sinapse-ai 9.3.0 → 9.4.0

package/squads/squad-storytelling/knowledge-base/video-storytelling.md

@@ -0,0 +1,252 @@
+# Video Storytelling
+
+> Version: 1.0.0 | Domain: video, short-form, YouTube, scripting
+
+## Overview
+
+Video is the highest-engagement narrative medium available to brands and creators. It combines visual, auditory, and temporal elements into a single experience — activating more of the brain than any static medium.
+
+This file covers the storytelling principles specific to video: YouTube long-form strategy, short-form narrative structure, scripting patterns, and the psychology of video attention.
+
+**Research source:** MS-008 Content, Section 7.
+
+---
+
+## The Attention Architecture of Video
+
+### The First 3 Seconds Rule
+
+65% of viewers decide in the first 3 seconds whether to continue watching. This is not a creative challenge — it's a biological one. The brain's threat detection system is constantly filtering: "Is this worth my time and attention?"
+
+The hook must answer yes immediately.
+
+**Hook types by mechanism:**
+
+| Hook Type | Mechanism | Example |
+|-----------|-----------|---------|
+| **Controversial statement** | Pattern interrupt — breaks expectation | "Content marketing is dead. Here's what works instead." |
+| **Question the audience already has** | Curiosity gap — they need the answer | "Why do 90% of blogs never generate a single lead?" |
+| **Statistic with high surprise factor** | Dissonance — challenges a belief | "This technique tripled our traffic in 30 days" |
+| **Before/after transformation** | Aspiration — show the destination first | "Six months ago I had 200 newsletter subscribers. Today, 50,000." |
+| **Tutorial promise** | Value signal — explicit payoff | "In 60 seconds you'll know how to write headlines that convert" |
+| **Pattern interrupt** | Novelty detection — unexpected start | Unusual visual, unexpected sound, fast cut from silence |
+
+### The Attention Curve
+
+Video attention follows a predictable decay curve:
+- Peak attention: Seconds 0-30
+- Sustained attention: Maintained by value delivery, pattern interruption, and forward promises
+- Drop-off zones: Transitions, slow explanations, unstructured segments
+- Re-engagement: New hooks, reveals, and payoff moments can re-spike attention
+
+**Storytelling implication:** Great video isn't a single narrative arc — it's a series of mini-arcs, each with its own tension and resolution, chained together to sustain engagement across the full run time.
+
+---
+
+## YouTube Long-Form Narrative Structure
+
+### The PAS-Driven Video Structure
+
+For educational/authority content (5-20 minutes):
+
+```
+HOOK (0:00-0:30)
+  → Most compelling moment, question, or promise
+  → "By the end of this video, you will know [specific thing]"
+
+CONTEXT ESTABLISHING (0:30-2:00)
+  → Why this topic matters right now
+  → Quick credibility signal
+
+PROBLEM DEEP-DIVE (2:00-5:00)
+  → The specific pain points in detail
+  → Make the audience say "yes, that's me"
+
+SOLUTION DELIVERY (5:00-15:00)
+  → The core content — structured in steps or reveals
+  → Mini-hooks between each section ("And in section 3, I'll show you the part most people miss")
+
+SYNTHESIS (15:00-18:00)
+  → Bring it together — how the pieces connect
+
+CALL TO ACTION (18:00-20:00)
+  → One specific action
+  → Community invitation / subscribe prompt
+```
+
+### The "Open Loop" Technique
+
+Open loops are unresolved tensions planted early that are resolved later. The brain compels viewers to stay until the loop closes.
+
+**How to use:**
+- "I'll show you the most important thing at the end of this video"
+- "There's one mistake I see everyone make — I'll show you exactly what it is in section 4"
+- "Most people stop here, but if you keep watching, I'll show you the part they miss"
+
+### Chapter Structure for Retention
+
+YouTube chapters (via timestamps in description) serve two purposes:
+1. SEO (creates "key moments" in Google search results)
+2. Navigation (reduces abandonment by letting viewers jump to relevant sections)
+
+**Chapter naming:** Make chapter names descriptive and interesting, not just labels. "The Mistake That Kills Conversion" is better than "Section 3."
+
+---
+
+## Short-Form Video Narrative (Reels, TikTok, Shorts)
+
+### The 60-Second Story
+
+Short-form video (< 60 seconds) demands extreme compression. Every second must carry narrative weight.
