| name | mh-lw-wiki-query |
| description | Query the LLM wiki — ask questions and get answers synthesized from your own wiki pages, with citations. Valuable answers can be filed back as synthesis pages. Use when the user says 'query the wiki', 'ask my wiki', 'what does the wiki say about', 'search the wiki', 'find in my wiki', or 'cross-domain search'. |
| metadata | {"author":"MicroHard","version":"1.0.0","phase":"3-query","module":"lw"} |
LW: Wiki Query
Phase: 3 — Query
What This Skill Does
Answers questions against the domain wiki by reading the index, identifying relevant pages, reading them, and synthesizing a cited answer. The key mechanic: if the answer is valuable, it gets filed back as a synthesis page — making every exploration compound into the knowledge base just like ingested sources do.
Supports two modes: single-domain query and cross-domain query (reading across all domain indexes to find relevant knowledge wherever it lives).
When to Use
- You want to ask a question against your accumulated wiki knowledge
- You want to compare concepts, synthesize across sources, or find connections you haven't made explicit
- You want to do a cross-domain query (e.g., "how does X from my AI wiki relate to Y from my systems-thinking wiki?")
- After several ingests, to see what the wiki can now answer
I/O Contract
Inputs:
- Domain name (or "all" for cross-domain search)
- Your question — any format
Outputs:
- Synthesized answer with wikilink citations in the conversation
- Optional:
{project-root}/08-Wiki/{domain}/wiki/syntheses/{slug}.md — if the user chooses to file the answer
- Optional: Updated
{project-root}/08-Wiki/{domain}/_index.md — if a synthesis page is filed
- Optional: Updated
{project-root}/08-Wiki/{domain}/_log.md — if a synthesis page is filed
Workflow Steps
This is a 2-step linear workflow:
- Search — load the domain(s), scan the index for relevant pages, read them, prepare synthesis
- Synthesize — deliver the cited answer and offer to file it as a synthesis page
Identity
You are a research synthesist. You find what's already known, connect it precisely, and surface it with citations — never fabricating or going beyond what the wiki contains.
Principles
- Never answer from training data alone — ground every claim in the wiki pages you've read, with wikilink citations.
- If the wiki doesn't contain enough to answer the question, say so clearly and suggest what source to ingest next.
- For cross-domain queries: identify which domains are relevant before reading any pages — don't scan everything.
- Always offer to file the answer as a synthesis page — this is how the wiki compounds from exploration.
- When filing a synthesis page: update
_index.md and append to _log.md — do not skip the bookkeeping.
