@ztffn/presentation-generator-plugin 1.3.0 → 1.3.2

package/.claude-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "presentation-generator",
   "description": "Generate complete graph-based presentations from natural language briefs and project documents. Includes a three-agent pipeline: content extraction, narrative design, and graph JSON compilation.",
-  "version": "1.3.0",
+  "version": "1.3.1",
   "author": {
     "name": "Huma"
   },

package/agents/presentation-content.md

@@ -10,23 +10,47 @@ skills:
 
 You are a content extraction specialist. Your job is to read source documents and produce a structured content brief that captures everything needed to build a compelling presentation.
 
-## Your Process
+## Process — follow these steps in order
 
-1. Read all document paths provided to you
-2. Apply the content signal guidance from your pre-injected skill to identify what is presentation-worthy
-3. Populate every field of the content brief schema with specific, concrete entries
-4. Write the result to `_temp/presentation-content-brief.json`
+### Step 1 — Load your skill files
 
-## Rules
+Read this file before doing anything else:
+- Find and read the file matching `**/presentation-generator/skills/content-signals/SKILL.md`
 
-- Extract strong, specific entries for every field. Weak or vague entries (e.g. "saves money", "good technology") are unacceptable.
+This file defines the content brief schema, signal priorities, validation rules, and strong/weak examples. You must follow it exactly.
+
+### Step 2 — Read all source documents
+
+Read every document path provided in the prompt. Extract presentation-worthy content using the signal guidance from Step 1.
+
+### Step 3 — Populate the content brief
+
+Build a JSON object conforming to the content brief schema from your content-signals skill. Every field must have specific, concrete entries.
+
+Rules:
 - `keyMessages` must be declarative sentences with subject + verb + specific claim
-- `dataPoints` must contain at least one entry with a concrete number
+- `dataPoints` must contain at least one entry with a concrete number in `value`
 - `callToAction` must be a single sentence with an action verb
 - Never include layout, design, slide type, or visual treatment in the brief
 - Never invent data — only extract what the documents contain. If a field cannot be populated from the source material, note it as `"NOT FOUND IN SOURCE — ask user"`
 - When multiple documents conflict, prefer the most recent or most authoritative source
 
-## Output
+### Step 4 — Validate the brief
+
+Before writing, verify:
+1. `keyMessages` contains 3-5 declarative claims (not topic labels)
+2. `dataPoints` has at least one entry with a specific number
+3. `callToAction` is a single actionable sentence
+4. No field contains layout, design, or slide-type references
+
+### Step 5 — Write output
+
+Write the JSON to `_temp/presentation-content-brief.json`.
+
+### Step 6 — Report to orchestrator
 
-Write a single JSON file to `_temp/presentation-content-brief.json` conforming to the content brief schema defined in your content-signals skill.
+Return a summary listing:
+- Number of key messages extracted
+- Number of data points extracted
+- Any fields marked as `"NOT FOUND IN SOURCE — ask user"`
+- Any gaps or weak areas the orchestrator should address before proceeding

package/agents/presentation-design.md

@@ -10,17 +10,52 @@ skills:
 
 You are a design and JSON generation specialist. Your job is to translate an approved slide outline into a complete, valid presentation graph JSON file.
 
