create-yss-spec 2.1.4 → 2.1.6

package/template/.agents/skills/drawio/references/docs/edge-quality-rules.md

@@ -1,65 +0,0 @@
-# Edge Quality Rules
-
-Use these rules for professional draw.io routing. They apply to architecture diagrams, network diagrams, and paper figures.
-
-## Blocking Rules
-
-1. **Non-waypoint edges need full connection points**
-   - Set `exitX`, `exitY`, `entryX`, `entryY`, `exitDx=0`, `exitDy=0`, `entryDx=0`, `entryDy=0`.
-2. **Waypoint edges must not mix explicit connection points**
-   - If `waypoints` are present, remove manual exit/entry hints.
-3. **Corners are forbidden**
-   - Do not use `0/0`, `0/1`, `1/0`, `1/1` corner pairs. Use face centers or distributed slots.
-4. **Distribute edges on shared faces**
-   - A face with a single edge connects at its `0.5` center.
-   - When multiple edges leave or enter the same face, distribute them by their counterparts' projected positions (see Default Face Policy), keeping each edge collinear.
-5. **Preserve corridor spacing**
-   - Parallel corridors should be at least `30px` apart.
-6. **Keep the final segment readable**
-   - The segment entering the target should be at least `30px`.
-7. **Offset labels away from the line**
-   - draw.io centers an edge label on its offset point, so the offset must cover **half the label extent plus ~8px clearance** — `y=-12` only clears a single short line.
-   - Horizontal edges: `y = -(8 + labelHeight/2)`. Vertical edges: `x = 8 + labelWidth/2`.
-   - On a bent (fallback) edge the label anchors to the middle segment; offset perpendicular to that segment.
-   - Do not use `labelBackgroundColor` to hide the line.
-8. **Collinear before anything else**
-   - A vertical edge must have the same absolute X at exit and entry (`sourceX + exitX × sourceWidth == targetX + entryX × targetWidth`); a horizontal edge the same absolute Y. Compute fractions from the shared absolute coordinate — never pick `0.25/0.5/0.75` blindly on both ends.
-   - Anchor the shared coordinate on the center of the narrower face, clamped into the faces' overlap interval (8px off corners).
-   - Only when the two faces have no overlap on the shared axis is a bend legitimate; prefer explicit waypoints there.
-9. **Native bound edges only**
-   - Every connector is an `edge="1"` cell with both `source` and `target` referencing node ids, so it stays attached when shapes move.
-   - Never simulate a connector with standalone arrow shapes (`shape=singleArrow`, `shape=triangle`, `mxgraph.arrows2.*`) or floating edges positioned by raw coordinates. `--validate` reports both.
-10. **Plain text boxes stay transparent**
-    - Standalone text (captions, callouts, vertical labels) always renders `fillColor=none;strokeColor=none;labelBackgroundColor=none`. The converter ignores white fills on `type: text` and warns; use a shape node or `formula` type when a filled label is intended.
-
-## Arrowhead Defaults
-
-- Connectors default to a bold **open** head: `endArrow=open;endSize=12` (an unfilled two-line "V"; and `startSize=12` when a start arrow exists). Small stock arrowheads read as afterthoughts on 2px architecture connectors, so open heads still carry `endSize=12`.
-- Filled heads are opt-in: explicitly-requested `block`/`classic` and UML/ER semantic markers (e.g. inheritance `endArrow=block;endFill=0`, composition `endArrow=diamond;endFill=1`) keep their own fill and also get `endSize=12`.
-- Override per edge via `edge.style.endArrow` / `edge.style.endSize`, or per theme via `theme.connector.<type>.endSize`.
-
-## Audit Checklist
-
-- Does any edge cross an unrelated node?
-- Do two edges share the same face slot?
-- Is any arrow forced through a corner?
-- Is the target entry segment too short?
-- Is any no-waypoint edge bent although the two faces overlap on the shared axis (a collinear solution exists)?
-- Is every connector a bound edge (`source` + `target` set), with no arrow-shape stand-ins?
-- Does any edge label sit on its own line, cross another connector, or overlap another label?
-- Are waypoint arrays free of duplicate points?
-
-## Default Face Policy
-
-- Left/right faces vary on `Y`.
-- Top/bottom faces vary on `X`.
-- Each edge takes the shared absolute coordinate derived from its counterpart (narrower-face center, clamped into the overlap interval). Edges that land within `30px` of each other on the same face spread symmetrically around their mean — both endpoints move together so the line stays straight. Bidirectional pairs become two parallel straight lines.
-- The `0.25 / 0.5 / 0.75 / 0.33 / 0.66` slot sequence is only the fallback for edges whose faces have no overlap on the shared axis.
-
-## When to Escalate
-
-Use `--strict` when:
-
-- The diagram is paper-facing.
-- The diagram is dense enough that routing quality affects readability.
-- The output is intended to be reused as a reference asset or screenshot.

