agentleFS
inkparkinkpark/skills/llm-wiki/SKILL.md

llm-wiki

Skill1 starsChanged 7 months agoApache-2.0View on GitHub

Maintain a reviewable LLM Wiki from immutable raw notes, including language-aware ingest planning, querying, linting, and durable wiki updates.

95 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 9 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

  • 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

  • Ivy-Chen1999Skill

    graph-colorize

    Color-code the Obsidian graph view by rewriting `.obsidian/graph.json` colorGroups. Use this skill when the user says "color my graph", "color code obsidian", "colorize the graph", "color the graph by tag", "color by category", "highlight visibility in graph", "make the graph colorful", "distinguish tags in graph", or wants nodes in Obsidian's graph view tinted by tag, folder, or visibility. Generates a `colorGroups` array from the vault's actual tags/categories and merges it into the existing graph.json without clobbering other graph settings. Always backs up first.

    Also in 3 projects that use this

  • +3 more from Ivy-Chen1999

The file

Show its rules and text

We split this skill into the rules it gives an agent

34 rules. 0 are common, 16 are distinctive, and 18 are very specific to this project.

  • 0

    Common

    Most of the 77 edited versions we read keep these.

  • 16

    Distinctive

    This author's own choices. Worth a look.

  • 18

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  • could be automatic: a setting could enforce it
  1. Very specific:

    Use this skill from any capable AI agent when the user asks to ingest notes into an LLM Wiki, maintain or query the wiki, lint wiki …

    line 8
  2. Very specific:

    When the user invokes only llm-wiki or otherwise provides no additional task details, do not show a workflow menu.

    line 10
  3. Very specific:

    If the user does not explicitly specify a working directory or knowledge repository path, operate on the current session working directory.

    line 12
  4. Very specific:

    Treat raw/ as immutable source evidence: never edit, move, delete, normalize, or write generated files under it.

    line 16
  5. Very specific:

    Treat wiki/ as the reviewable compiled Markdown layer.

    line 17
  6. Distinctive:

    Treat the session working directory as the knowledge repository root unless the user explicitly provides another directory.

    line 18
  7. Very specific:

    Every generated or updated wiki page that is derived from raw material must link to its corresponding raw/ file(s), not only list them in …

    line 19
  8. Distinctive:

    Do not install packages, call models/APIs, or generate a full wiki unless the user explicitly requests that execution step.

    line 20
  9. Very specific:

    Prefer small, reviewable wiki updates and record meaningful changes in wiki/log.md.

    line 24
  10. Distinctive:

    Read only the references needed for the current request:

    line 31
  11. Very specific:

    Use this workflow when the user invokes the skill without any other instruction, for example just llm-wiki.

    line 41
  12. Distinctive:

    if the user specified a language in the current request, use it.

    line 44
  13. Very specific:

    If the user selected a language and wiki/config.md is absent or stale, create/update it before other wiki writes.

    line 48
  14. Distinctive:

    Use the user-specified working directory when provided.

    line 49
  15. Very specific:

    Create or update wiki/manifest.json with raw file paths, content hashes, target source pages, language, and status.

    line 51
  16. Distinctive:

    If the selected raw set is larger than 30 files, or the pending changed/new set is larger than the batch size, enter batch mode and process …

    line 52
  17. Distinctive:

    After the source-page pass, run a Concept synthesis pass using sub-agents/equivalent isolated workers:

    line 57
  18. Distinctive:

    ask for reusable cross-source concepts, merge candidates, and evidence links.

    line 59
  19. Very specific:

    Update wiki/index.md, wiki/manifest.json, and append wiki/log.md with the language, changed pages, raw coverage, concept synthesis status, …

    line 61
  20. Very specific:

    Keep the index compact: summarize raw coverage counts and link to wiki/manifest.json or a dedicated source catalog when needed instead of …

    line 62
  21. Very specific:

    Read selected raw/*.md files as evidence only.

