llm-wiki

Run retrieval benchmarks (LME / LoCoMo / ConvoMem) from the vault; optional LLM rerank via API or local CLI.

Build static graph viewer + wiki-data.json into llm-wiki/wiki/.og/

Change vault settings after setup — MCP wiki_configure, CLI configure -i, or wiki-setup sections.

Show vault git diff (working tree or staged).

Audit vault git history by lifecycle phase (ingest, wiki, build, …) for flow progression.

Vault git log — history for the llm-wiki vault only.

Commit current vault state with a message (vault-scoped git).

Vault-scoped git status (requires git.enabled in llm-wiki/config.json).

Generate a knowledge/cluster graph — pages colored by undirected link component (relational clusters).

Generate an on-demand D3 link graph of llm-wiki/wiki into .tmp/llm-wiki-graph (GitHub-dark styled).

Land sources in raw/ via llm-wiki ingest (any adapter), then merge via wiki-ingest. Generic web/source principles and phased playbook.

Configure optional ingest integrations — Firecrawl, Playwright (headless browser), Brave Search, Perplexity, Twitter, PDF adapters, Anna's Archive, and more. Shows status, walks through setup, validates keys.

Health-check the wiki — contradictions, orphans, gaps.

MCP server + editor config — stdio JSON-RPC (default) or HTTP transport; search/KG backends per llm-wiki/config.json.

List, search, save, or prune per-chat session memory under raw/memory/ (see wiki-session-memory).

Answer from the wiki with citations; optionally file the answer back into wiki/.

Validate and clean raw/ markdown; log preparation; optional vault git [prepare] commit.

Run recurring research tasks from research-tasks.json / research-tasks.yaml (wiki-research-loop skill).

Ad-hoc research on a topic — discover sources, ingest into raw/, merge into wiki/, with stepped progress and logging (wiki-research skill).

Root setup entry — vault (llm-wiki) and/or session memory. Full wizard body lives in skills/wiki-setup/SKILL.md.

Vault health — setup flag, integrations, tools, MCP search/KG; follow wiki-status skill.

Long-running research passes over many URLs or HN items; writes raw artifacts then hands off to wiki-ingest patterns.

Large multi-file edits across llm-wiki/wiki — index, cross-links, batch updates. Prefer llm-wiki git CLI when git.enabled.

Cleans and validates noisy raw/ markdown (HTML/PDF/OCR) into maintainable Markdown before wiki-ingest.

Extract Crunchbase company profiles, funding rounds, and investor data into raw/. Use for startup, investor, or market research.

Extract text from EPUB, MOBI, PDF, and AZW3 ebook files into raw/. Use when user has a local ebook or PDF to add to the vault.

Extract product listings, pricing, and auction data from Amazon, eBay, Etsy, Shopify into raw/. Use for product research or price tracking.

Extract GitHub repo README, issues, PRs, releases, and changelogs into raw/. Use when a GitHub URL is shared or tracking a project.

Extract LinkedIn profiles, jobs, and posts into raw/. Uses Firecrawl, archive.ph, Google cache, or manual export bypass hierarchy.

Extract Substack, Beehiiv, Ghost, Buttondown newsletter issues into raw/. Handles subscriber-only content via archives, RSS, and email.

Extract patent text, claims, and citations from Google Patents, Espacenet, USPTO into raw/. Use for IP and prior art research.

Bypass paywalls to get full article text into raw/. Uses Freedium, archive.ph, Wayback Machine, 12ft.io. Auto-invoked on 403/paywall.

Extract podcast episodes from RSS and transcribe via Whisper or AssemblyAI into raw/. Use for podcast URLs or episode links.

Extract Wikipedia articles and Wikidata infobox data into raw/ via REST API and SPARQL. Use for encyclopedia-level reference content.

Extract YouTube transcripts, captions, and metadata into raw/. Uses yt-dlp or pytube. Supports audio transcription when no captions exist.

Index and routing for source-specific vault extractors (wiki-extract-*). Use to pick which extractor or reference applies to a URL, file, or intent.

Orchestrate fetching a URL or research query into the vault using the best available adapter.

Merges new raw sources into llm-wiki/wiki/ with index and log updates. Use after llm-wiki ingest or when user drops files into raw/. Works for any source on the public web or local files.

Cross-session memory for the vault — learn, recall, and review patterns, preferences, pitfalls, and next steps in llm-wiki/.agent-memory.md.

Audits llm-wiki/wiki for orphans, stale claims, missing links, and contradictions. Use when user asks to health-check the wiki.

Maintains the LLM Wiki vault — index, log, cross-links, contradictions. Use when editing llm-wiki/wiki/, merging ingested sources, or keeping the wiki consistent.

End-to-end vault workflow — status, research/fetch, raw prepare, wiki ingest, lint, build-site, validate — with optional user gates between stages.

Answers questions using llm-wiki/wiki pages with citations. Use when the user asks about vault content or synthesized knowledge.

Validate and clean raw/ markdown (HTML/PDF/OCR) with deterministic checks + LLM formatting. Records in preparation-log.jsonl, aligns with git prepare.

Research academic papers from arXiv, DOI, or Semantic Scholar and ingest into raw/academic/. Invoked by wiki-research for paper requests.

Multi-pass research — runs sub-skills in sequence, writes synthesis draft to outputs/, merges into wiki/ after review gate. For broad/complex topics.

Monitor RSS/Atom feeds or OPML lists and ingest recent items into raw/feeds/. Invoked by wiki-research for feed/newsletter research.

Batch research from research-tasks.json — fetch, ingest, wiki merge. Use when research_loop.enabled for automated passes (e.g. HN front page).

Research current events using Perplexity or news APIs and ingest into raw/news/. Invoked by wiki-research for recency-first queries.

Fetch Twitter/X threads, Hacker News discussions, or Reddit posts into raw/social/. Invoked by wiki-research for social/community content.

Fetch web URLs into raw/ using Firecrawl CLI/REST or stdlib fallback. Invoked by wiki-research for general web pages.

Ad-hoc research orchestrator — classifies request, picks sub-skill, ingests, and merges into wiki/. Not wiki-research-loop (batch tasks).

Vault retrospective — activity, source mix, topic coverage, health, open questions, and recommendations from logs, git, and wiki-lint.

Per-chat session memory in raw/memory/ — save, recall, list, prune. Use when the user wants continuity across Claude sessions or to audit past work.

Interactive vault setup wizard — asks questions, previews config.json, scaffolds directories. Use for "setup", "configure", or "initialize".

Vault health dashboard — setup state, integrations, CLI, packages, MCP, git, config. Use for "status", "health check", or pre-research.

Update the llm-wiki plugin repo (git pull), re-run setup, and sanity-check vault config against the new version.

Search and download ebooks, papers, and documents from Anna's Archive into raw/. Requires ANNAS_ARCHIVE_KEY.

6 hooks across 4 events