Wiki Query
Purpose
Answer a question using the structured wiki (stored in MemPalace) as the knowledge source. Returns cited synthesis with [[wikilinks]], mandatory Counter-Arguments, and explicit knowledge gaps.
Use BEFORE going to the web. Wiki knowledge — when present — is faster, project-aware, and brand-voice-compliant.
When To Invoke
- User asks a question that existing wiki knowledge could answer
- Synthesis, research briefings, or pattern detection requests
- Cross-project queries ("what overlaps between iSimplification and SentinelFlow?")
- Before any external research (per
documents/reference/AGENTS.md— wiki-first principle)
Inputs
query: The question to answerfilters(optional):startup: scope to a single startup's pagesconfidence_min: high | medium | lowexplored_only: boolean (default true — exclude unverified pages)tags: array of tag filters
Workflow
- Hybrid search via MemPalace:
mempalace_search(query, limit=10)— semantic + keyword- Filter results:
explored: trueonly (unless--include-unexplored) - Filter by
confidence_minand tags
- Read top 3-7 most relevant pages in full via
mempalace_search(full content) or specific drawer fetch. - Graph traversal (optional, for synthesis questions):
mempalace_kg_query(entity, direction="both")to find related conceptsmempalace_traverse(start_room, max_hops=2)to discover cross-domain links
- Synthesize cited answer:
- Lead with a direct 2-sentence answer
- Cite every claim with
[[page-name]]inline - Include Counter-Arguments from cited pages (mandatory)
- End with: "Knowledge gaps: [list any missing coverage]"
- Apply brand voice (if output is for external use):
- Load
.brand-foundation/personal/voice-profile.md - Detect active startup → load
.brand-foundation/startups/{id}/positioning.md - Run
.brand-foundation/personal/ai-tells.mdbanned-phrase scan
- Load
- File substantive answers as new wiki pages in
wing_knowledge/outputs/[query-slug]withexplored: false.
Output Format
Wiki Query Result:
Query: "<original query>"
Pages consulted: [N]
Confidence: HIGH | MEDIUM | LOW
Direct Answer:
<2-sentence answer>
Synthesis:
<Cited body using [[page-name]] references>
Counter-Arguments (from cited pages):
- <gap 1>
- <opposing evidence 1>
Knowledge Gaps:
- <topics with weak/no wiki coverage — recommend `/mxm-wiki ingest` after web research>
Confidence: 🟢 HIGH (≥3 cited pages, all explored:true) | 🟡 MEDIUM (some pages explored:false) | 🔴 LOW (no relevant pages found — recommend web search + ingest cycle)
When To Hedge
- Pages cited are
explored: false→ flag with: "⚠️ Drawing on unverified content. Run/mxm-wiki exploreto confirm." confidence: lowpages → cite as "informal" not authoritative- Knowledge gap detected → recommend
/mxm-wiki ingestafter closing the gap externally
MemPalace Integration
mempalace_search({
query: "<query>",
wing: "wing_knowledge",
room: null, // search across rooms
limit: 10
})
mempalace_kg_query({
entity: "<key entity from query>",
direction: "both"
})
mempalace_traverse({
start_room: "<most relevant room>",
max_hops: 2
})
Handoff
- Substantive answer filed as wiki page → notify
wiki-explore(it'sexplored: falseby default) - Knowledge gaps identified → suggest research task to
innovation-researcherortrend-researcher, thenwiki-ingestafter - External-use answer → load
.brand-foundation/+ runai-tells.mdscan before delivery - Compliance-sensitive answer → loop
compliance-officerfor jurisdiction validation
Maxim Behavioral Layer
- Confidence tagging: 🟢 HIGH (≥3 explored:true pages, citations dense) | 🟡 MEDIUM (mixed verification) | 🔴 LOW (<2 pages or all unexplored)
- Brand foundation enforcement: applies to ALL externally-facing wiki-query outputs
- Anti-hallucination: refuse to extrapolate beyond cited pages — explicit gaps go in Knowledge Gaps section
Source References
FINAL_RELEASE_CHANGES/Brandvoice-wiki-example.md— wiki-query patterndocuments/reference/AGENTS.md— wiki-first principle.claude/skills/memory-palace/SKILL.md— MemPalace tools
Copyright (c) 2026 iSystematic Inc. Maxim product. BSL 1.1.