+
+**The 60-second structure:**
+
+```
+[0-3 sec]  HOOK — The reason to watch
+[3-15 sec] TENSION — The problem or the contrast
+[15-45 sec] DELIVERY — The insight, the transformation, the how
+[45-55 sec] PAYOFF — The specific outcome or revelation
+[55-60 sec] CTA — Follow, share, comment, link in bio
+```
+
+### Hook Compression for Short-Form
+
+Short-form hooks must be instant. The viewer's swipe threshold is essentially zero — if the opening frame doesn't arrest attention, they're gone.
+
+**Effective short-form openers:**
+- Start mid-sentence or mid-action (creates disorientation → curiosity)
+- Text overlay with the premise (works with sound off — 85% of viewers)
+- Verbal hook + text overlay simultaneously
+- "If you [specific situation], watch this"
+- Contrarian opener: "Everyone says [X]. They're wrong."
+
+### Platform-Specific Short-Form Storytelling
+
+| Platform | Tone | Pacing | Best Story Types |
+|----------|------|--------|-----------------|
+| **TikTok** | Raw, casual, informal | Fast (2-3 sec cuts) | Personal transformation, behind-the-scenes, contrarian takes |
+| **Instagram Reels** | More polished, aspirational | Moderate | Brand storytelling, lifestyle narrative, tutorial |
+| **YouTube Shorts** | Educational, authority | Varies | Compressed tutorials, data insights, quick tips |
+
+---
+
+## Video Scripting Frameworks
+
+### The 3-Act Script
+
+For any video longer than 3 minutes:
+
+**Act 1: Setup (20% of runtime)**
+- Establish the situation
+- Introduce the character (the viewer's proxy — could be yourself, a customer, or a concept)
+- Plant the central tension
+
+**Act 2: Conflict (60% of runtime)**
+- The journey through the problem
+- Escalating complexity
+- The search for solution
+- The key insight (the Midpoint — must come exactly at the middle)
+
+**Act 3: Resolution (20% of runtime)**
+- The synthesis
+- The transformation
+- The forward-looking vision
+- The clear next step
+
+### The Interview/Talking Head Script
+
+For direct-to-camera narrative:
+
+```
+SETUP LINE: One sentence that establishes the topic and stakes
+PERSONAL HOOK: A specific moment from your experience that grounds the topic
+THE PROBLEM: Describe the situation the viewer is in (use "you" language)
+THE INSIGHT: The core thing they need to know
+THE EVIDENCE: Proof (data, case study, personal example)
+THE APPLICATION: How to use this
+THE FORWARD HOOK: The next thing they'll want to know
+THE CTA: The specific action
+```
+
+### The Mini-Documentary Script
+
+For brand stories, case studies, and origin stories:
+
+```
+COLD OPEN: Start in the middle of the story — the most dramatic moment
+REWIND: "Here's how we got here" — the backstory
+THE CHALLENGE: What made this hard
+THE TURNING POINT: The moment everything changed
+THE RESULT: What happened
+THE MEANING: What this means for the viewer / the brand
+```
+
+---
+
+## Visual Storytelling Principles for Video
+
+### Show, Don't Tell
+The most basic principle, most violated in business video. If you're explaining a concept, show an example. If you're describing an outcome, show a before/after. If you're making a claim, show evidence.
+
+**Test:** If you turned off the audio, could the viewer still understand the story from the visuals alone? Great video storytelling passes this test.
+
+### The B-Roll Narrative
+B-roll (supplementary footage) carries the visual story while the voiceover or talking head carries the verbal story. Great B-roll:
+- Shows the environment being discussed (not just generic stock)
+- Shows the character doing the thing being described
+- Provides visual metaphors for abstract concepts
+- Creates emotional context (human faces, hands, environments)
+
+### Text Overlay as Storytelling Element
+With 85% of social video watched without sound, text overlay is not a caption — it's a storytelling layer. It should:
+- Highlight the most important words (not transcribe everything)
+- Create rhythm and emphasis
+- Stand alone as a readable story
+
+---
+
+## The YouTube Hook-Content-CTA Formula
+
+**Hook (3 seconds):** The reason to stay
+**Content (bulk):** The fulfillment of the hook's promise
+**CTA (last 30 seconds):** The next step — always one, never many
+
+**The CTA hierarchy:**
+1. Subscribe (channel growth)
+2. Comment (engagement signal — algorithms favor it)
+3. Share (distribution)
+4. Link in bio / description (conversion)
+
+**The rule:** Ask for one action. Multiple CTAs compete and reduce total compliance.