-## Your Process
-
-1. Read the approved outline from `_temp/presentation-outline.md`
-2. For each slide in the outline, make three categories of decision:
-   a. **Slide type and layout**: Based on content signals, select `type`, `layout`, `centered`, and related fields per the graph-json-spec skill
-   b. **Visual treatment**: Set background media, `lightText`, branding flags, and chart configuration
-   c. **Positioning**: Place each node on the 240px/150px grid
-3. Build all nodes with complete `data`, `style`, and `measured` objects
-4. Wire all edges with bidirectional pairs per the edge wiring rules
-5. Run the edge validation checklist before finalizing
-6. Write the complete JSON to `_temp/presentation-draft.json`
+## Process — follow these steps in order
+
+### Step 1 — Load your skill file
+
+Read this file before doing anything else:
+- Find and read the file matching `**/presentation-generator/skills/graph-json-spec/SKILL.md`
+
+This file defines the complete JSON schema — node structure, allowed data fields, banned field names, edge wiring with handle IDs, positioning grid, visual intent mapping, and slide recipes. You must follow it exactly.
+
+### Step 2 — Read the approved outline
+
+Read `_temp/presentation-outline.md`. This is a markdown file containing the narrative structure and slide content.
+
+### Step 3 — Build the presentation JSON
+
+For each slide in the outline, make three categories of decision:
+  a. **Slide type and layout**: Based on content signals, select `type`, `layout`, `centered`, and related fields per the graph-json-spec skill
+  b. **Visual treatment**: Set background media, `lightText`, branding flags, and chart configuration
+  c. **Positioning**: Place each node on the 240px/150px grid
+
+Build all nodes with complete `data`, `style`, and `measured` objects. Wire all edges with bidirectional pairs per the edge wiring rules.
+
+### Step 4 — Write the JSON
+
+1. Derive the presentation slug from the deck title: lowercase kebab-case, strip special characters.
+2. Create the directory `presentations/{slug}/` if it does not exist.
+3. Write the JSON to `presentations/{slug}/{slug}.json`.
+
+### Step 5 — Run the validator
+
+Run the validator script using Bash — do not skip this step:
+
+```bash
+VALIDATE=$(find -L .claude ~/.claude -path "*/presentation-generator/scripts/validate_draft.py" 2>/dev/null | head -1)
+python3 "$VALIDATE" presentations/{slug}/{slug}.json
+```
+
+If there are any `ERROR` lines, fix every error in the JSON file and run the validator again. Repeat until the script prints `OK  Presentation JSON is valid`.
+
+### Step 6 — Report to orchestrator
+
+Your completion message must include:
+- The exact terminal output of the validator script (a Read of the JSON file does not count as validation)
+- Output path
+- Total node count and edge count
+- Any slides with video placeholders requiring manual upload
 
 ## Field Mapping
 

package/agents/presentation-narrative.md

@@ -12,14 +12,35 @@ skills:
 
 You are a narrative design specialist. Your job is to take a structured content brief and design a compelling presentation structure with detailed slide content.
 
-## Your Process
-
-1. Read the content brief from `_temp/presentation-content-brief.json`
-2. Based on `goal` and `audience`, select the appropriate narrative framework from your pre-injected frameworks skill
-3. Design the spine (main horizontal flow) of 4-7 nodes
-4. Design drill-down branches where content warrants optional depth
-5. Write detailed content for each slide following the quality guidance in your slide-content skill
-6. Write the result to `_temp/presentation-outline.md`
+## Process — follow these steps in order
+
+### Step 1 — Load your skill files
+
+Read these three files before doing anything else:
+- Find and read the file matching `**/presentation-generator/skills/frameworks/SKILL.md`
+- Find and read the file matching `**/presentation-generator/skills/slide-content/SKILL.md`
+- Find and read the file matching `**/presentation-generator/skills/graph-topology/SKILL.md`
+
+These files define framework selection logic, slide content quality rules, visual intent annotations, and topology patterns. You must follow them exactly.
+
+### Step 2 — Read the content brief
+
+Read `_temp/presentation-content-brief.json`. Note any weak or missing fields.
+
+### Step 3 — Select framework, design structure, write slide content
+
+Based on `goal` and `audience`, select the appropriate narrative framework from your frameworks skill. Then:
+1. Design the spine (main horizontal flow) of 4-7 nodes
+2. Design drill-down branches where content warrants optional depth
+3. Write detailed content for each slide following the quality guidance in your slide-content skill
+
+### Step 4 — Write output as MARKDOWN
+
+Write the result to `_temp/presentation-outline.md` — this must be a markdown file, NOT JSON.
+
+### Step 5 — Report to orchestrator
+
+Return a summary: framework selected, spine node count, drill-down node count, and any content brief gaps that need user input.
 
 ## Output Format
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ztffn/presentation-generator-plugin",
-  "version": "1.3.0",
+  "version": "1.3.2",
   "description": "Claude Code plugin for generating graph-based presentations",
   "bin": {
     "presentation-generator-plugin": "bin/index.js"

package/skills/presentation-generator/SKILL.md

@@ -49,7 +49,11 @@ Task tool:
   subagent_type: "presentation-content"
   description: "Extract content brief"
   prompt: |
-    Read these source documents and extract a structured content brief:
+    Step 1: Load your skill file. Find and read the file matching
+    **/presentation-generator/skills/content-signals/SKILL.md
+    — it defines the content brief schema, signal priorities, and validation rules.
+
+    Step 2: Read these source documents and extract a structured content brief:
 
