agentleFS
catalsqlb-cmdcatalsqlb-cmd/Prism-H/skills/llm-wiki/SKILL.md

llm-wiki

Skill16 starsChanged 6 months agoMITView on GitHub

General-purpose personal knowledge base built on Karpathy's LLM Wiki pattern: an immutable raw/ source layer, a generated human-readable wiki/ layer, and a CLAUDE.md schema/contract. Portable, stdlib-only, Obsidian-compatible. Use when the user says \"个人知识库\", \"llm wiki\", \"personal wiki\", \"ingest into wiki\", \"建/查个人知识库\", or wants a durable cross-project notes graph that is NOT research-lifecycle specific. For paper/idea/experiment/claim research graphs use /research-wiki instead.

94 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use touchdesigner-mcp, teams-meeting-pipeline and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • john-data-chenSkill

    touchdesigner-mcp

    Control a running TouchDesigner instance via twozero MCP — create operators, set parameters, wire connections, execute Python, build real-time visuals. 36 native tools.

    Also in 7 projects that use this

  • john-data-chenSkill

    teams-meeting-pipeline

    Operate the Teams meeting summary pipeline via Hermes CLI — summarize meetings, inspect pipeline status, replay jobs, manage Microsoft Graph subscriptions.

    Also in 9 projects that use this

  • mae616Skill

    p5js

    p5.js(クリエイティブコーディング)のProcessing由来「コードでスケッチ」思想を軸に、setup/drawループ・座標系・インタラクション・パフォーマンス最適化を整理する。p5.js、Processing、クリエイティブコーディング、ジェネラティブアート、インタラクティブ表現、ビジュアルプログラミング、Canvas描画、パーティクル、フラクタル、ノイズ表現に関する相談で必ず使うこと。ユーザーが「コードでアート」「動くビジュアル」「インタラクティブな描画」と言った場合にも適用する。

    Also in 12 projects that use this

  • aivrarSkill

    competitor-news-monitor

    Watch named companies for material news; cited digests.

    Also in 6 projects that use this

  • math-incSkill

    heartmula

    Set up and run HeartMuLa, the open-source music generation model family (Suno-like). Generates full songs from lyrics + tags with multilingual support.

    Also in 7 projects that use this

  • Ivy-Chen1999Skill

    claude-history-ingest

    Ingest Claude Code conversation history into the Obsidian wiki. Use this skill when the user wants to mine their past Claude conversations for knowledge, import their ~/.claude folder, extract insights from previous coding sessions, or says things like "process my Claude history", "add my conversations to the wiki", "what have I discussed with Claude before". Also triggers when the user mentions their .claude folder, Claude projects, session data, past conversation logs, local-agent-mode sessions, or audit logs.

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

11 rules. 0 are common, 4 are distinctive, and 7 are very specific to this project.

  • 0

    Common

    Most of the 76 edited versions we read keep these.

  • 4

    Distinctive

    This author's own choices. Worth a look.

  • 7

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    All structure and search logic lives in two stdlib-only scripts (per the integration contract — one implementation, not copy-pasted prose):

    line 62
  2. Very specific:

    After init, tell the user: drop sources into raw/, then run /llm-wiki ingest.

    line 86
  3. Very specific:

    Add a one-line entry under the right section of wiki/index.md for any new page.

    line 107
  4. Distinctive:

    Always run this before broad external search.

    line 116
  5. Very specific:

    Run lint to find orphans and missing-index entries.

    line 135
  6. Very specific:

    Add any orphaned durable pages to wiki/index.md.

    line 136
  7. Distinctive:

    Prefer updating an existing page over creating a near-duplicate.

    line 144
  8. Very specific:

    Every write appends wiki/log.md.

    line 145
  9. Very specific:

    Every substantive claim carries a status tag.

    line 146
  10. Very specific:

    Use Obsidian-style [[topics/foo]] links so the graph stays navigable.

    line 147
  11. Distinctive:

    When linking a project to a local wiki, this is the general-purpose target.

    line 153

“Common” is measured: kept by at least 6 in 10 of the 76 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

LLM Wiki: Portable Personal Knowledge Base

Subcommand: $ARGUMENTS

Overview

/llm-wiki builds and maintains a general-purpose personal knowledge base that follows Karpathy's LLM Wiki pattern: compile knowledge once into durable, linked Markdown; keep it current; never re-derive on every query.

