Wiki Brain Operating Model

The wiki should behave like a compiled brain: coherent articles, stable retrieval, generated inventories, explicit provenance, and routing that makes the next edit obvious.

This is the shape spec for broad wiki changes. Read it before consolidation, recategorization, page creation, or any rewrite that could change the graph. Then use Filing Decision Tree for directory placement.

For corpus-wide merge, fold, archive, redirect, regeneration, split, or deletion work, use Wiki Corpus Refactor Program. This page defines the desired brain shape; the refactor program defines how the existing corpus migrates without losing evidence or retrieval. Source: User request, 2026-07-15

The durable unit is not a captured thought, source, skill, or folder path. The durable unit is a coherent mental object: a page, section, ledger row, generated view, or redirect that future agents can retrieve and update without reconstructing the original chat.

Current Brain State

The current wiki is no longer just flat article pages. It has canonical synthesis pages, operational protocols, a typed brain graph, workflow manifests, generated skill/source views, identity/cadence extractions, and redirect ledgers. The filing model has to account for all of them. Source: User request, 2026-08-10

Surface Role
wiki/RESOLVER.md Directory-level page filing.
wiki/meta/ Operating model, protocols, quality gates, sync, automation, generated-index docs.
wiki/resolver/ Compact retrieval companions for filing rules.
wiki/skill-resolver/ Compact retrieval companions for task/tool/skill dispatch.
wiki/soul/, wiki/user/, wiki/heartbeat/ Extracted identity, Kevin model, and cadence expansions.
wiki/agent-docs/ Compact repository-discovery pointers into canonical system owners.
wiki/skills/<category>/ Generated source views for executable skills.
wiki/concepts/*-skills.md and Skill Registry Readable skill-family synthesis and generated inventory.
wiki/meta/wiki-redirects.json and Wiki Redirect Index Retired slug retrieval.
Generated Surface Contract Source/projection ownership for indexes, graph views, registries, redirects, qmd, and stateful ledgers.

Thesis

Flatten meaning. Use folders to clarify ownership, not to manufacture ontology.

A good wiki page should be useful to a human reader and to an agent. It should explain what is true now, preserve why we believe it, and make the next action obvious. If a page only exists because a source was captured once, it is probably the wrong shape.

Design Targets

Target Rule
Readability A human should understand the article without knowing the ingestion history.
Retrieval Old names, aliases, and redirects must still resolve after merging.
Compression Merge low-signal pages into stronger parent articles, ledgers, timelines, or tables.
Evidence Preserve source trails in timelines, citations, source pages, raw captures, or generated ledgers.
Agent usability Every article should answer "when do I use this knowledge?"
Goal effect Important knowledge should name the goal, project, decision, workflow, or output it can improve; weak current relevance changes priority, not retention.
Maintenance Generated inventories stay generated; hand-written pages own synthesis.

Brain Architecture

The raw source archive, compiled wiki, user/interaction memory, and session/run state are separate owners. The wiki precomputes durable synthesis so agents do not rediscover structure on every query. It does not replace exact source retrieval, and it does not automatically remember Kevin's later correction, a rejected approach, or a workflow outcome unless those interaction facts are explicitly written to the scoped operating graph. Source: X/@hwchase17, “Wiki Memory,” 2026-06-30; X/@mem0ai, “The State of Agent Wikis,” 2026-07-21