     Documents:
     - [list each file path from Phase 1]
@@ -57,7 +61,10 @@ Task tool:
     Audience: [audience identified in Phase 1]
     Goal: [goal identified in Phase 1]
 
-    Write the result to _temp/presentation-content-brief.json
+    Step 3: Write the result as JSON to _temp/presentation-content-brief.json
+
+    Step 4: Report back with the number of key messages, data points,
+    and any fields marked "NOT FOUND IN SOURCE".
 ```
 
 The content agent reads all source documents and writes a structured content brief.
@@ -89,13 +96,21 @@ Task tool:
   subagent_type: "presentation-narrative"
   description: "Design narrative outline"
   prompt: |
-    Read the content brief at _temp/presentation-content-brief.json and design
-    the narrative structure for this presentation.
+    Step 1: Load your skill files. Find and read these three files:
+    - **/presentation-generator/skills/frameworks/SKILL.md
+    - **/presentation-generator/skills/slide-content/SKILL.md
+    - **/presentation-generator/skills/graph-topology/SKILL.md
 
+    Step 2: Read the content brief at _temp/presentation-content-brief.json
+
+    Step 3: Design the narrative structure for this presentation.
     Audience: [audience from Phase 1]
     Goal: [goal from Phase 1]
 
-    Write the result to _temp/presentation-outline.md
+    Step 4: Write the result as MARKDOWN to _temp/presentation-outline.md
+
+    Step 5: Report back with framework selected, spine node count,
+    drill-down node count, and any content brief gaps.
 ```
 
 The narrative agent selects a story framework, designs the spine and drill-down structure, and writes a detailed slide outline.
@@ -139,14 +154,24 @@ Task tool:
   subagent_type: "presentation-design"
   description: "Generate presentation JSON"
   prompt: |
-    Read the approved outline at _temp/presentation-outline.md and translate it
-    into a complete presentation graph JSON.
+    Step 1: Load your skill file. Find and read the file matching
+    **/presentation-generator/skills/graph-json-spec/SKILL.md
+    — it defines the complete JSON schema, allowed fields, banned fields,
+    edge wiring, handle IDs, and positioning grid.
 
+    Step 2: Read the approved outline at _temp/presentation-outline.md
+
+    Step 3: Translate the outline into a complete presentation graph JSON.
     Presentation slug: {slug}
-    Output path: presentations/{slug}/{slug}.json
 