    line 68
  22. Very specific:

    Update wiki/manifest.json before writing pages.

    line 69
  23. Distinctive:

    If there are more than 30 raw files or more pending files than the batch size, process only one batch at a time.

    line 70
  24. Distinctive:

    After source pages are drafted, run Concept synthesis with sub-agents/equivalent isolated context workers before the final index/log update.

    line 73
  25. Very specific:

    Update wiki/index.md, wiki/manifest.json, and append an entry to wiki/log.md for accepted writes.

    line 75
  26. Distinctive:

    Stop before full wiki generation unless execution scope explicitly allows it.

    line 77
  27. Very specific:

    Use this workflow whenever raw/ contains more than 30 Markdown files, or when the changed/new pending set is larger than the configured …

    line 81
  28. Distinctive:

    After each source batch, run a bounded Concept synthesis pass with sub-agents/equivalent isolated workers over the completed source pages …

    line 94
  29. Very specific:

    After every batch, update wiki/manifest.json, wiki/index.md, and wiki/log.md before continuing.

    line 95
  30. Distinctive:

    If work remains after a batch, report the remaining counts and next batch recommendation instead of silently attempting an unbounded full …

    line 97
  31. Very specific:

    Prefer wiki/index.md and relevant wiki//*.md pages for answered/cited knowledge.

    line 101
  32. Distinctive:

    After answering, decide whether the result contains durable knowledge: a reusable synthesis, comparison, decision, source map, or workflow …

    line 103
  33. Distinctive:

    If durable knowledge was produced, do not leave it only in chat history:

    line 104
  34. Distinctive:

    when the user requested wiki maintenance, compounding, or autonomous execution, perform the scoped update instead of merely offering it, …

    line 106

“Common” is measured: kept by at least 6 in 10 of the 77 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

Use this skill from any capable AI agent when the user asks to ingest notes into an LLM Wiki, maintain or query the wiki, lint wiki consistency, or update wiki indexes/logs/cross-links.

When the user invokes only llm-wiki or otherwise provides no additional task details, do not show a workflow menu. Treat the bare invocation as the default bootstrap/refresh pipeline: resolve the wiki language, then generate or update the wiki directly from raw/.

If the user does not explicitly specify a working directory or knowledge repository path, operate on the current session working directory. Resolve raw/, wiki/, and all relative links from that directory.

Safety contract

  • Treat raw/ as immutable source evidence: never edit, move, delete, normalize, or write generated files under it.
  • Treat wiki/ as the reviewable compiled Markdown layer.
  • Treat the session working directory as the knowledge repository root unless the user explicitly provides another directory.
  • Every generated or updated wiki page that is derived from raw material must link to its corresponding raw/ file(s), not only list them in frontmatter.
  • Do not install packages, call models/APIs, or generate a full wiki unless the user explicitly requests that execution step.
  • Before creating or updating wiki/ pages, determine the target wiki language. If the user did not explicitly specify it in the current request or accepted page plan, open a short choice/input dialog when the agent UI supports it, or ask in chat, and wait for the user to choose or specify the language.
  • Remember the selected wiki language in wiki/config.md using language: <value>, and reuse it on later invocations unless the user overrides it.
  • For many raw sources, use wiki/manifest.json for incremental batching instead of trying to generate the whole wiki in one pass. Default large-source threshold: more than 30 raw files. Default batch size: 20 raw files unless the user specifies another batch size.
  • Prefer small, reviewable wiki updates and record meaningful changes in wiki/log.md.
  • Keep wiki/index.md as a compact navigation catalog, not an ever-growing provenance dump. Do not append every raw file into an index Sources field/list. Store complete raw coverage in wiki/manifest.json, keep raw provenance on the relevant source/concept pages, and rebuild index source links from current wiki/sources/*.md pages with deduplication.
  • Source generation: generate or refresh source pages with sub-agents or equivalent isolated context workers whenever the agent environment supports them. Each worker should receive only a bounded raw-file slice plus the page schema and return source-page drafts/summaries; the main context should coordinate, review, and integrate rather than ingesting the entire raw corpus.
  • After generating or refreshing source pages, always run a bounded Concept synthesis pass. Use sub-agents or equivalent isolated context workers for this pass whenever the agent environment supports them, so the main context handles only plans, source-page summaries, proposed concepts, and final integration rather than every raw/source detail.

