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
- StevenGuo42/llm-wiki-skillstheir own version, first seen Nov 15, 2026
- jellydn/my-ai-toolstheir own version, first seen Nov 15, 2026
- infranodus/skillstheir own version, first seen Nov 11, 2026
- Damon-GSY/skillstheir own version, first seen Nov 8, 2026
- DracoUnion/skillstheir own version, first seen Nov 8, 2026
- Ar9av/obsidian-wikitheir own version, first seen Nov 8, 2026
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-chenSkilltouchdesigner-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-chenSkillteams-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
mae616Skillp5js
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-incSkillheartmula
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-Chen1999Skillclaude-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-Chen1999Skillgraph-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
Other versions95 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 95
- Ar9av/obsidian-wiki★ 3.7ksame name
- sdyckjq-lab/llm-wiki-skill★ 2.5ksame name
- staruhub/ClaudeSkills★ 705same name
- lewislulu/llm-wiki-skill★ 654same name
- zosmaai/pi-llm-wiki★ 622same name
- Tommy-yw/RunbookHermes★ 546same name
- luotwo/llm-wiki★ 218same name
- sammcj/agentic-coding★ 169same name
- MehmetGoekce/llm-wiki★ 163same name
- HybridAIOne/hybridclaw★ 162same name
85 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
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
- Very specific: line 8
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 …
- Very specific: line 10
When the user invokes only llm-wiki or otherwise provides no additional task details, do not show a workflow menu.
- Very specific: line 12
If the user does not explicitly specify a working directory or knowledge repository path, operate on the current session working directory.
- Very specific: line 16
Treat raw/ as immutable source evidence: never edit, move, delete, normalize, or write generated files under it.
- Very specific: line 17
Treat wiki/ as the reviewable compiled Markdown layer.
- Distinctive: line 18
Treat the session working directory as the knowledge repository root unless the user explicitly provides another directory.
- Very specific: line 19
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 …
- Distinctive: line 20
Do not install packages, call models/APIs, or generate a full wiki unless the user explicitly requests that execution step.
- Very specific: line 24
Prefer small, reviewable wiki updates and record meaningful changes in wiki/log.md.
- Distinctive: line 31
Read only the references needed for the current request:
- Very specific: line 41
Use this workflow when the user invokes the skill without any other instruction, for example just llm-wiki.
- Distinctive: line 44
if the user specified a language in the current request, use it.
- Very specific: line 48
If the user selected a language and wiki/config.md is absent or stale, create/update it before other wiki writes.
- Distinctive: line 49
Use the user-specified working directory when provided.
- Very specific: line 51
Create or update wiki/manifest.json with raw file paths, content hashes, target source pages, language, and status.
- Distinctive: line 52
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 …
- Distinctive: line 57
After the source-page pass, run a Concept synthesis pass using sub-agents/equivalent isolated workers:
- Distinctive: line 59
ask for reusable cross-source concepts, merge candidates, and evidence links.
- Very specific: line 61
Update wiki/index.md, wiki/manifest.json, and append wiki/log.md with the language, changed pages, raw coverage, concept synthesis status, …
- Very specific: line 62
Keep the index compact: summarize raw coverage counts and link to wiki/manifest.json or a dedicated source catalog when needed instead of …
- Very specific: line 68
Read selected raw/*.md files as evidence only.
- Very specific: line 69
Update wiki/manifest.json before writing pages.
- Distinctive: line 70
If there are more than 30 raw files or more pending files than the batch size, process only one batch at a time.
- Distinctive: line 73
After source pages are drafted, run Concept synthesis with sub-agents/equivalent isolated context workers before the final index/log update.
- Very specific: line 75
Update wiki/index.md, wiki/manifest.json, and append an entry to wiki/log.md for accepted writes.
- Distinctive: line 77
Stop before full wiki generation unless execution scope explicitly allows it.
- Very specific: line 81
Use this workflow whenever raw/ contains more than 30 Markdown files, or when the changed/new pending set is larger than the configured …
- Distinctive: line 94
After each source batch, run a bounded Concept synthesis pass with sub-agents/equivalent isolated workers over the completed source pages …
- Very specific: line 95
After every batch, update wiki/manifest.json, wiki/index.md, and wiki/log.md before continuing.
