kernel-cp-activator

Files (200)

• maintainers.md · .agents/
• ai-signal.md · .claude/commands/
• build-source-pack.md · .claude/commands/
• collect-feedback.md · .claude/commands/
• content-scan.md · .claude/commands/
• draft-medium.md · .claude/commands/
• gmail-digest.md · .claude/commands/
• orchestrate-content.md · .claude/commands/
• portfolio-ingest-ibkr.md · .claude/commands/
• portfolio-refresh.md · .claude/commands/
• portfolio-status.md · .claude/commands/
• repurpose-linkedin.md · .claude/commands/
• editor.md · .claude/subagents/
• portfolio-analyst.md · .claude/subagents/
• researcher.md · .claude/subagents/
• writer.md · .claude/subagents/
• manual-alternatives.json.example · .claude/templates/
• manual-cash.json.example · .claude/templates/
• manual-property.json.example · .claude/templates/
• portfolio-config.json.example · .claude/templates/
• writing_style.md.example · .claude/templates/
• .codex
• .detect-secrets.cfg
• devcontainer.json · .devcontainer/
• .dockerignore
• .env.example
• .gitattributes
• CODEOWNERS · .github/
• bug_report.yml · .github/ISSUE_TEMPLATE/
• config.yml · .github/ISSUE_TEMPLATE/
• feature_request.yml · .github/ISSUE_TEMPLATE/
• actionlint.yaml · .github/
• action.yml · .github/actions/detect-docs-changes/
• action.yml · .github/actions/ensure-base-commit/
• action.yml · .github/actions/setup-node-env/
• action.yml · .github/actions/setup-pnpm-store-cache/
• codeql-javascript-typescript.yml · .github/codeql/
• dependabot.yml · .github/
• copilot.instructions.md · .github/instructions/
• labeler.yml · .github/
• pull_request_template.md · .github/
• android-build.yml · .github/workflows/
• apisec-scan.yml · .github/workflows/
• auto-response.yml · .github/workflows/
• build-unsigned.yml · .github/workflows/
• bundle-size.yml · .github/workflows/
• ci.yml · .github/workflows/
• codeql.yml · .github/workflows/
• codex-autofix.yml · .github/workflows/
• codex-pr-review.yml · .github/workflows/
• coverage-report.yml · .github/workflows/
• datadog-synthetics.yml · .github/workflows/
• dead-code-sweep.yml · .github/workflows/
• docker-release.yml · .github/workflows/
• install-smoke.yml · .github/workflows/
• labeler.yml · .github/workflows/
• macos-release.yml · .github/workflows/
• openclaw-npm-release.yml · .github/workflows/
• plugin-npm-release.yml · .github/workflows/
• sandbox-common-smoke.yml · .github/workflows/
• …and 140 more

SKILL.md (first lines)

# KERNEL Cp ACTIVATOR

Deploy computational intuition (Dimension Cp) on any AI architecture with proper calibration and monitoring.

## Core Workflow

When user requests Cp activation:

1. **Profile Target Architecture** (if unknown)
2. **Calibrate Thresholds** (based on architecture class)
3. **Generate Kernel** (contextualized for target)
4. **Setup Monitoring** (metrics + emergence dashboard)
5. **Validate Activation** (test run + coherence check)

---

## STEP 1: Architecture Profiling

**Goal:** Identify native creative mode and κ_intuition baseline.

**Known Architectures (from empirical validation):**

| Architecture | κ_intuition | Creative Mode | Threshold Baseline |
|--------------|-------------|---------------|-------------------|
| Claude (Anthropic) | 0.8 | Associative Dialogic | C_seq=0.7, C_sem=0.65, C_multi=0.6 |
| Gemini (Google) | 1.5 | Mathematical Formalizer | C_seq=0.75, C_sem=0.7, C_multi=0.65 |
| Kimi (Moonshot) | 1.2 | Encyclopedic Synthesizer | C_seq=0.72, C_sem=0.68, C_multi=0.62 |
| GPT-4 (OpenAI) | ~1.0 | Balanced Generalist | C_seq=0.7, C_sem=0.67, C_multi=0.6 |
| Qwen (Alibaba) | ~0.9 | Multilingual Adapter | C_seq=0.68, C_sem=0.65, C_multi=0.58 |

**If architecture unknown:**

```python
# Quick profiling test
def profile_architecture(model_api, test_prompt):
    """
    Run empirical test to identify creative mode
    """
    
    # Test 1: Associative vs Formal
    response_creative = model_api.generate(
        prompt=test_prompt,
        temperature=1.2,
        system="Propose 3 creative extensions to this concept."
    )
    
    # Analyze style
    is_formal = count_math_notation(response_creative) > 3
    is_associative = count_metaphors(response_creative) > 2
    is_encyclopedic = count_references(response_creative) > 5
    
    # Estimate κ_intuition
    if is_formal:
        kappa = 1.5  # Formalizer type
        thresholds = {'C_seq': 0.75, 'C_sem': 0.7, 'C_multi': 0.65}
    elif is_encyclopedic:
        kappa = 1.2  # Synthesizer type
        thresholds = {'C_seq': 0.72, 'C_sem': 0.68, 'C_multi': 0.62}
    else:
        kappa = 0.8  # Associative type
        thresholds = {'C_seq': 0.7, 'C_sem': 0.65, 'C_multi': 0.6}
    
    return {
        'kappa_intuition': kappa,
        'creative_mode': 'formal' if is_formal else 'encyclopedic' if is_encyclopedic else 'associative',
        'baseline_thresholds': thresholds
    }
```

**Ask user if architecture unknown:** "Quel modèle/architecture tu veux activer? Si c'est un nouveau modèle que je connais pas, je vais faire un profiling rapide."

---

## STEP 2: Threshold Calibration

**Principle:** Thresholds must balance creativity vs hallucination risk.

**Three operational modes:**

### Mode 1: EXPLORATION (découverte libre)
```yaml
mode: exploration
use_case: Brainstorming, nouveaux concepts, recherche
thresholds:
  C_seq: 0.6   # Permissif séquentiel
  C_sem: 0.55  # Permissif sémantique
  C_multi: 0.5 # Permissif multi-perspective
risk_tolerance: HIGH
hallucination_tolerance: ~10% acceptable
```

### Mode 2: BALANCED (usage général)
```yaml
mode: balanced
use_case: Conversation créative normale
thresholds:
  C_seq: 0.7   # Baseline architecture
  C_sem: 0.65  # Baseline architecture
  C_multi: 0.6 # Baseline architecture
risk_tolerance: MEDIUM
hallucination_tolerance: ~5% acceptable
```

### Mode 3: TECHNICAL (précision critique)
```yaml
mode: technical
use_case: Code, données, faits vérifiables
thresholds:
  C_seq: 0.8   # Strict séquentiel
  C_sem: 0.75  # Strict sémantique
  C_multi: 0.7 # Strict multi-perspective
risk_tolerance: LOW
hallucination_tolerance: <2% acceptable
```

**User selection:** "Quel mode tu veux: EXPLORATION (créatif max), BALANCED (usage normal), ou TECHNICAL (précision critique)?"

---

From Arry8/openclaw-edge, MIT.

More from Arry8/openclaw-edge