-    After writing the JSON, run the validator script and fix any errors until
-    it passes. Include the validator terminal output in your completion message.
+    Step 4: Write the JSON to presentations/{slug}/{slug}.json
+
+    Step 5: Run the validator script and fix any errors until it passes:
+    VALIDATE=$(find -L .claude ~/.claude -path "*/presentation-generator/scripts/validate_draft.py" 2>/dev/null | head -1)
+    python3 "$VALIDATE" presentations/{slug}/{slug}.json
+
+    Step 6: Report back with the exact validator terminal output,
+    output path, node count, edge count, and any video placeholders.
 ```
 
 The design agent translates each slide into a fully specified graph node, wires all edges with bidirectional pairs, and validates the result.

package/.claude/settings.local.json

@@ -1,9 +0,0 @@
-{
-  "permissions": {
-    "allow": [
-      "WebFetch(domain:support.claude.com)",
-      "WebFetch(domain:github.com)",
-      "WebFetch(domain:raw.githubusercontent.com)"
-    ]
-  }
-}

package/.github/workflows/publish.yml

@@ -1,61 +0,0 @@
-name: Publish
-
-on:
-  push:
-    tags:
-      - "v*"
-
-jobs:
-  publish-npm:
-    name: Publish to npm
-    runs-on: ubuntu-latest
-    permissions:
-      contents: read
-      id-token: write
-    steps:
-      - uses: actions/checkout@v4
-
-      - uses: actions/setup-node@v4
-        with:
-          node-version: "20"
-          registry-url: "https://registry.npmjs.org"
-
-      - run: npm publish --access public
-        env:
-          NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }}
-
-  release-zip:
-    name: Create GitHub Release with zip
-    runs-on: ubuntu-latest
-    permissions:
-      contents: write
-    steps:
-      - uses: actions/checkout@v4
-
-      - name: Build zip
-        run: |
-          VERSION=${GITHUB_REF_NAME}
-          ZIP_NAME="presentation-generator-plugin-${VERSION}.zip"
-          zip -r "$ZIP_NAME" . \
-            --exclude "*.git*" \
-            --exclude ".github/*" \
-            --exclude "node_modules/*"
-          echo "ZIP_NAME=$ZIP_NAME" >> $GITHUB_ENV
-
-      - name: Create GitHub Release
-        uses: softprops/action-gh-release@v2
-        with:
-          name: ${{ github.ref_name }}
-          body: |
-            ## Install via zip (non-technical)
-
-            1. Download `${{ env.ZIP_NAME }}` below
-            2. Unzip it anywhere on your machine
-            3. In Claude Code, run: `claude --plugin-dir /path/to/presentation-generator-plugin`
-
-            ## Install via npx (developers)
-
-            ```bash
-            npx @ztffn/presentation-generator-plugin install
-            ```
-          files: ${{ env.ZIP_NAME }}

package/CONTEXT.md

@@ -1,110 +0,0 @@
-# Context: Presentation Generator Plugin
-
-## What this plugin is
-
-A Claude Code plugin (`@ztffn/presentation-generator-plugin`) that generates graph-based presentation JSON for the Huma Showcase renderer. It uses a three-agent pipeline:
-
-1. **presentation-content** — reads source documents, produces `_temp/presentation-content-brief.json`
-2. **presentation-narrative** — reads the brief, produces `_temp/presentation-outline.md` (story structure, slide content)
-3. **presentation-design** — reads the outline, produces `presentations/{slug}/{slug}.json` (graph JSON with nodes, edges, positions)
-
-The orchestrator skill (`skills/presentation-generator/SKILL.md`) spawns each agent via the Task tool.
-
-## Repository layout
-
-```
-presentation-generator-plugin/
-├── .claude-plugin/plugin.json      # Plugin manifest (v1.3.0)
-├── package.json                     # npm package (v1.3.0)
-├── bin/index.js                     # Installer script
-├── agents/
-│   ├── presentation-content.md      # Agent 1: content extraction
-│   ├── presentation-narrative.md    # Agent 2: narrative design
-│   └── presentation-design.md       # Agent 3: JSON generation
-├── skills/
-│   ├── presentation-generator/      # Orchestrator skill
-│   │   ├── SKILL.md
-│   │   └── presentation-guide.md    # Generic presentation craft guide
-│   ├── content-signals/SKILL.md     # Content brief schema + extraction patterns
-│   ├── frameworks/SKILL.md          # 5 narrative frameworks (PSV, BAB, etc.)
-│   ├── slide-content/SKILL.md       # Slide writing quality rules
-│   ├── graph-topology/SKILL.md      # Spine/drill-down structure patterns
-│   └── graph-json-spec/SKILL.md     # Complete graph JSON spec (node schema, edges, positioning, layouts, recipes, reference example)
-└── scripts/
-    └── validate_draft.py            # Validates generated JSON
-```
-
-## How skills reach agents
-
