v3.7 Pass 3d: scrub Claude from published spec; EXTENDS rewritten as a generic predicate

User directive: "extends is generic". The EXTENDS predicate is a generic
capability-mounting mechanism, not Claude-specific. Plus the wider
follow-up to Pass 3c: scrub all remaining Claude/CK.Claude references
from the published v3.7 spec and ontology.

extends.md (full rewrite, 404 -> 291 lines):
- Title now "EXTENDS Predicate" (no "and CK.Claude")
- "Design Problem: Capability Mounting" — frames EXTENDS as the answer
  for any heavy capability (LLM, search, transcoding, simulation), not
  just LLM. Three reasons (not every kernel needs it / different shapes
  / composability) generalised away from Claude.
- "Capability Provider Kernels" replaces the CK.Claude-specific section.
  Defines the provider contract: declare boundary, optionally publish
  template registry, honour dispatch contract.
- "Behavioural Templates (Provider-Defined)" replaces the 5 hardcoded
  Claude personas. Templates are provider-defined; their schema is
  owned by the provider, not by EXTENDS.
- Edge declaration syntax now uses {provider-kernel} placeholder and
  `config.template` (renamed from `config.persona` — persona was a
  Claude-specific term). `cklib/actions.py` _extends metadata renamed
  accordingly.
- Runtime dispatch mermaid uses {source}/{provider} placeholders.
- "Why EXTENDS Instead of Direct Invocation" replaces "Why Not Just
  Call Claude Directly".
- Conformance requirements rephrased — capability-provider contract
  instead of "kernel with LLM capability or claude_agent_sdk".
- Architectural consistency Q&A rephrased generically.

Other docs (Claude scrub):
- index.md: tagline drops "Claude integration"; feature card "Claude
  as Subagent" replaced with "EXTENDS Capability Mounting".
- introduction.md: release-train de-Claudified; layer diagram
  simplified to five layers (no Subagent / Streaming / EXTENDS+Claude
  layers — just CRDs, Auth, EXTENDS, Consensus, Graph).
- taxonomy.md: "Capability: CK.Claude" archetype replaced with
  generic "Capability Provider (project-specific)" archetype; system
  kernel identity table drops CK.Claude row; edge topology shows
  "{capability provider}" placeholder.
- consensus.md: "Talk to the kernel via Claude Code" intro removed;
  consensus loop diagram uses "Authorized executor" instead of
  "Headless Claude Code"; EXTENDS edge to capability provider made
  optional/deployment-specific; review action description rephrased.
- task-engine.md: title description, all body references, lifecycle
  state, validation rationale, post-execution flow, conformance —
  all rephrased around an "authorised executor" instead of "headless
  Claude Code".
- edges.md: EXTENDS example uses {provider-kernel} placeholder and
  `config.template`; cross-reference to extends.md updated.
- nats.md: stream.{kernel} purpose rephrased ("progressive token
  stream from an `agent`-type kernel" instead of "Claude streaming").
- message-envelope.md: "render Claude responses in real time" -> "render
  progressive output in real time".
- graph.md: SPARQL example uses {provider-kernel} placeholder.
- changelog.md: title description, v3.5.8/9/10/11 release notes
  rephrased; v3.5.8 (Subagent) and v3.5.13/14/15/16 references aligned
  with the deferred-chapters note. Frontmatter mention of
  "Claude Code" removed.

Ontology TTLs (Claude scrub):
- core.ttl: ckp:AgentKernel rdfs:comment + dc:description rephrased
  generically (behavioural-template registry instead of "persona
  templates from storage/personas/"; no "CK.Claude is canonical");
  ckp:ModelConfiguration precedence chain rephrased.
- edges.ttl: EXTENDS rdfs:comment broadened beyond "LLM capabilities
  via CK.Claude" to any capability-provider runtime.
- kernel-metadata.ttl: examples list trimmed of CK.Claude.
- processes.ttl: TaskExecutionProcess executor rephrased.
- rbac.ttl: ProcessAgent example uses {capability-provider-kernel}.
- base-instances.ttl: kernel_class example uses generic placeholder.
- self-improvement.ttl: AIAnalyst role + analyzedBy rephrased.
- schema.yaml: AgentKernel description, ModelConfiguration precedence,
  task lifecycle state — all rephrased.
- index.html: abstract + module table descriptions rephrased.

