/SKILL.md

name, description

name

description

all-in-one-llm-wiki

Build an all-in-one AI-readable LLM Wiki from personal exports and app data. Use when a user wants to ingest digital-life data into a private wiki for AI agents, including Obsidian notes, AI memory exports, Bilibili data via CLI, Garmin data via script, Apple Health export, Spotify, Douban, browser history, YouTube/Google Takeout, Steam, calendars, coffee or lifestyle logs, service accounts, NAS/media libraries, and then create a privacy-safe narrative or PPT prompt about the system.

All in One LLM Wiki

Purpose

Create a private, AI-readable wiki that helps agents understand a user's long-term context. Optimize for agent retrieval, provenance, privacy, and downstream tasks, not for human-perfect publishing.

Do not copy another user's personal profile into outputs. Treat all examples as schema examples, not facts about the current user.

Core Workflow

1. Inventory available sources and classify them as notes, AI memory, media taste, browsing/tooling, health, lifestyle, or accounts.
2. Read references/privacy.md before handling raw exports, browser history, health data, family data, account data, or any publishable artifact.
3. Use references/data-sources.md for source-specific ingestion methods.
4. Use references/wiki-schema.md to create wiki pages, index entries, raw source storage, and logs.
5. Summarize each source into agent-facing pages: facts, preferences, recurring behaviors, constraints, evidence, and open questions.
6. For a new wiki, use scripts/init_wiki.py to create the skeleton; use --demo only for synthetic public-safe examples.
7. Create a privacy-safe story or presentation only after the wiki layer exists. Use references/presentation.md and assets/ppt-prompt-template.md.
8. Before publishing any demo, deck, screenshot, or prompt, use PUBLISHING_CHECKLIST.md.

Output Shape

Prefer this folder layout:

.wiki/
  SCHEMA.md
  index.md
  log.md
  raw/
    articles/
    ai-memory/
    health/
    platform-exports/
  entities/
  concepts/
  queries/
  _archive/

Every useful page should include:

• frontmatter: title, created, updated, type, tags, sources, confidence
• an overview table
• concrete agent affordances: "What an agent can now do with this"
• cross-links to related entities
• a privacy note when the page is sensitive

Ingestion Rules

• Keep raw exports private by default.
• Store exact raw data only when the user explicitly wants it and the destination is private.
• For browser, health, family, account, and location data, prefer aggregate summaries plus checksums.
• Record source date, extraction method, command, and parser assumptions in log.md.
• Distinguish facts from inferred preferences. Use confidence: low|medium|high.
• Add a "needs confirmation" section for inferred or ambiguous items.

Agent Affordance Pattern

For each imported source, write what it enables:

## Agent affordances

- Recommend actions based on real user history, not generic popularity.
- Avoid suggestions that conflict with known constraints.
- Explain why a recommendation fits, citing the source page.
- Ask for confirmation before using sensitive or low-confidence inferences.

Examples:

• Music data lets an agent choose playlists by task, mood, or context.
• Douban or movie/book data lets an agent recommend media by actual taste and prior ratings.
• Bilibili/YouTube data lets an agent understand current learning and entertainment streams.
• Browser aggregates let an agent understand tooling, projects, and attention distribution.
• Health data lets an agent suggest rest, training, or routine changes with realistic context.
• AI memory exports let an agent merge scattered self-knowledge across platforms.

Resources

• references/data-sources.md: source-specific import methods.
• references/wiki-schema.md: page schema, index, log, linking, and confidence conventions.
• references/privacy.md: redaction and publishability rules.
• references/presentation.md: converting the wiki process into a video/PPT narrative.
• assets/ppt-prompt-template.md: prompt template for Kimi, Claude, Gamma, or other PPT tools.
• PUBLISHING_CHECKLIST.md: final public-release redaction checklist.
• examples/demo-wiki/: synthetic demo wiki for first-run testing.
• scripts/init_wiki.py: create a private .wiki/ skeleton or copy the synthetic demo.
• scripts/garmin_to_wiki_example.py: generic Garmin Connect to wiki JSON/Markdown starter script.