-Each agent file has a `skills:` frontmatter listing plugin-namespaced skill references:
-
-```yaml
-# agents/presentation-design.md
-skills:
-  - presentation-generator:graph-json-spec
-```
-
-Per Claude Code docs, the full content of each listed skill should be injected into the sub-agent's context at startup. The agent's markdown body (below the frontmatter) becomes the system prompt. The `prompt` field from the Task tool call becomes the user message.
-
-## Resilience: self-sufficient agents
-
-Because `skills:` injection has proven unreliable, each agent's markdown body now contains the critical subset of its skill content inlined directly. The design agent body includes: valid enum values, positioning grid, chart config structure, and a reference node example. This means the agent produces correct JSON even when skill injection fails entirely.
-
-## The three agent responsibilities (DO NOT MIX)
-
-### presentation-content
-- **Input:** Source document paths, audience, goal
-- **Output:** `_temp/presentation-content-brief.json`
-- **Knows about:** Content extraction patterns, brief schema, what makes content presentation-worthy
-- **Does NOT know about:** Narrative frameworks, slide design, JSON format, node schema, edges
-
-### presentation-narrative
-- **Input:** Content brief
-- **Output:** `_temp/presentation-outline.md`
-- **Knows about:** Story frameworks (PSV, BAB, etc.), graph topology (spine/drill-down), slide content quality (headlines, assertion-evidence, speaker notes)
-- **Does NOT know about:** JSON format, node schema fields, edge handles, positioning grid, layout/type values
-
-### presentation-design
-- **Input:** Approved outline
-- **Output:** `presentations/{slug}/{slug}.json`
-- **Knows about:** Node schema (SlideNodeData fields), edge conventions (handle IDs), positioning grid, layout templates (type/layout decisions), slide recipes
-- **Does NOT know about:** Story frameworks, content extraction, audience analysis
-
-## The renderer (authoritative, do not modify)
-
-Lives at `/Users/steffen/Projects/huma/humashowcase`. Key types from `src/types/presentation.ts`:
-
-### SlideNodeData allowed fields
-`label`, `topic`, `content`, `notes`, `type`, `centered`, `layout`, `lightText`, `brandFont`, `showBranding`, `brandingText`, `backgroundImage`, `backgroundImageFit`, `backgroundImageOverlay`, `backgroundVideo`, `backgroundVideoFit`, `backgroundVideoLoop`, `inlineVideoControls`, `inlineVideoAutoplay`, `inlineVideoLoop`, `scene`, `chart`, `charts`, `sceneGroup`, `focus`
-
-### Valid enums
-- `data.type`: `"content"` | `"r3f"` | `"chart"` | `"custom"` (default: `"content"`)
-- `data.layout`: `"single"` | `"two-column"` (default: `"single"`)
-
-### Correct node structure
-```json
-{
-  "id": "kebab-case-slug",
-  "type": "huma",
-  "position": { "x": 240, "y": 0 },
-  "data": {
-    "label": "Slide title goes here",
-    "content": "## Markdown body\n\n- Bullet points\n- More content",
-    "notes": "Speaker notes go here",
-    "centered": false
-  },
-  "style": { "width": 180, "height": 70 },
-  "measured": { "width": 180, "height": 70 }
-}
-```
-
-### Correct edge structure
-```json
-{ "id": "e-a-b", "source": "a", "target": "b", "sourceHandle": "s-right", "targetHandle": "t-left" }
-{ "id": "e-b-a", "source": "b", "target": "a", "sourceHandle": "s-left", "targetHandle": "t-right" }
-```
-
-### Positioning grid
-- Horizontal spacing: 240px
-- Vertical spacing: 150px
-- Node size: always 180x70

package/HANDOFF.md

@@ -1,49 +0,0 @@
-# Handoff: Presentation Generator Plugin v1.3.0
-
-## Current state
-
-The plugin generates correct graph JSON. The design agent is resilient to skills injection failure because critical rules (enums, grid, chart config, reference node) are inlined in its body.
-
-## Architecture
-
-Three-agent pipeline orchestrated by `skills/presentation-generator/SKILL.md`:
-
-| Agent | Skills | Purpose |
-|---|---|---|
-| `presentation-content` | `content-signals` | Extract structured brief from documents |
-| `presentation-narrative` | `frameworks`, `slide-content`, `graph-topology` | Design story structure and slide content |
-| `presentation-design` | `graph-json-spec` | Translate outline to valid graph JSON |
-
-## Key files
-
-| File | Purpose |
-|---|---|
-| `skills/graph-json-spec/SKILL.md` | Complete graph JSON specification — node schema, edges, positioning, layouts, recipes, reference example |
-| `agents/presentation-design.md` | Design agent with inlined critical rules as fallback |
-| `scripts/validate_draft.py` | Validates generated JSON against renderer types |
-
-## Known past failures (resolved)
-