Verified: zero "Claude" matches across docs/v3.7/** and
docs/public/ontology/v3.7/**. Build clean (37 pages reduced to 31 by
Pass 3c is unchanged here; this commit only edits content).

Author: styk-tv

docs/public/ontology/v3.7/base-instances.ttl

@@ -49,7 +49,7 @@ ckp:kernel_class a owl:DatatypeProperty , owl:FunctionalProperty ;
     rdfs:domain ckp:InstanceManifest ;
     rdfs:range xsd:string ;
     rdfs:label "kernel class"@en ;
-    rdfs:comment "The class name of the kernel that produced this instance (e.g., LOCAL.ClaudeCode)"@en .
+    rdfs:comment "The class name of the kernel that produced this instance (e.g., a project-specific bridge or capability-provider kernel)"@en .
 
 ckp:kernel_id a owl:DatatypeProperty , owl:FunctionalProperty ;
     rdfs:domain ckp:InstanceManifest ;

docs/public/ontology/v3.7/core.ttl

@@ -94,8 +94,8 @@ ckp:StaticKernel a owl:Class ;
 ckp:AgentKernel a owl:Class ;
     rdfs:subClassOf ckp:Kernel ;
     rdfs:label "Agent Kernel"@en ;
-    rdfs:comment "A kernel of type 'agent' that supports long-running conversational sessions with LLM inference. Agent kernels MUST support streaming via stream.{kernel}, multi-turn sessions, and persona templates from storage/personas/. CK.Claude is the canonical AgentKernel."@en ;
-    dc:description "v3.6 addition. Type: agent. NATS pattern: persistent subscriber with streaming. Example: CK.Claude." ;
+    rdfs:comment "A kernel of type 'agent' that supports long-running conversational sessions with inference-style runtime. Agent kernels MUST support streaming via stream.{kernel}, multi-turn sessions, and a behavioural-template registry served from their DATA organ (e.g., data/templates/, data/personas/, data/profiles/). Concrete provider kernels are project-specific and not part of the protocol specification."@en ;
+    dc:description "v3.6 addition. Type: agent. NATS pattern: persistent subscriber with streaming." ;
     ckp:implementationStatus "PARTIAL"^^xsd:string .
 
 ### Kernel Properties
@@ -400,7 +400,7 @@ ckp:personaTemperature a owl:DatatypeProperty ;
 ckp:ModelConfiguration a owl:Class ;
     rdfs:subClassOf bfo:0000019 ;  # bfo:Quality
     rdfs:label "Model Configuration"@en ;
-    rdfs:comment "A quality describing LLM model selection and effort level for a kernel or action. v3.7 defines a three-level precedence chain: CK.Claude default < persona template < EXTENDS edge config."@en ;
+    rdfs:comment "A quality describing model selection and effort level for an agent-type kernel or action backed by one. The precedence chain is provider-defined and typically: provider default < behavioural template < EXTENDS edge config."@en ;
     dc:description "v3.7 addition. Properties: defaultModel, modelEffort. Values: haiku, sonnet, opus; low, medium, high, max." .
 