It is discipline-agnostic and Obsidian-compatible. Use it for durable domain knowledge, reading notes, distilled answers, and any long-lived personal KB. For the research-lifecycle graph (papers / ideas / experiments / claims with typed edges) use /research-wiki instead — the two do not share directories and should not be merged.

This skill matches the external-wiki shape documented in shared-references/local-wiki-link.md, so any wiki created here is auto-detectable by the rest of Prism.

Three layers

LayerPathMutabilityPurpose
Rawraw/immutable — never edit/deletesource material: PDFs, clippings, transcripts, dumps
Wikiwiki/agent-generateddistilled, linked, durable knowledge
SchemaCLAUDE.mdhuman-ownedhow the wiki is built & linted
<root>/
  raw/                    immutable sources
  wiki/
    index.md              navigation catalog
    log.md                append-only timeline
    topics/  concepts/  sources/  memos/
  scripts/query_wiki.py   shipped search helper
  scripts/llm_wiki.py     shipped scaffold/lint helper
  CLAUDE.md               schema + maintenance contract

Status tags

Tag every non-trivial claim so future reads know how much to trust it:

  • 已确认 — confirmed, stable prior knowledge
  • 有争议 — contested; preserve competing positions
  • 推论 — inference; usable as reasoning, not as a source
  • 待核验 — unverified; do NOT cite as settled support

Claims with no source anchor MUST be tagged 待核验.

Canonical helpers

All structure and search logic lives in two stdlib-only scripts (per the integration contract — one implementation, not copy-pasted prose):

  • tools/llm_wiki.py — init, lint, log
  • tools/query_wiki.py — keyword search

The repo copies live under Prism's tools/; init also copies both into the wiki's own scripts/ so the wiki is self-contained and portable.


Subcommands

init <path> — scaffold a new wiki

python3 tools/llm_wiki.py init <path>

Creates the three-layer structure, seed index.md / log.md, a starter CLAUDE.md schema, an example concept page, and copies the search/scaffold helpers into <path>/scripts/. Idempotent — existing files are never overwritten.

After init, tell the user: drop sources into raw/, then run /llm-wiki ingest.

ingest [raw-path] — fold a source into the wiki

For each new item under raw/ (or the given path):

  1. Read the raw source (do not modify it).
  2. Query first to avoid duplication:
    python3 <root>/scripts/query_wiki.py --wiki <root>/wiki --query "<topic>" --top 8
  3. Write/update a wiki/sources/<slug>.md summary page with a link or anchor back to the raw/ item.
  4. Fold durable facts into the relevant wiki/topics/ or wiki/concepts/ page. Prefer updating an existing page over creating a near-duplicate. Tag each claim with a status. Add [[wiki-links]] to related pages.
  5. Append the log and update the index:
    python3 <root>/scripts/llm_wiki.py log <root> --msg "ingest — <pages> — <source>"
    Add a one-line entry under the right section of wiki/index.md for any new page.

query <topic> — search prior knowledge

python3 <root>/scripts/query_wiki.py --wiki <root>/wiki --query "<topic>" --top 8

Always run this before broad external search. Read only the top relevant pages. Treat conclusions by their status tag (see above).

lint — hygiene check

python3 <root>/scripts/llm_wiki.py lint <root>

Reports broken links, missing H1 titles, untagged substantive pages, stale pages (>180d), conflicting status on one line, orphan pages (no incoming links and not in index), and index references to missing pages. Exit code 1 on error-level findings (missing-page references). Fix warnings opportunistically; fix errors before relying on the wiki.

sync — reconcile index/log with disk

Rebuild coverage after manual edits:

  1. Run lint to find orphans and missing-index entries.
  2. Add any orphaned durable pages to wiki/index.md.
  3. Append a log.md entry summarizing the reconciliation.

Write-back rules (hard)

  1. Never modify or delete anything under raw/.
  2. Prefer updating an existing page over creating a near-duplicate.
  3. Every write appends wiki/log.md; new pages also update wiki/index.md.
  4. Every substantive claim carries a status tag; unsourced claims are 待核验.
  5. Use Obsidian-style [[topics/foo]] links so the graph stays navigable.

Relationship to other Prism wiki systems

  • External wiki protocol (shared-references/local-wiki-link.md): a wiki created by /llm-wiki init is exactly the shape that protocol auto-detects. When linking a project to a local wiki, this is the general-purpose target.
  • /research-wiki: the research-lifecycle graph (papers/ideas/experiments/ claims + graph/edges.jsonl). Use both when useful; do not merge directories. Query /llm-wiki for durable domain knowledge; update /research-wiki for the project's research graph.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.