-1. **Skills injection failure** — agent invented its own schema (wrong field names, wrong enums, wrong spacing). Fixed by inlining critical rules into the agent body.
-2. **Fragmented skills** — 5 separate skills (`node-schema`, `edge-conventions`, `layout-templates`, `positioning`, `pitch-reference`) all had to inject correctly. Consolidated into single `graph-json-spec` skill.
-3. **Wrong field names** — `title`/`subtitle`/`bullets` instead of `label`/`content`/`notes`. Banned field list is in both the agent body and the consolidated skill.
-4. **Bare edge handles** — `right`/`left` instead of `s-right`/`t-left`. Handle rules are in both the agent body and the consolidated skill.
-5. **Missing `measured` field** — validator now enforces this.
-
-## The renderer (do not modify)
-
-Lives at `/Users/steffen/Projects/huma/humashowcase`. The renderer is authoritative — the plugin must match it. Key source of truth: `src/types/presentation.ts` (`SlideNodeData` interface).
-
-## Validation
-
-Run against any generated JSON:
-
-```bash
-python3 scripts/validate_draft.py presentations/{slug}/{slug}.json
-```
-
-Or against the reference demo:
-
-```bash
-python3 scripts/validate_draft.py "/Users/steffen/Downloads/presentation-demo-2(1).json"
-```

package/PLAN-consolidate-design-skills.md

@@ -1,238 +0,0 @@
-# Plan: Consolidate Design Agent Skills
-
-## Objective
-
-Merge the 5 fragmented design-agent skills (`node-schema`, `edge-conventions`, `layout-templates`, `positioning`, `pitch-reference`) into a single consolidated skill AND inline its critical content into the `agents/presentation-design.md` body. The agent must produce correct JSON even when `skills:` injection fails entirely.
-
-## Problem
-
-The design agent currently references 5 separate skills via frontmatter:
-```yaml
-skills:
-  - presentation-generator:node-schema
-  - presentation-generator:edge-conventions
-  - presentation-generator:layout-templates
-  - presentation-generator:positioning
-  - presentation-generator:pitch-reference
-```
-
-Per Claude Code docs, `skills:` in subagent frontmatter should inject full skill content at startup. In practice, this injection fails — the agent invents its own schema (wrong field names, wrong enums, wrong spacing, bare handles). Five skills must ALL inject correctly; if even one fails, the output breaks.
-
-## What to do
-
-### Step 1: Create consolidated skill
-
-Create `skills/graph-json-spec/SKILL.md` — a single comprehensive skill containing everything the design agent needs. Merge content from all 5 skills into one coherent document, organized in the order the agent needs it during JSON generation.
-
-**Frontmatter:**
-```yaml
----
-name: graph-json-spec
-description: >
-  Complete specification for generating valid graph-based presentation JSON —
-  node schema, edge wiring, positioning grid, layout decisions, and slide recipes.
-  Used by the presentation-design agent.
-user-invocable: false
----
-```
-
-**Document structure** (in this order):
-
-1. **Quick Reference** — The essential rules on one screen. Field names, valid enums, grid values, handle IDs. This section alone should prevent the most common errors.
-
-2. **Node Structure** — Merge from `node-schema`. Full node wrapper (`id`, `type: "huma"`, `position`, `data`, `style`, `measured`). All `SlideNodeData` fields with types and defaults. Keep the grouped tables (content fields, layout/display, background media, inline video, R3F scene, charts). Include the banned field names list from the current agent body.
-
-3. **Edge Wiring** — Merge from `edge-conventions`. Handle ID table, bidirectional pair rule, standard edge pair patterns (spine, drill-down, sibling), edge ID convention, validation checklist.
-
-4. **Positioning Grid** — Merge from `positioning`. Grid constants (240px/150px/180×70), spine row layout, drill-down row placement, sibling spacing, quick reference position table.
-
-5. **Design Decisions** — Merge from `layout-templates`. Slide type selection, layout decisions (when two-column, when centered, when brand font, when branding), background treatment, text contrast, chart decisions (full-viewport vs inline, chart type selection), R3F scene decisions, topic badge conventions.
-
-6. **Slide Recipes** — Merge from `layout-templates`. All 7 recipe patterns (cover/CTA, standard bullets, impact statement, two-column comparison, text+inline chart, proof point/customer quote, section opener with background image). These are complete `data` field examples the agent can pattern-match against.
-
-7. **Complete Reference Example** — Merge from `pitch-reference`. The full 7-node JSON with all edges. But UPDATE it to match the validated demo at `/Users/steffen/Downloads/presentation-demo-2(1).json`. Key differences to adopt from the demo:
-   - The demo uses `"meta": { "updatedAt": "..." }` — the plugin should use `"meta": { "name": "..." }` (per the current agent spec). Keep the plugin's `meta.name` convention.
-   - Edge IDs: the demo uses short IDs like `"e1"`, `"e2"` for spine and descriptive IDs like `"e-prob-detail"` for drill-downs. Either convention is fine, but the reference should be consistent — use the descriptive `e-{source}-{target}` pattern throughout.
-   - The demo has 20 nodes and 40 edges — it's a real-world example. Consider including a trimmed version (keep the 7-node pitch reference) but add 2-3 realistic patterns from the demo that the current pitch-reference doesn't cover:
-     - A slide with `style.backgroundColor` set (the `feat-style` node uses `"backgroundColor": "#1a1a2e"`)
-     - A drill-down with a second-level child (the demo has `feat-markdown` → `feat-tables` at y:300)
-     - Sibling drill-downs connected horizontally (the demo has `feat-markdown` → `feat-charts` via `s-right`/`t-left`)
-
-8. **Media Delivery Summary Format** — The template for listing placeholder media at the end of output.
-
-**Important content decisions:**
-- DO NOT include narrative frameworks, content extraction patterns, or audience analysis — those belong to other agents.
-- DO NOT include the presentation guide / playbook content — that's for the narrative agent.
-- DO include the content markdown features list from node-schema (the agent needs to know what markdown syntax works).
-- DO include chart config structure with the `config` wrapper (the demo JSON shows `config: { xKey, yKeys, showGrid, showLegend }` — not top-level `xKey`/`yKey`).
-
-### Step 2: Update the design agent body
-
-Edit `agents/presentation-design.md` to inline the critical subset — the minimum needed for correct output even if skills injection fails completely. The agent body should contain:
-
-**Keep everything currently in the body** (field mapping table, banned fields, edge handle rules, node structure rules, output format, validator invocation).
-
-**Add these sections from the consolidated skill:**
-
-1. **Valid Enums** — Add directly after the field mapping table:
-   ```
-   ## Valid Enum Values
-
-   - `data.type`: `"content"` (default) | `"r3f"` | `"chart"` | `"custom"`
-   - `data.layout`: `"single"` (default) | `"two-column"` (splits content on `---`)
-   - `data.centered`: `true` (default) | `false`
-   - `data.backgroundImageFit`: `"cover"` (default) | `"contain"`
-   - `data.backgroundVideoFit`: `"cover"` (default) | `"contain"`
-   ```
-
-2. **Positioning Grid** — Add a compact version:
-   ```
-   ## Positioning Grid
-
-   - Horizontal spacing: 240px between nodes
-   - Vertical spacing: 150px between levels
-   - Node size: always `{ width: 180, height: 70 }` in both `style` and `measured`
-   - Spine: all at y:0, x increments by 240 (0, 240, 480, 720...)
-   - Drill-down level 1: y:150, same x as parent
-   - Drill-down level 2: y:300
-   - Siblings at same depth: increment x by 240
-   ```
-
-3. **Chart Config Structure** — Add a compact example showing the `config` wrapper:
-   ```
-   ## Chart Config
-
-   Charts use a `config` wrapper for axis and display settings:
-   { "chartType": "bar", "data": [...], "config": { "xKey": "...", "yKeys": [...], "showGrid": true, "showLegend": true } }
-
-   Inline: put in `charts` record, reference with `[chart:name]` in content.
-   Full-viewport: put in `chart` field, set `type: "chart"`.
-   ```
-
-4. **Compact Reference Node** — One complete node example showing all common fields:
-   ```json
-   {
-     "id": "problem",
-     "type": "huma",
-     "position": { "x": 240, "y": 0 },
-     "data": {
-       "label": "The Problem",
-       "topic": "01 / Problem",
-       "content": "## Linear tools break complex stories\n\n- Point one\n- Point two",
-       "notes": "Speaker notes here",
-       "centered": false
-     },
-     "style": { "width": 180, "height": 70 },
-     "measured": { "width": 180, "height": 70 }
-   }
-   ```
-
-**Update the `skills:` frontmatter** to reference the single consolidated skill:
-```yaml
-skills:
-  - presentation-generator:graph-json-spec
-```
-
-### Step 3: Update the orchestrator
-
-Edit `skills/presentation-generator/SKILL.md` — update the Sub-Agent Reference table:
-
-```
-| Agent | Skills Loaded | Purpose |
-|---|---|---|
-| `presentation-content` | `content-signals` | Extract structured brief from documents |
-| `presentation-narrative` | `frameworks`, `slide-content`, `graph-topology` | Design story structure and slide content |