+
+---
+
+## Brand Storytelling in Video
+
+### The 3-Video Brand Narrative Stack
+
+For brands building authority through video:
+
+| Video Type | Purpose | Length | Frequency |
+|------------|---------|--------|-----------|
+| **Brand story video** | Who we are, why we exist | 2-5 min | Once (hero piece) |
+| **Customer story videos** | Social proof through narrative | 1-3 min | Monthly |
+| **Educational authority videos** | Prove expertise, build trust | 5-15 min | Weekly |
+
+### Emotional Arc by Video Goal
+
+| Goal | Emotional Journey | Narrative Structure |
+|------|-----------------|-------------------|
+| Build brand | Curiosity → Recognition → Trust → Affinity | Brand story arc |
+| Drive conversion | Pain → Hope → Belief → Action | PAS arc |
+| Build community | Isolation → Belonging → Commitment | Us story arc |
+| Educate | Confusion → Clarity → Capability → Growth | Tutorial arc |
+
+---
+
+## Cross-Reference
+
+- **Kindra Hall's 4 Stories** provides the story types that power the strongest video content (founder, value, customer, purpose stories)
+- **Blake Snyder's Opening Image** principle applies to the visual hook — the first frame establishes the before
+- **Park Howell's ABT** can diagnose the narrative logic of any video script
+- **Nancy Duarte's DataStory** applies when video content is data-driven or analytical

package/squads/claude-code-mastery/data/swarm-orchestration-patterns.yaml

@@ -1,378 +0,0 @@
-# Swarm Orchestration Patterns for AI Agents
-# Squad: claude-code-mastery
-# Source: Research 07-skills-agents-swarm (llm-files-swarm-agi.md)
-# Last updated: 2026-04-10
-
-version: "1.0.0"
-source_research: "07-skills-agents-swarm/llm-files-swarm-agi.md"
-
-# ---------------------------------------------------------------------------
-# SWARM INTELLIGENCE FUNDAMENTALS
-# ---------------------------------------------------------------------------
-swarm_intelligence:
-  definition: >
-    Collective behavior emerging from decentralized, self-organized agents
-    following simple local rules — producing globally intelligent behavior
-    without central coordination.
-  origin: "Dorigo & Colorni (1992) — Ant Colony Optimization"
-  key_principle: "Local simplicity → Global intelligence"
-
-  classic_algorithms:
-    - name: "Ant Colony Optimization (ACO)"
-      mechanism: "Pheromone trails"
-      best_for: "Route optimization, graph search"
-      ai_analogy: "Shared memory blackboard for successful solutions"
-
-    - name: "Particle Swarm Optimization (PSO)"
-      mechanism: "Velocity + personal/global best"
-      best_for: "Continuous parameter optimization"
-      ai_analogy: "Agents sharing best-found strategies"
-
-    - name: "Bee Algorithm"
-      mechanism: "Scout + forager specialization"
-      best_for: "Parallel exploration + exploitation"
-      ai_analogy: "Parallel agent spawning for task exploration"
-
-    - name: "Flocking (Boids)"
-      mechanism: "Separation + alignment + cohesion"
-      best_for: "Emergent coordination without hierarchy"
-      ai_analogy: "Peer agent coordination without orchestrator"
-
-# ---------------------------------------------------------------------------
-# AI AGENT SWARM TOPOLOGIES (4 Core Patterns)
-# ---------------------------------------------------------------------------
-swarm_topologies:
-
-  - name: "Hierarchical Swarm"
-    also_known_as: "Orchestrator-Worker"
-    pattern: |
-      Orchestrator
-          ├── Worker Agent 1
-          ├── Worker Agent 2
-          └── Worker Agent 3
-    characteristics:
-      - "Single point of coordination at top"
-      - "Clear delegation chain"
-      - "Predictable task routing"
-    strengths:
-      - "Easy to reason about"
-      - "Clear authority and accountability"
-      - "Good for complex domain decomposition"
-    weaknesses:
-      - "Single point of failure at orchestrator"
-      - "Bottleneck on orchestrator throughput"
-    best_for: "Complex multi-domain tasks with clear specializations"
-    sinapse_example: "sinapse-orqx → squad-orqx → specialist agents"
-    frameworks: ["CrewAI", "AutoGen", "LangGraph", "SINAPSE"]
-
-  - name: "Democratic / Flat Swarm"