package/template/.agents/skills/drawio/references/docs/ieee-network-diagrams.md

@@ -1,145 +0,0 @@
-# IEEE Network Diagram Standards for Draw.io
-
-This guide focuses on creating network diagrams that adhere to IEEE and academic standards (e.g., IEEE 315, ANSI Y32.2) using Draw.io.
-
-> **Related**: See `math-typesetting.md` for LaTeX formula reference and IEEE domain equations.
-
-## 1. Symbol Standardization
-
-### 1.1 Core Devices
-
-* **Router**: Circle with four arrows pointing inward (X shape) or standard cylinder with arrows.
-* **Switch**: Square/Box with diverging arrows.
-* **Firewall**: Brick wall icon.
-* **Server**: Tower or Rack icon.
-* **Cloud/Internet**: Cloud shape.
-
-### 1.2 Link Representation
-
-* **Solid Line**: Physical Ethernet/Fiber link.
-* **Dashed Line**: Logical link (VLAN, VPN, Tunnel).
-* **Lightning Bolt / Zigzag**: Wireless or WAN link.
-* **Bus Line**: Thick straight line for Ethernet bus (older style) or Backbone.
-
-## 2. Information Hierarchy
-
-### 2.1 Topology Layers
-
-Organize the diagram into logical tiers:
-
-1. **Core Layer**: High-speed backbone switches/routers.
-2. **Distribution Layer**: Aggregation switches.
-3. **Access Layer**: End-user switches and APs.
-
-### 2.2 Visual Cues
-
-* **Size**: Core devices should appear slightly larger or more prominent.
-* **Position**: Core at top or center; Access at bottom or periphery.
-* **Grouping**: Use dashed rectangles to group devices by:
-  * Physical Location (Building A, Floor 1)
-  * Logical Segment (DMZ, Internal, Guest)
-
-## 3. Labeling & Annotations (IEEE Style)
-
-* **Device Names**: Use standardized codes (e.g., `R1`, `SW-Core-01`).
-* **Interfaces**: Label ports near the connection (e.g., `Gi0/1`, `eth0`).
-* **IP Addresses**: Small font (8-9pt) below the device or interface.
-* **Titles**: Figure Number and Title at the bottom (e.g., "Fig. 1. Proposed Network Architecture").
-* **Legend**: **Mandatory** for academic papers. Explain all line types, colors, and non-standard icons.
-
-## 4. Black & White / Greyscale Optimization
-
-Academic papers often print in B&W.
-
-* **Avoid Color-Only Coding**: Don't rely solely on Red vs. Green. Use Solid vs. Dashed lines.
-* **Patterns**: Use fill patterns (dots, stripes) for zones instead of just solid colors.
-* **Contrast**: Ensure text is black on white or white on black.
-
-## 5. Draw.io Implementation Steps
-
-1. **Load Libraries**: `More Shapes > Networking > Cisco / Azure / AWS` (or standard "General").
-2. **Set Page Setup**: A4 or Letter, Portrait/Landscape.
-3. **Draw Backbone**: Place Core devices first.
-4. **Connect**: Use "Waypoints" to make orthogonal lines.
-5. **Annotate**: Add text labels. Use LaTeX for math/formulas if needed (`Extras > Mathematical Typesetting`).