 #################################################################

docs/public/ontology/v3.7/edges.ttl

@@ -118,7 +118,7 @@ ckpe:PRODUCES a owl:NamedIndividual , ckpe:EdgePredicate ;
 
 ckpe:EXTENDS a owl:NamedIndividual , ckpe:EdgePredicate ;
     rdfs:label "EXTENDS"@en ;
-    rdfs:comment "Capability mounting. Source gains NEW actions backed by target's runtime. Actions defined in edge config, not target SKILL.md. Instances sealed in source's DATA loop. Primary mechanism for mounting LLM capabilities via CK.Claude."@en ;
+    rdfs:comment "Capability mounting. Source gains NEW actions backed by target's runtime. Actions defined in edge config, not target SKILL.md. Instances sealed in source's DATA loop. Primary mechanism for mounting any kind of capability provider — LLM inference, search, transcoding, simulation — without coupling the source kernel's identity to a specific runtime."@en ;
     ckpe:compositionStyle "capability mounting"^^xsd:string ;
     ckpe:instanceOwnership "source kernel writes all"^^xsd:string ;
     ckpe:activationModel "on-demand via persona"^^xsd:string ;

docs/public/ontology/v3.7/index.html

@@ -59,7 +59,7 @@ <h1>CKP Ontology v3.7 <span class="status status-release">v3.7</span></h1>
 <div class="prefix-box">@prefix ckp: &lt;https://conceptkernel.org/ontology/v3.7/&gt; .</div>
 
 <h2>Abstract</h2>
-<p class="abstract">v3.7 extends the CKP ontology with the agent kernel type (CK.Claude), five formal edge predicates (COMPOSES, TRIGGERS, PRODUCES, EXTENDS, LOOPS_WITH), 20 compliance check types, a four-layer consensus governance engine, task execution via headless Claude Code, dynamic kernel spawning, multi-user NATS sessions, and persona-based LLM capability mounting. All classes maintain BFO 2020 grounding through the four-layer import chain.</p>
+<p class="abstract">v3.7 extends the CKP ontology with the `agent` kernel type (long-running, conversational), five formal edge predicates (COMPOSES, TRIGGERS, PRODUCES, EXTENDS, LOOPS_WITH), 20 compliance check types, a four-layer consensus governance engine, a task execution dispatch contract, multi-user NATS sessions, and capability mounting via the EXTENDS edge predicate. Concrete capability-provider kernels are project-specific and out of scope for the protocol. All classes maintain BFO 2020 grounding through the four-layer import chain.</p>
 
 <h2>Ontology Modules</h2>
 <table>
@@ -87,7 +87,7 @@ <h2>v3.7 Additions</h2>
   <tr><td><code>ckp:LOOPS_WITH</code></td><td>core.ttl</td><td>ckp:RelationshipType</td><td>Bidirectional cooperation edge predicate</td></tr>
   <tr><td><code>edges.ttl</code></td><td>NEW MODULE</td><td>bfo:Quality</td><td>5 edge predicates with composition semantics</td></tr>
   <tr><td><code>consensus.ttl</code></td><td>NEW MODULE</td><td>prov:Activity</td><td>Proposals, votes, 4 validation layers, decisions</td></tr>
-  <tr><td><code>ckpp:TaskExecutionProcess</code></td><td>processes.ttl</td><td>bfo:Process</td><td>Headless Claude Code task execution</td></tr>
+  <tr><td><code>ckpp:TaskExecutionProcess</code></td><td>processes.ttl</td><td>bfo:Process</td><td>Consensus-driven task execution by an authorised executor</td></tr>
   <tr><td><code>ckpp:SpawningProcess</code></td><td>processes.ttl</td><td>bfo:Process</td><td>Dynamic kernel creation from discovered concepts</td></tr>
   <tr><td><code>ckpp:SessionProcess</code></td><td>processes.ttl</td><td>bfo:Process</td><td>Multi-user NATS collaborative sessions</td></tr>
   <tr><td><code>ckpp:ConsensusEvaluationProcess</code></td><td>processes.ttl</td><td>bfo:Process</td><td>Four-layer consensus evaluation</td></tr>

docs/public/ontology/v3.7/kernel-metadata.ttl

@@ -325,7 +325,7 @@ ckp:SystemKernel a owl:Class ;
     rdfs:subClassOf ckp:Kernel ;
     rdfs:label "System Kernel"@en ;
     rdfs:comment "Core system kernel (CK.*) providing infrastructure services"@en ;
-    dc:description "Examples: CK.Operator, CK.ComplianceCheck, CK.Claude, CK.Consensus" .
+    dc:description "Examples: CK.Operator, CK.ComplianceCheck, CK.Consensus, and project-specific capability-provider kernels reached via the EXTENDS predicate." .
 
