokens per session scan A 23be1a70ac53
llm-brain CLAUDE.md is an instructions file published in the GitHub repository kimsanguine/llm-brain (10 stars, last pushed 17d ago), licensed MIT. It adds 3,289 tokens to every session, about $0.0164 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other instructions, from other repositories
cmds-llm-wiki AGENTS.md
Schema and harness document for the CMDS LLM Wiki vault. Defines the 3-layer architecture (Raw Sources / Wiki / Schema), ingest-query-lint operations, file conventions, and frontmatter standards. This is the single source of truth for LLM behavior in this vault.
cmds-llm-wiki CLAUDE.md
Schema and harness document for the CMDS LLM Wiki vault. Defines the 3-layer architecture (Raw Sources / Wiki / Schema), ingest-query-lint operations, file conventions, and frontmatter standards. This is the single source of truth for LLM behavior in this vault.
claude-obsidian GEMINI.md
Gemini CLI instructions for AgriciDaniel/claude-obsidian: Read AGENTS.md as the canonical host-neutral contract. Skills live in skills/ /SKILL.md and the portable core lives in claudeobsidian/.
claude-obsidian AGENTS.md
AGENTS.md instructions for AgriciDaniel/claude-obsidian, covering claude-obsidian: agent instructions, product and vault boundaries, bootstrap, canonical skills and mutation protocol.
claude-obsidian copilot-instructions.md
Copilot instructions for AgriciDaniel/claude-obsidian: Read AGENTS.md before proposing changes. The repository is an Agent Skills package, Claude Code adapter, standard-library Python core, and deterministic vault template—not the default live user vault.
obsidian-llm-wiki AGENTS.md
Instructions for green-dalii/obsidian-llm-wiki, covering llm wiki plugin project development standards, 🛡️ six-gate quality closure, gate 1: five-gate automated, gate 2: no side effects and gate 3: no breaking changes.