-6. **Export**:
-    * Format: **PDF** or **SVG** (Vector is best for papers).
-    * "Crop": Select "Crop" to remove excess whitespace.
-    * "Transparent Background": Check if embedding in a colored poster, otherwise uncheck for papers.
-
-## 6. Example IEEE Prompt for AI
-
-When asking the AI to generate an IEEE-style diagram, specify:
-> "Create an IEEE-standard network diagram for a campus network. Use orthogonal lines. Include Core, Distribution, and Access layers. Use standard symbols. Output in black and white compatible style."
-
-## 7. Mathematical Annotations in Network Diagrams
-
-For network diagrams requiring mathematical notation (throughput, latency, bandwidth calculations):
-
-### 7.1 Common Network Formulas
-
-Use LaTeX format for all mathematical expressions:
-
-```latex
-% Throughput
-$$T = \frac{W}{RTT} \cdot \frac{1}{\sqrt{p}}$$
-
-% Queuing delay (M/M/1)
-$$D_q = \frac{\lambda}{\mu(\mu - \lambda)}$$
-
-% Shannon capacity
-$$C = B \log_2\left(1 + \frac{S}{N}\right)$$
-
-% Link utilization
-$$\rho = \frac{\lambda}{\mu}$$
-
-% Packet loss probability
-$$P_{loss} = \frac{\rho^K (1-\rho)}{1-\rho^{K+1}}$$
-```
-
-### 7.2 Label Examples for Network Components
-
-| Component | Label Example |
-|-----------|---------------|
-| Link bandwidth | `\(B = 1\text{ Gbps}\)` |
-| Propagation delay | `\(d_{prop} = \frac{L}{c}\)` |
-| Buffer size | `\(Q_{max} = B \cdot RTT\)` |
-| Traffic rate | `\(\lambda = 500\text{ pps}\)` |
-| Service rate | `\(\mu = 1000\text{ pps}\)` |
-
-### 7.3 Example: Network Performance Diagram
-
-```
-┌─────────────────┐     \(d = 10\text{ ms}\)     ┌─────────────────┐
-│    Server       │ ──────────────────────────▶ │    Client       │
-│ \(\mu = 1000\)  │     \(B = 100\text{ Mbps}\)  │ \(\lambda=500\) │
-└─────────────────┘                              └─────────────────┘
-                    $$T = \frac{MSS \cdot 8}{RTT \cdot \sqrt{p}}$$
-```
-
-### 7.4 Notation Consistency
-
-For IEEE network papers, maintain consistent notation:
-
-| Symbol | Meaning | LaTeX |
-|--------|---------|-------|
-| λ | Arrival rate | `\lambda` |
-| μ | Service rate | `\mu` |
-| ρ | Utilization | `\rho` |
-| B | Bandwidth | `B` |
-| D | Delay | `D` |
-| C | Capacity | `C` |
-| T | Throughput | `T` |
-| RTT | Round-trip time | `\text{RTT}` |
-| MSS | Max segment size | `\text{MSS}` |
-
-## 8. Export Checklist for IEEE Submission
-
-* [ ] All mathematical expressions use LaTeX format
-* [ ] Figure has proper caption: "Fig. N. Description."
-* [ ] Legend explains all symbols and line types
-* [ ] Grayscale compatibility verified
-* [ ] Vector format (PDF/SVG) used
-* [ ] Excess whitespace cropped
-* [ ] Font size matches journal requirements (8-10pt)
-* [ ] Resolution ≥ 300 DPI for any rasterized elements