 ckp:ApplicationKernel a owl:Class ;
     rdfs:subClassOf ckp:Kernel ;

docs/public/ontology/v3.7/processes.ttl

@@ -518,8 +518,8 @@ ckpp:precedes a owl:ObjectProperty ;
 ckpp:TaskExecutionProcess a owl:Class ;
     rdfs:subClassOf ckpp:Process ;
     rdfs:label "Task Execution Process"@en ;
-    rdfs:comment "A process for executing consensus-approved tasks via headless Claude Code. Tasks are generated by CK.Consensus and executed by the task execution engine. Lifecycle: pickup -> execution -> validation -> commit."@en ;
-    dc:description "v3.7 addition. Executor: headless Claude Code (claude -p or claude_agent_sdk)." ;
+    rdfs:comment "A process for executing consensus-approved tasks via an authorised executor. Tasks are generated by CK.Consensus and dispatched by the task execution engine to the executor declared in the task. Lifecycle: pickup -> execution -> validation -> commit."@en ;
+    dc:description "v3.7 addition. Executor: deployment-specific (typically a capability-provider kernel reached via EXTENDS)." ;
     ckp:implementationStatus "PARTIAL"^^xsd:string .
 
 ### TaskExecution Participants

docs/public/ontology/v3.7/rbac.ttl

@@ -60,7 +60,7 @@ ckp:ProcessAgent a owl:Class ;
     rdfs:comment "A kernel executing as an autonomous agent capable of performing actions."@en ;
     dc:description "URN: ckp://Agent/process:{KernelName}" ;
     dc:description "Example: ckp://Agent/process:System.Consensus (governance agent)" ;
-    dc:description "Example: ckp://Agent/process:ConceptKernel.LLM.Claude (autonomous LLM)" .
+    dc:description "Example: ckp://Agent/process:{capability-provider-kernel} (an autonomous capability provider reached via EXTENDS)" .
 
 ### Agent Properties
 

docs/public/ontology/v3.7/schema.yaml

@@ -54,7 +54,8 @@ classes:
     description: >
       LLM-capable kernel supporting long-running conversational sessions,
       streaming via stream.{kernel}, multi-turn sessions, and persona
-      templates from storage/personas/. CK.Claude is the canonical AgentKernel.
+      behavioural templates from data/templates/ (or an equivalent
+      provider-defined path). Concrete provider kernels are project-specific.
     attributes:
       default_model: { description: "Default LLM model (sonnet, opus, haiku)" }
       default_effort: { description: "Default effort level (low, medium, high, max)" }
@@ -100,7 +101,7 @@ classes:
   ModelConfiguration:
     description: >
       LLM model selection quality. Three-level precedence:
-      CK.Claude default < persona template < EXTENDS edge config.
+      provider default < behavioural template < EXTENDS edge config.
     attributes:
       model: { description: "LLM model: haiku, sonnet, opus, or full model name" }
       effort: { description: "Effort level: low, medium, high, max" }
@@ -419,7 +420,7 @@ enums:
     description: "Lifecycle status of a consensus-generated task"
     permissible_values:
       pending: { description: "Generated by consensus, awaiting execution" }
-      executing: { description: "Headless Claude Code is working on it" }
+      executing: { description: "Authorised executor is working on the task" }
       completed: { description: "Output validated, changes committed" }
       failed: { description: "Validation failed or execution error" }
 