- Distinctive: line 97
If work remains after a batch, report the remaining counts and next batch recommendation instead of silently attempting an unbounded full …
- Very specific: line 101
Prefer wiki/index.md and relevant wiki//*.md pages for answered/cited knowledge.
- Distinctive: line 103
After answering, decide whether the result contains durable knowledge: a reusable synthesis, comparison, decision, source map, or workflow …
- Distinctive: line 104
If durable knowledge was produced, do not leave it only in chat history:
- Distinctive: line 106
when the user requested wiki maintenance, compounding, or autonomous execution, perform the scoped update instead of merely offering it, …
“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.
Files
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.mdusinglanguage: <value>, and reuse it on later invocations unless the user overrides it. - For many raw sources, use
wiki/manifest.jsonfor 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.mdas a compact navigation catalog, not an ever-growing provenance dump. Do not append every raw file into an indexSourcesfield/list. Store complete raw coverage inwiki/manifest.json, keep raw provenance on the relevant source/concept pages, and rebuild index source links from currentwiki/sources/*.mdpages 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.
- 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.mdwhen present; - otherwise, if
wiki/index.mdor existing maintained pages have a consistentlanguagevalue, reuse it and write it towiki/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.
- If the user selected a language and
wiki/config.mdis absent or stale, create/update it before other wiki writes. - Use the user-specified working directory when provided; otherwise use the current session working directory as the knowledge repository root.
- Inventory
raw/*.mdand existingwiki/**/*.mdcoverage without alteringraw/. - Create or update
wiki/manifest.jsonwith raw file paths, content hashes, target source pages, language, and status. - 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.
- 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.
- 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.
- Update
wiki/index.md,wiki/manifest.json, and appendwiki/log.mdwith 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.jsonor a dedicated source catalog when needed instead of expanding an unbounded raw-source list.
- Keep the index compact: summarize raw coverage counts and link to
Workflow: ingest plan
- Identify the requested source set. For new or large sets, list proposed wiki pages before writing.
- 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. - Read selected
raw/*.mdfiles as evidence only; do not alter them. - Update
wiki/manifest.jsonbefore writing pages. Mark raw files asnew,changed,unchanged,missing-page,done, orskipped. - 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".
- Map each raw source in the current batch to at least one
wiki/sources/*.mdpage and at most a small justified set of concept/workflow pages. - Draft source pages in the chosen wiki language through sub-agents/equivalent isolated workers, with provenance and
languagein 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. - 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.
- 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.
- Update
wiki/index.md,wiki/manifest.json, and append an entry towiki/log.mdfor accepted writes.
- Rebuild the index deterministically from maintained pages; do not append duplicate or historical
Sourcesentries.
- 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.
- Build or refresh
wiki/manifest.jsonby hashing everyraw/**/*.mdfile. Prefer SHA-256 when local tooling is available. - 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.
- Select the next batch in this priority order:
changed,missing-page,new. - Process at most the batch size. Default: 20 raw files. User override examples:
batch size 50,每批 10 个. - 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.
- 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.
- After every batch, update
wiki/manifest.json,wiki/index.md, andwiki/log.mdbefore continuing.- For large sets,
wiki/index.mdshould show counts, batch status, and links to source pages or grouped source catalogs; the exhaustive raw-file inventory belongs inwiki/manifest.json.
- For large sets,
- If work remains after a batch, report the remaining counts and next batch recommendation instead of silently attempting an unbounded full generation.
Workflow: query
- Prefer
wiki/index.mdand relevantwiki/**/*.mdpages for answered/cited knowledge. - Cite wiki pages and their raw provenance.
- After answering, decide whether the result contains durable knowledge: a reusable synthesis, comparison, decision, source map, or workflow clarification that future queries should find.
- 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.
- For accepted or performed durable updates, write the synthesis into the smallest suitable
wiki/page, updatewiki/index.mdwhen navigation changes, and append awiki/log.mdentry.
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.mdentries; - missing
wiki/log.mdentries for wiki maintenance; - contradictions between summaries and raw provenance;
- missing or inconsistent
languagevalues across maintained pages. - stale or missing
wiki/manifest.jsonentries, including raw hashes that no longer match source pages.
Report proposed fixes before large rewrites.
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
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.