-    also_known_as: "Peer-to-Peer"
-    pattern: |
-      Agent A ←→ Agent B
-          ↕           ↕
-      Agent C ←→ Agent D
-    characteristics:
-      - "No hierarchy — all agents equal"
-      - "Agents vote or negotiate on decisions"
-      - "Consensus-based routing"
-    strengths:
-      - "Resilient to single agent failure"
-      - "Emergent specialization"
-    weaknesses:
-      - "Harder to coordinate for complex tasks"
-      - "Potential deadlock on disagreements"
-    best_for: "Creative tasks, brainstorming, peer review"
-    frameworks: ["AutoGen multi-agent", "Society of Mind patterns"]
-
-  - name: "Market-Based / Auction Swarm"
-    also_known_as: "Contract Net Protocol"
-    pattern: |
-      Task Broadcaster → [Agent bids] → Best fit wins → Executes
-    characteristics:
-      - "Agents bid on tasks based on capability"
-      - "Dynamic load balancing"
-      - "Self-organizing specialization"
-    strengths:
-      - "Optimal resource allocation"
-      - "Agents naturally specialize"
-    weaknesses:
-      - "Complex bid evaluation"
-      - "Overhead from negotiation"
-    best_for: "Dynamic workloads where agent capability varies"
-    frameworks: ["OpenHands task routing", "Agent SDK multi-agent"]
-
-  - name: "Stigmergic Swarm"
-    also_known_as: "Shared Memory / Blackboard"
-    pattern: |
-      Agent A → [Shared Context] ← Agent B
-                      ↑
-                 Agent C reads/writes
-    characteristics:
-      - "Agents coordinate via shared artifacts"
-      - "No direct agent-to-agent communication needed"
-      - "State visible to all agents"
-    strengths:
-      - "Decoupled — agents don't need to know each other"
-      - "Naturally persistent context"
-    weaknesses:
-      - "Shared state can become stale"
-      - "Write conflicts without locking"
-    best_for: "Long-running workflows where context must persist across agent switches"
-    sinapse_example: "Story files in docs/stories/ + scratchpad in .sinapse/scratchpad/"
-    frameworks: ["SINAPSE story system", "AIOX .aiox/ directory", "Blackboard architecture"]
-
-  - name: "Pipeline Swarm"
-    also_known_as: "Sequential Chain"
-    pattern: |
-      Agent 1 → output → Agent 2 → output → Agent 3 → final output
-    characteristics:
-      - "Each agent processes previous output"
-      - "Clear handoff protocol"
-      - "Stage gates between agents"
-    strengths:
-      - "Predictable, auditable"
-      - "Easy to insert QA gates"
-    weaknesses:
-      - "Sequential bottleneck"
-      - "Early errors propagate"
-    best_for: "Document processing, code review pipeline, editorial workflows"
-    sinapse_example: "@sprint-lead → @product-lead → @developer → @quality-gate → @devops"
-    frameworks: ["SINAPSE SDC workflow", "LangChain chains", "AIOX SDC"]
-
-  - name: "Parallel / Fan-out Swarm"
-    also_known_as: "Map-Reduce for Agents"
-    pattern: |
-      Orchestrator
-          ├── [spawn] Agent A → result A
-          ├── [spawn] Agent B → result B
-          └── [spawn] Agent C → result C
-                          ↓
-                    [merge] Aggregator
-    characteristics:
-      - "Tasks split across parallel agents"
-      - "Results aggregated at end"
-      - "Dramatically reduces wall-clock time"
-    strengths:
-      - "Speed via parallelism"
-      - "Scales with task complexity"
-    weaknesses:
-      - "Complex result merging"
-      - "Cost multiplies with agents"
-    best_for: "Research tasks, parallel code generation, multi-file refactoring"
-    implementation: "Claude Code Agent tool with parallel spawning"
-    sinapse_example: "obra/superpowers 'dispatching-parallel-agents' skill"
-
-# ---------------------------------------------------------------------------
-# AGENT COMMUNICATION PROTOCOLS
-# ---------------------------------------------------------------------------
-communication_protocols:
-
-  - name: "Message Passing"
-    description: "Agents send structured messages to each other"
-    formats: ["JSON tool calls", "Structured YAML", "Markdown artifacts"]
-    latency: "Low (synchronous)"