docs/public/ontology/v3.7/self-improvement.ttl

@@ -40,7 +40,7 @@ ckpi:SelfInspectorRole
 ckpi:AIAnalystRole
   rdf:type bfo:0000023 ;  # BFO:Role
   rdfs:label "AI Analyst" ;
-  rdfs:comment "Role borne by LLM kernels (Fabric, Claude) that analyze validation issues and propose solutions" ;
+  rdfs:comment "Role borne by capability-provider kernels (typically `agent`-type, project-specific) that analyse validation issues and propose solutions" ;
   ckp:roleContext "ai-analysis" ;
   ckp:grants ckpi:permission-analyze-issues ;
   ckp:grants ckpi:permission-propose-solutions .
@@ -256,7 +256,7 @@ ckpi:analyzedBy
   rdfs:domain ckpi:ValidationIssue ;
   rdfs:range ckp:Kernel ;
   rdfs:label "analyzed by" ;
-  rdfs:comment "Links issue to AI analyst kernel (Fabric, Claude)" .
+  rdfs:comment "Links issue to the analyst kernel (a project-specific capability-provider kernel)" .
 
 ckpi:submittedToConsensus
   rdf:type owl:ObjectProperty ;

docs/v3.7/changelog.md

@@ -1,6 +1,6 @@
 ---
 title: Changelog
-description: Full changelog for CKP v3.7 -- the complete 10-part, 48-chapter specification covering identity, ontology, runtime, infrastructure, Claude integration, and governance.
+description: Full changelog for CKP v3.7 -- the nine-part specification covering identity, ontology, runtime, infrastructure, edges & composition, system kernels, and governance.
 ---
 
 # Changelog
@@ -38,7 +38,7 @@ v3.5-alpha6 was the last deployed incremental release. v3.7 adds:
 | IX | Governance & Accumulation: CK.Consensus, task engine, ontological graph, sessions, PROV-O provenance |
 
 ::: info Deferred (not part of the v3.7 normative specification)
-Six chapters drafted during v3.5.x development -- *CK as Subagent*, *Streaming*, *Web Shell*, *CK Loop Evolution*, *Agent Teams*, *Dynamic Spawning* -- have been moved out of the normative v3.7 spec. They exposed vendor-specific integrations (Claude Code) and aspirational features that belong in tooling specs rather than the protocol specification. The consolidated content is preserved for future revisiting at `parked/3.7-deferred.md` in the source tree; it is intentionally not served on the website.
+Six chapters drafted during v3.5.x development -- *CK as Subagent*, *Streaming*, *Web Shell*, *CK Loop Evolution*, *Agent Teams*, *Dynamic Spawning* -- have been moved out of the normative v3.7 spec. They were authored around vendor-specific integrations and aspirational features that belong in tooling specs rather than the protocol specification. The consolidated content is preserved for future revisiting at `parked/3.7-deferred.md` in the source tree; it is intentionally not served on the website.
 :::
 
 ---
@@ -135,47 +135,39 @@ All notable changes in the development increments that compose v3.7. Each versio
 
 - CK.Consensus kernel: ontological governance engine
 - Five actions: propose, evaluate, approve, decisions, review
-- Review action EXTENDS CK.Claude with `strict-auditor` persona
+- Review action EXTENDS a capability provider (deployment-specific) with a strict-auditor-style template
 - Proposal evaluation against ontology + SHACL + fleet topology
-- Task generation for headless Claude execution
+- Task generation for downstream execution by an authorised executor
 - Every decision is a `prov:Activity` with full audit chain
 - `spec-parts/consensus-loop.md`: normative definition
 
-### v3.5.10 -- EXTENDS Predicate + CK.Claude <Badge type="tip" text="DEPLOYED" />
+### v3.5.10 -- EXTENDS Predicate <Badge type="tip" text="DEPLOYED" />
 