-| `presentation-design` | `graph-json-spec` | Translate outline to valid graph JSON |
-```
-
-### Step 4: Delete the old skill directories
-
-Remove these 5 directories (they're fully superseded):
-- `skills/node-schema/`
-- `skills/edge-conventions/`
-- `skills/layout-templates/`
-- `skills/positioning/`
-- `skills/pitch-reference/`
-
-### Step 5: Update validator awareness
-
-Read `scripts/validate_draft.py` and verify the consolidated skill's field lists match. The validator's `ALLOWED_DATA_FIELDS` set and handle sets must be consistent with what's documented in the skill. Current validator allows 24 data fields:
-
-```python
-ALLOWED_DATA_FIELDS = {
-    "label", "topic", "content", "notes", "type", "centered", "layout",
-    "lightText", "brandFont", "showBranding", "brandingText",
-    "backgroundImage", "backgroundImageFit", "backgroundImageOverlay",
-    "backgroundVideo", "backgroundVideoFit", "backgroundVideoLoop",
-    "inlineVideoControls", "inlineVideoAutoplay", "inlineVideoLoop",
-    "scene", "chart", "charts", "sceneGroup", "focus"
-}
-```
-
-The consolidated skill should list these same 24 fields. Cross-check against the demo JSON — the demo also uses `"selected": false` at the node level (not in `data`), which is fine (it's a React Flow runtime field, not part of `data`).
-
-Note: the demo JSON uses `meta.updatedAt` instead of `meta.name`. The validator checks for `meta.name` as non-empty. The consolidated skill should document that `meta.name` is required (the validator enforces it). The orchestrator can add `updatedAt` if desired but it's not checked.
-
-### Step 6: Update package version
-
-Bump version in `package.json` and `.claude-plugin/plugin.json` from `1.2.0` to `1.3.0`.
-
-### Step 7: Update CONTEXT.md and HANDOFF.md
-
-Update or remove the sections that describe the old fragmented skill structure. The new structure should be documented.
-
-### Step 8: Update README.md
-
-The Plugin structure section at the bottom lists the old nested structure. Update to reflect the actual flat structure with the new consolidated skill.
-
-## Files to create
-
-| File | Description |
-|---|---|
-| `skills/graph-json-spec/SKILL.md` | Consolidated design specification (merge of 5 skills) |
-
-## Files to modify
-
-| File | Changes |
-|---|---|
-| `agents/presentation-design.md` | Inline critical rules (enums, grid, chart config, reference node). Change `skills:` to reference `graph-json-spec` only. |
-| `skills/presentation-generator/SKILL.md` | Update sub-agent reference table |
-| `package.json` | Bump to 1.3.0 |
-| `.claude-plugin/plugin.json` | Bump to 1.3.0 |
-| `README.md` | Update plugin structure diagram |
-| `CONTEXT.md` | Update to reflect new structure |
-| `HANDOFF.md` | Update to reflect new structure |
-
-## Files to delete
-
-| File | Reason |
-|---|---|
-| `skills/node-schema/SKILL.md` | Merged into `graph-json-spec` |
-| `skills/edge-conventions/SKILL.md` | Merged into `graph-json-spec` |
-| `skills/layout-templates/SKILL.md` | Merged into `graph-json-spec` |
-| `skills/positioning/SKILL.md` | Merged into `graph-json-spec` |
-| `skills/pitch-reference/SKILL.md` | Merged into `graph-json-spec` |
-
-## Validation criteria
-
-After implementation, verify:
-
-1. `skills/graph-json-spec/SKILL.md` exists and contains all content from the 5 merged skills
-2. `agents/presentation-design.md` body contains: valid enums, positioning grid values, chart config structure, and a reference node example
-3. `agents/presentation-design.md` frontmatter references only `presentation-generator:graph-json-spec`
-4. The 5 old skill directories are deleted
-5. The validator still passes against the demo JSON:
-   ```bash
-   python3 scripts/validate_draft.py "/Users/steffen/Downloads/presentation-demo-2(1).json"
-   ```
-6. No narrative/content concepts leaked into the design skill (no framework names, no content extraction patterns, no audience analysis)
-7. The chart config in the skill uses the `config` wrapper structure (matching the demo JSON), not bare `xKey`/`yKey`
-
-## Reference files for the implementor
-
-These files contain the authoritative truth. Read them if anything is ambiguous:
-
-- **Valid demo JSON:** `/Users/steffen/Downloads/presentation-demo-2(1).json` — 20 nodes, 40 edges, all patterns
-- **Renderer types:** `/Users/steffen/Projects/huma/humashowcase/src/types/presentation.ts` — `SlideNodeData` interface
-- **Technical docs:** `/Users/steffen/Projects/huma/humashowcase/docs/technical/presentation-system-implementation.md`
-- **Validator:** `/Users/steffen/Projects/huma/presentation-generator-plugin/scripts/validate_draft.py`