Reference map

Read only the references needed for the current request:

  • SKILL.md — canonical workflow behavior for language gating, batching, manifest lifecycle, Source generation, and Concept synthesis.
  • references/wiki-schema.md — directory layout, frontmatter, page types, index/log/config rules, page-plan guardrails.
  • references/ai-agent-integration.md — portable setup patterns for agents beyond the named Claude Code reference.
  • references/claude-code.md — Claude Code installation and usage notes.
  • references/first-pass-page-plan.md — initial page plan and sample style for the current <knowledge-repo> raw sources.

Workflow: bare invocation

Use this workflow when the user invokes the skill without any other instruction, for example just llm-wiki.

  1. Resolve the target wiki language:
    • if the user specified a language in the current request, use it;
    • otherwise read the remembered language from wiki/config.md when present;
    • otherwise, if wiki/index.md or existing maintained pages have a consistent language value, reuse it and write it to wiki/config.md;
    • otherwise ask: "请选择或指定本次 wiki 使用的语种(例如 zh-CN、en、bilingual 或其他)" and wait for the answer.
  2. If the user selected a language and wiki/config.md is absent or stale, create/update it before other wiki writes.
  3. Use the user-specified working directory when provided; otherwise use the current session working directory as the knowledge repository root.
  4. Inventory raw/*.md and existing wiki/**/*.md coverage without altering raw/.
  5. Create or update wiki/manifest.json with raw file paths, content hashes, target source pages, language, and status.
  6. If the selected raw set is larger than 30 files, or the pending changed/new set is larger than the batch size, enter batch mode and process only the next batch of pending raw files.
  7. Generate or update source pages with sub-agents/equivalent isolated workers, keeping the first pass small and reviewable unless the user explicitly requested full generation:
    • give each worker a bounded raw-file slice, target source-page paths, language, and source-page schema;
    • require clickable raw provenance links and concise evidence-bound summaries;
    • have workers return changed page paths, short page summaries, and any uncertainty/blockers for main-agent integration.
  8. After the source-page pass, run a Concept synthesis pass using sub-agents/equivalent isolated workers:
    • give each worker a bounded set of source pages or source-page summaries, not the whole raw corpus;
    • ask for reusable cross-source concepts, merge candidates, and evidence links;
    • integrate only accepted concepts into wiki/concepts/*.md, avoiding duplicate or over-fragmented pages.
  9. Update wiki/index.md, wiki/manifest.json, and append wiki/log.md with the language, changed pages, raw coverage, concept synthesis status, batch status, and verification.
    • Keep the index compact: summarize raw coverage counts and link to wiki/manifest.json or a dedicated source catalog when needed instead of expanding an unbounded raw-source list.