 **Date:** 2026-04-06
 
 - EXTENDS predicate implementation in `cklib/actions.py`
 - EXTENDS creates NEW actions on source kernel from edge config (not inherited from target)
-- `get_effective_actions()` enriches EXTENDS actions with persona + constraints metadata
-- CK.Claude kernel: agent-type with persona templates (analytical-reviewer, friendly-assistant, strict-auditor)
-- CK.Claude `processor.py`: handles message/analyze/summarize via `claude -p` with persona loading
+- `get_effective_actions()` enriches EXTENDS actions with the edge's `template` and `constraints` metadata
+- First capability-provider kernel deployed (vendor-specific; not part of the protocol — see [parked deferred chapters](https://github.com/ConceptKernel/conceptkernel.github.io/blob/main/parked/3.7-deferred.md))
 - `resolve_composed_actions()`: COMPOSES inherits target actions; EXTENDS creates new from edge config
 - `spec-parts/extends-predicate.md`: full normative definition
 
-### v3.5.9 -- Claude Streaming via NATS <Badge type="tip" text="DEPLOYED" />
+### v3.5.9 -- Stream Topic Added <Badge type="tip" text="DEPLOYED" />
 
 **Date:** 2026-04-06
 
-- `stream.{kernel}` topic added to NatsKernelLoop (`cklib/nats_loop.py`)
+- `stream.{kernel}` topic added to NatsKernelLoop (`cklib/nats_loop.py`) for progressive output from `agent`-type kernels
 - `stream_event()` callback passed to handlers for per-token NATS publishing
 - Handler signature extended: `handler_fn(body, nc=, trace_id=, stream=)` -- backwards compatible
 - Structured JSON logging replaces `[rx]`/`[tx]` print statements
 - `_log()` method for spec-compliant JSON output (ts, level, kernel, event)
 - `spec-parts/structured-logging.md`: stream topic definition + conformance
 
-### v3.5.8 -- CK as Claude Code Subagent <Badge type="tip" text="DEPLOYED" />
+### v3.5.8 -- Subagent Tooling <Badge type="info" text="DEFERRED" />
 
 **Date:** 2026-04-06
 
-- Updated ck-agent skill (`~/.claude/skills/ck-agent/SKILL.md`) for CKP v3.5 layout
-- CLAUDE.md at root (not llm/), SKILL.md loading, ontology.yaml summary
-- Three-loop discipline enforced: CK read-only, TOOL read-only, DATA writable
-- Memory persistence: `data/memory/MEMORY.md` (DATA loop)
-- Multi-root search: `$CK_CONCEPTS_DIR`, `./concepts/`, `~/git/delve_workspace/concepts/`
-- Fuzzy kernel resolution: `CK.*`, `Delvinator.*`, `CS.*`, `Hello.*`
-- NATS bridge: optional dispatch to live kernel via `nats pub`
-- `data/memory/` directories created for all kernels
+This increment landed a vendor-specific subagent integration (talking to a kernel through an external IDE-style assistant). The chapter is preserved at [parked/3.7-deferred.md](https://github.com/ConceptKernel/conceptkernel.github.io/blob/main/parked/3.7-deferred.md) for future revisiting; it is not part of the v3.7 normative specification.
 
 ### v3.5.7 -- Hello.Greeter Kernel <Badge type="tip" text="DEPLOYED" />
 
@@ -234,7 +226,7 @@ All notable changes in the development increments that compose v3.7. Each versio
 
 **Date:** 2026-04-05
 
-- `SPEC.CKP.v3.5.4.delta.md`: CK-as-subagent (D1-D4), Claude streaming (D5), LOCAL.ClaudeCode bridge (D6), Consensus loop (D7), EXTENDS predicate (D8)
+- `SPEC.CKP.v3.5.4.delta.md`: subagent (D1-D4, deferred), stream topic (D5), local-bridge (D6, deferred), Consensus loop (D7), EXTENDS predicate (D8)
 - Consensus services integration: `PLAN.v002.md`, updated CLAUDE.md + README.md in `ref-consensus-services/`
 
 ### v3.5.3 -- Auth/Web Shell Delta Spec <Badge type="info" text="SPEC" />