Page Layers

Layer Purpose Authoring rule
Canonical synthesis article Readable mental object. Hand-written, source-backed, revised in place.
Directory hub or README Local ownership and navigation. Hand-written and short.
Tool/ecosystem page Facts, setup, tradeoffs, interfaces, alternatives. Hand-written, current enough, source-backed.
Workflow page Human/team process or operational runbook. Hand-written checklist plus failure modes.
Executable skill Actual agent procedure. Lives in skills/{engineering,productivity,personal,misc,in-progress,deprecated}/<slug>/SKILL.md.
Generated skill source page Browseable view of executable skill source. Generated under wiki/skills/<category>/; do not hand-edit generated bodies.
Skill registry/family page Inventory plus readable capability synthesis. Registry is generated; family pages are synthesis-backed.
Source/evidence entry Provenance, raw extraction, or timeline material. Preserve evidence; do not pretend it is synthesis.
Ledger Compact store for low-signal entities. Table/list with aliases and promotion criteria.
Brain graph projection Machine-readable goals, loops, object bindings, and proof. Update the canonical graph inputs and validate the runtime.
Redirect Retired slug that should still resolve. Store in wiki/meta/wiki-redirects.json; regenerate Wiki Redirect Index.

Category Shape

Use Filing Decision Tree for final placement. This model describes why the placement exists.

Category family Owns
Article categories architecture/, career/, concepts/, decisions/, design/, interview-prep/, people/, philosophies/, postmortems/, projects/, research/, stories/, style/, tools/, workflows/.
Operational categories meta/, resolver/, skill-resolver/, soul/, user/, heartbeat/, inbox/.
Generated or asset categories agent-docs/, skills/, assets/.
Dormant schema categories places/, patterns/, tensions/, eras/, transitions/ when a real cluster justifies them.

Everything else should justify itself. If a category mostly contains stubs, fold those stubs into a stronger article, ledger, timeline, or generated view.

Flattening Rule

Prefer a strong article with sections and decision tables over a deep tree of thin pages. Use subfolders only when they improve browsing, enforce generation boundaries, or preserve a mature domain.

Good reasons for subfolders:

  • Generated namespaces such as wiki/skills/<category>/ and wiki/agent-docs/.
  • Resolver companion pages that provide retrieval handles but point back to canonical authority.
  • Identity/cadence expansions that keep boot files readable.
  • Asset provenance and media groupings.
  • Mature domains with multiple real standalone articles.

Folders are navigation. Identity lives in canonical slugs, titles, aliases, redirects, and qmd retrieval.

Article Standard

Every canonical article should include the subset of these sections that helps the topic:

Section Purpose
One-line thesis What the page lets the reader understand or decide.
Routing summary When humans or agents should open it.
Core model The actual explanation.
Decision matrix Tradeoffs, routes, comparisons, ownership boundaries.
Tool/skill routing What to load or use when the page changes behavior.
Failure modes Common mistakes and confusions.
Evidence/timeline Source trail without turning the article into a diary.
Related pages Neighboring mental objects.

Small pages do not need every section. Long pages need enough structure to be scanned.

Combining Rule

When merging pages, preserve ideas by changing their form:

Old page shape New form
One-off thought Section in a canonical article.
Skill mirror Generated source page plus row in Skill Registry.
Similar tools Comparison table in an ecosystem/tool article.
Minor person/entity Row in a ledger with aliases and promotion criteria.
Source summary Timeline entry, raw source reference, or provenance note.
Duplicate concept Alias, redirect, and merged section.
Narrow procedure Executable skill or workflow subsection.
Resolver micro-rule Companion page under wiki/resolver/ that points back to Filing Decision Tree.

No idea gets deleted because it is small. It gets promoted, folded, generated, or redirected.

Rewrite Procedure

  1. Pick a cluster, not a pair of pages.
  2. List pages, aliases, source files, skills, tools, workflow bindings, graph objects, and redirects in the cluster.
  3. Choose the canonical article owner.
  4. Move useful claims into the owner as sections, rows, matrices, or timeline entries.
  5. Preserve source trails, citations, dates, and embeds.
  6. Add aliases for old titles.
  7. Add redirects for retired slugs.
  8. Update wikilinks when the old link should no longer resolve through redirects.
  9. Regenerate indexes, qmd, registries, graph projections, and audits.
  10. Log the rewrite.