Workflow: ingest plan

  1. Identify the requested source set. For new or large sets, list proposed wiki pages before writing.
  2. Determine the target wiki language before any wiki write. If it is not explicit, first reuse wiki/config.md; if no remembered language exists, ask: "请选择或指定本次 wiki 使用的语种(例如 zh-CN、en、bilingual 或其他)" and do not proceed with wiki writes until the user answers.
  3. Read selected raw/*.md files as evidence only; do not alter them.
  4. Update wiki/manifest.json before writing pages. Mark raw files as new, changed, unchanged, missing-page, done, or skipped.
  5. If there are more than 30 raw files or more pending files than the batch size, process only one batch at a time. Default batch size is 20; respect explicit user input such as "batch size 50".
  6. Map each raw source in the current batch to at least one wiki/sources/*.md page and at most a small justified set of concept/workflow pages.
  7. Draft source pages in the chosen wiki language through sub-agents/equivalent isolated workers, with provenance and language in frontmatter, concise claim summaries, and explicit Markdown links to the corresponding raw file(s). The main agent should pass bounded raw slices to workers and integrate their drafts, not read every raw source into one context.
  8. After source pages are drafted, run Concept synthesis with sub-agents/equivalent isolated context workers before the final index/log update. Each worker should inspect only a bounded slice of source pages plus existing concept titles/summaries, then return candidate concepts with evidence links.
  9. Draft concept/workflow pages in the chosen wiki language when they synthesize multiple sources, merge repeated ideas, or encode reusable procedure; concept pages must link to the relevant source pages and raw file(s). If no new concept qualifies, record that the Concept pass found no durable synthesis.
  10. Update wiki/index.md, wiki/manifest.json, and append an entry to wiki/log.md for accepted writes.
  • Rebuild the index deterministically from maintained pages; do not append duplicate or historical Sources entries.
  1. Stop before full wiki generation unless execution scope explicitly allows it.

Workflow: large raw sets

Use this workflow whenever raw/ contains more than 30 Markdown files, or when the changed/new pending set is larger than the configured batch size.

  1. Build or refresh wiki/manifest.json by hashing every raw/**/*.md file. Prefer SHA-256 when local tooling is available.
  2. Compare each raw file with the manifest:
    • new: raw file not in manifest;
    • changed: hash differs from manifest;
    • unchanged: hash matches and target source page exists;
    • missing-page: hash matches but target source page is missing;
    • done: processed and verified in the current language;
    • skipped: intentionally not ingested, with a reason.
  3. Select the next batch in this priority order: changed, missing-page, new.
  4. Process at most the batch size. Default: 20 raw files. User override examples: batch size 50, 每批 10 个.
  5. Generate source pages first with sub-agents/equivalent isolated workers over bounded raw-file slices. Do not load all raw files into the main context for source-page drafting.
  6. After each source batch, run a bounded Concept synthesis pass with sub-agents/equivalent isolated workers over the completed source pages for that batch plus concise summaries of existing concepts. Do not load all raw files into the main context for concept discovery.
  7. After every batch, update wiki/manifest.json, wiki/index.md, and wiki/log.md before continuing.
    • For large sets, wiki/index.md should show counts, batch status, and links to source pages or grouped source catalogs; the exhaustive raw-file inventory belongs in wiki/manifest.json.
  8. If work remains after a batch, report the remaining counts and next batch recommendation instead of silently attempting an unbounded full generation.

Workflow: query

  1. Prefer wiki/index.md and relevant wiki/**/*.md pages for answered/cited knowledge.
  2. Cite wiki pages and their raw provenance.
  3. After answering, decide whether the result contains durable knowledge: a reusable synthesis, comparison, decision, source map, or workflow clarification that future queries should find.
  4. If durable knowledge was produced, do not leave it only in chat history:
    • for ordinary query prompts, offer a scoped wiki update with a proposed page title/path and the provenance that would be filed;
    • when the user requested wiki maintenance, compounding, or autonomous execution, perform the scoped update instead of merely offering it, but first apply the wiki-language gate if the target language was not explicit.
  5. For accepted or performed durable updates, write the synthesis into the smallest suitable wiki/ page, update wiki/index.md when navigation changes, and append a wiki/log.md entry.

Workflow: lint

Check for:

  • source pages missing raw provenance;
  • raw sources not represented in wiki/sources/;
  • orphan or duplicate concept pages;
  • over-fragmented pages that should be merged;
  • stale wiki/index.md entries;
  • missing wiki/log.md entries for wiki maintenance;
  • contradictions between summaries and raw provenance;
  • missing or inconsistent language values across maintained pages.
  • stale or missing wiki/manifest.json entries, including raw hashes that no longer match source pages.

Report proposed fixes before large rewrites.

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.