-    best_for: "Tight orchestration, real-time coordination"
-    claude_code_impl: "Agent tool spawning with system prompt handoff"
-
-  - name: "Shared Memory / Blackboard"
-    description: "Agents read/write a common knowledge store"
-    implementations:
-      - "Story files (docs/stories/)"
-      - "Scratchpad (.sinapse/scratchpad/{story-id}/)"
-      - "MEMORY.md (auto-updated)"
-      - "Agent memory files (.claude/agent-memory/)"
-    latency: "Variable (async)"
-    best_for: "Cross-session persistence, multi-agent context sharing"
-
-  - name: "Handoff Artifacts"
-    description: "Structured summaries passed between agents on switch"
-    sinapse_spec: ".claude/rules/agent-handoff.md"
-    max_size: "500 tokens"
-    fields: ["story_context", "decisions", "files_modified", "blockers", "next_action"]
-    best_for: "Context window efficiency on agent switches"
-
-  - name: "Event Bus / Hooks"
-    description: "Lifecycle events trigger automated agent responses"
-    claude_code_events:
-      - "PreToolUse — before any tool execution"
-      - "PostToolUse — after tool completion"
-      - "Stop — agent turn ends"
-      - "PreCompact — context compaction approaching"
-      - "UserPromptSubmit — user input received"
-      - "SubagentStart / SubagentStop — parallel agent lifecycle"
-    codex_events: ["SessionStart", "PreToolUse", "PostToolUse", "UserPromptSubmit", "Stop"]
-    best_for: "Governance enforcement, automated QA, session state preservation"
-
-# ---------------------------------------------------------------------------
-# TASK DECOMPOSITION STRATEGIES
-# ---------------------------------------------------------------------------
-task_decomposition:
-
-  - strategy: "Recursive Decomposition"
-    description: "Break task into subtasks until atomic and delegatable"
-    pattern: |
-      Epic → Stories → Tasks → Subtasks
-    sinapse_impl: "Epic → story files → task checklist items"
-    best_for: "Large initiatives with unknown scope"
-
-  - strategy: "Domain Partitioning"
-    description: "Split by expertise domain — each domain gets one agent"
-    pattern: |
-      Full request → [route by domain] → specialist agents
-    sinapse_impl: "sinapse-orqx routes to squad orchestrators"
-    best_for: "Multi-domain requests (brand + content + growth)"
-
-  - strategy: "Pipeline Staging"
-    description: "Sequential stages with quality gates between them"
-    pattern: |
-      Stage 1 → [gate] → Stage 2 → [gate] → Stage 3
-    sinapse_impl: "SDC: create → validate → implement → QA gate"
-    best_for: "Quality-critical workflows"
-
-  - strategy: "Parallel Exploration"
-    description: "Multiple agents explore solution space simultaneously"
-    pattern: |
-      Problem → [fork N agents] → [merge best solution]
-    implementation: "Claude Code subagents via Agent tool"
-    best_for: "Research tasks, alternative solution generation"
-    obra_superpowers: "dispatching-parallel-agents skill"
-
-  - strategy: "Role-Based Specialization"
-    description: "Assign roles matching agent expertise"
-    roles_example:
-      - "Analyst: requirements gathering, research"
-      - "Architect: design decisions, patterns"
-      - "Developer: implementation"
-      - "QA: testing, review"
-      - "Orchestrator: routing, coordination"
-    sinapse_impl: "Squad agent definitions with specific personas"
-    best_for: "Complex development workflows"
-
-# ---------------------------------------------------------------------------
-# SWARM FRAMEWORKS COMPARISON (2026)
-# ---------------------------------------------------------------------------
-frameworks_comparison:
-
-  - name: "CrewAI"
-    type: "Python framework"
-    topology: "Hierarchical (Process.hierarchical) or Sequential"
-    agents_model: "YAML/Python defined with role + goal + backstory"
-    communication: "Task output passing"
-    strengths:
-      - "Clean API for multi-agent crews"
-      - "Built-in memory and caching"
-      - "Easy process definition"
-    weaknesses:
-      - "Python only"
-      - "Less suited for terminal/CLI workflows"
-    github_stars: "27K+ (2026)"
-    use_case: "Python-based multi-agent pipelines"