Promotion And Demotion

Action Trigger
Promote to standalone page Durable retrieval value, multiple sources, decisions, active use, or future agent trigger.
Keep as section The idea only makes sense inside a broader topic.
Keep as ledger row The entity matters but does not justify narrative context.
Keep as generated page It is inventory, executable source, or derived context.
Keep as evidence Provenance matters more than synthesis.
Keep as redirect Old slug/title still matters for retrieval.
Keep as session-only No durable retrieval value.

Validation Loop

After any structural rewrite:

npm run build-index
npm run wiki-quality
npm run routing-doctor
npm run brain
npx tsx scripts/doctor.ts --quiet
qmd update && qmd embed

Use npm run wiki:resolve -- <old-slug> to verify retired names still land on the intended article.


Timeline

83 pages link here

Active Stack (What to Actually Use)MetaAgent Company OSArchitectureAgent Docs FilingFiling Decision TreeAgent Docs System MapAgent DocsAgent HarnessConceptsAgent Harness, Runtime, Memory, and EvalsConceptsAgent Operations HubMetaAgent Self-Improvement Eval LibraryConceptsAgent Skill BundlesConceptsAgent SoulSOULAgent Soul HubSOULArchitecture FilingFiling Decision TreeAutomation Audit WorkflowWorkflowsBrain CapsulesMetaBrain Source FabricArchitectureCapture Ingest ProtocolMetaCapture Ingest QuickrefAgent DocsCapture Ingest WorkflowWorkflowsCareer FilingFiling Decision TreeCompiled Truth TimelineFiling Decision TreeCompound KnowledgeSOULConcept System MapConceptsConcepts FilingFiling Decision TreeConvention ChecklistHEARTBEATCross-Agent Harness PortabilityMetaDecisions FilingFiling Decision TreeDesign FilingFiling Decision TreeDormant Schema FilingFiling Decision TreeEntity IdentityFiling Decision TreeEntity Identity ProtocolMetaFactory AutoWikiToolsFiling Decision TreeFiling Decision TreeFiling Resolver HubFiling Decision TreeFull Corpus Workflow and Capability ProgramArchitectureGemini SparkToolsGenerated Surface ContractMetaGenerated Surface RefreshHEARTBEATHermes Agent (Nous Research)ToolsHiggsfield SupercomputerToolsHow to Update the WikiMetaIdentity, Session, and RoutingConceptsInbox FilingFiling Decision TreeInstaticToolsInterview Prep FilingFiling Decision TreeLLM Wiki Agent HarnessArchitectureMemory Routing ProtocolMetaMerge ProtocolFiling Decision TreeMeta FilingFiling Decision TreeMonthly CadenceHEARTBEATNotability GateFiling Decision TreeObsidian Second BrainToolsOpen Knowledge Format (OKF)ToolsOpenCodeToolsOperational HeartbeatHEARTBEATPage Consolidation AuditMetaPeople FilingFiling Decision TreePersonal Agent TemplateToolsPhilosophies FilingFiling Decision TreePostmortems FilingFiling Decision TreeProjects FilingFiling Decision TreeQuality GatesMetaResearch FilingFiling Decision TreeResolver Split DecisionDecisionsResolver vs Skill ResolverFiling Decision TreeRetired Skill Article LedgerMetaSession Startup ProtocolMetaSkill ResolverSkill ResolverSkills FilingFiling Decision TreeSkills, Tools, and Capability DiscoveryConceptsSource Compile WorkflowWorkflowsStories FilingFiling Decision TreeStyle FilingFiling Decision TreeTools FilingFiling Decision TreeWiki Brain Operating ModelMetaWiki Capsules And Agent Doc PacksMetaWiki Corpus Refactor ProgramMetaWiki Graph IntegrationSOULWiki MaintenanceUSERWiki Page FormatMetaWiki Redirect IndexMetaWorkflows FilingFiling Decision TreeX Bookmark Artifact AuditMetaX Bookmarks: AI Agents & Tools (Jan 2025 – Jun 2026)Tools