-
-  - name: "AutoGen (Microsoft)"
-    type: "Python framework"
-    topology: "Conversational (agents talk to each other)"
-    agents_model: "AssistantAgent + UserProxyAgent"
-    communication: "Direct message passing between agents"
-    strengths:
-      - "Flexible conversation flows"
-      - "Human-in-the-loop support"
-      - "Multi-modal agents"
-    weaknesses:
-      - "Complex configuration"
-      - "Token-heavy conversations"
-    use_case: "Research, multi-turn agent conversations"
-
-  - name: "LangGraph"
-    type: "Python framework (LangChain)"
-    topology: "Graph-based (directed + cycles supported)"
-    agents_model: "Nodes as agents, edges as transitions"
-    communication: "State passing through graph"
-    strengths:
-      - "Cycles and loops (unlike simple chains)"
-      - "Checkpointing for recovery"
-      - "Visual graph design"
-    weaknesses:
-      - "Steeper learning curve"
-      - "LangChain dependency"
-    use_case: "Complex stateful agent workflows"
-
-  - name: "Semantic Kernel (Microsoft)"
-    type: "Multi-language SDK (.NET, Python, Java)"
-    topology: "Plugin-based with planner"
-    agents_model: "Skills (plugins) + AI planner"
-    communication: "Function calling"
-    strengths:
-      - "Enterprise-ready (.NET)"
-      - "Multi-language"
-      - "Azure integration"
-    weaknesses:
-      - "Heavier abstraction"
-      - "Less suited for terminal agents"
-    use_case: "Enterprise .NET applications with AI"
-
-  - name: "Claude Agent SDK (Anthropic)"
-    type: "TypeScript/Python SDK"
-    topology: "Hierarchical via Agent tool spawning"
-    agents_model: "Subagents via claude --agent or Agent tool"
-    communication: "System prompt injection + tool results"
-    strengths:
-      - "Native to Claude Code"
-      - "Full tool access per agent"
-      - "No external dependency"
-    weaknesses:
-      - "Claude API only"
-      - "Cost scales with parallel agents"
-    use_case: "Claude Code subagent orchestration, SINAPSE agents"
-
-  - name: "SINAPSE"
-    type: "Meta-framework (Claude Code native)"
-    topology: "Hierarchical squads with stigmergic state (story files)"
-    agents_model: "Squad YAML + agent .md definitions + Constitution"
-    communication: "Handoff artifacts + story files + scratchpad"
-    strengths:
-      - "Constitutional governance"
-      - "Multi-domain (20+ squads)"
-      - "Quality gates automated"
-      - "CLI-first with hooks"
-    weaknesses:
-      - "Claude Code only currently"
-      - "Distribution still local"
-    use_case: "Full-stack product development with specialized domain squads"
-
-# ---------------------------------------------------------------------------
-# SINAPSE SWARM ARCHITECTURE SPECIFICS
-# ---------------------------------------------------------------------------
-sinapse_swarm_architecture:
-
-  topology: "Hierarchical + Stigmergic hybrid"
-
-  delegation_chain:
-    - level: 1
-      agent: "sinapse-orqx (Imperator)"
-      role: "Cross-domain routing, Constitutional enforcement"
-    - level: 2
-      agent: "{squad}-orqx"
-      role: "Intra-squad routing, specialist coordination"
-    - level: 3
-      agent: "Specialist agents"
-      role: "Domain execution (code, copy, research, etc.)"
-
-  stigmergic_state:
-    story_files: "docs/stories/ — persistent task state readable by all agents"
-    scratchpad: ".sinapse/scratchpad/{story-id}/ — inter-agent discoveries"
-    handoffs: ".sinapse/handoffs/ — compact context between switches"
-    agent_memory: ".claude/agent-memory/ — persistent learnings per agent"
-
-  quality_gates:
-    - gate: "Story validation (@product-lead)"
-      trigger: "Before any implementation begins"
-    - gate: "QA gate (@quality-gate)"
-      trigger: "After implementation, before PR"
-    - gate: "Hook governance (Constitution)"
-      trigger: "On every tool use"
-
-  constitutional_rules:
-    - "Mandatory delegation — orchestrators never execute domain work"
-    - "Documentation-first — story required before code"
-    - "Agent authority — @devops exclusive for git push"
-    - "No invention — all output traceable to requirements"