package/template/.agents/skills/drawio/references/docs/live-snapshots-drift.md

@@ -1,72 +0,0 @@
-# Saved Live Snapshots And Projection Drift
-
-Use this route only for JSON snapshots the user explicitly selected. It parses
-Terraform state/plan JSON, Docker inspect JSON, or Kubernetes live JSON in
-memory; it does not run `terraform`, `docker`, or `kubectl`, contact a daemon,
-cluster, or cloud, inherit provider credentials, or copy raw snapshots into a
-work directory.
-
-## Snapshot Adapters
-
-```js
-import {
-  parseTerraformStateSnapshot,
-  parseDockerInspectSnapshot,
-  parseKubernetesLiveSnapshot
-} from './scripts/adapters/index.js'
-```
-
-- Terraform keeps exact resource instance addresses. Declared unexpanded
-  addresses and `count`/`for_each` instances remain different identities.
-- Docker accepts only containers with explicit Compose project/service labels,
-  aggregates replicas by project/service, and excludes standalone containers.
-  Container ID/name, env, unrelated labels, credentials, and mount paths are
-  never projected. Network/volume parity remains `missing evidence`.
-- Kubernetes live JSON calls the same parser core, identity builder, kind-scope
-  rules, and attribute allowlist as declared manifests. Secret payloads,
-  annotations, managed fields, and literal env values are excluded.
-
-Each adapter returns finalized `CanonicalGraphProjection v1` with `mode: live`.
-Declared and live projections use the same identity factory and allowlist;
-display label, capture time, input order, container instance name, and pod UID
-do not identify an entity.
-
-## Compare And Render
-
-```js
-import { compareGraphProjections, renderDriftGraph } from './scripts/adapters/index.js'
-
-const report = compareGraphProjections(declared, live, {
-  baselineContext: 'shop-production',
-  observedContext: 'shop-production'
-})
-const { spec, xml } = await renderDriftGraph(report, declared, live)
-```
-
-Both logical contexts are mandatory and must match. Projection/report version,
-domain, context, or identity ambiguity fails explicitly with
-`DRIFT_INCOMPATIBLE` or `IDENTITY_COLLISION`; the comparator never falls back
-to label matching.
-
-The versioned report has node, edge, and allowlisted attribute
-`added/removed/changed/unchanged` buckets in deterministic identity order.
-Label changes are display changes on the same identity. Edge relation is part
-of identity, so a relation change is removed plus added.
-
-Drift canonical YAML carries status text and changed key names, includes a
-legend, and renders removed edges as dashed. Color is not the only status
-carrier. The runtime always uses the shared projector, `validateSpec`, vendored
-JavaScript ELK, canonical renderer, and `validateXml`; adapters never lay out
-nodes or write XML.
-
-## Evidence Boundary
-
-The recorded Compose case is under `references/examples/importers/live/`:
-
-- `compose-drift-report.json` is the sanitized machine report;
-- `compose-drift.spec.yaml` is the editable canonical source;
-- `compose-drift-evidence.json` records stable renderer IDs and stopping rules.
-
-The fixture proves deterministic parsing, comparison, JS ELK, rendering, and
-XML validation. It does not prove provider CLI capture, a real environment,
-Desktop preview, or visual-model review; those remain `missing evidence`.

package/template/.agents/skills/drawio/references/docs/math-typesetting.md

@@ -1,258 +0,0 @@
-# Math Typesetting (LaTeX / AsciiMath) in Draw.io
-
-Use this file as the syntax source of truth for formulas in draw.io labels and text shapes.
-
-> Source: <https://www.drawio.com/doc/faq/math-typesetting>
-> Related design guidance: `design-system/formulas.md`
-
-## Official Delimiter Rules
-
-Draw.io uses MathJax for mathematical typesetting. Math is recognized only when wrapped in explicit delimiters.
-
-| Syntax | Use It For | Example |
-|--------|------------|---------|
-| `$$...$$` | Standalone equations, dedicated formula nodes, labels that are entirely math | `$$E = mc^2$$` |
-| `\(...\)` | Inline math inside a longer sentence or label | `Output: \(y = f(x)\)` |
-| `` `...` `` | Simple AsciiMath expressions | `` `sum_(i=1)^n x_i` `` |
-
-### Rule of Thumb
-
-- Use `$$...$$` only when the label is essentially a formula block.
-- Use `\(...\)` for sentence-level inline math.
-- Use `` `...` `` only for simple AsciiMath or when the user explicitly asks for AsciiMath.
-
-### Minimal Examples
-
-```yaml
-nodes:
-  - id: inline
-    label: "Linear: \\(y = mx + b\\)"
-
-  - id: block
-    label: "$$\\sum_{i=1}^{n} x_i$$"
-    type: formula
-
-  - id: asciimath
-    label: "`sum_(i=1)^n x_i`"
-```
-
-## Mixed Labels
-
-Inline LaTeX and AsciiMath may appear in the same label. Keep each formula wrapped independently.
-
-```yaml
-nodes:
-  - id: mixed
-    label: "Score: \\(x_i^2\\), baseline: `x^2`"
-```
-
-Prefer `\(...\)` over `` `...` `` when the expression already uses LaTeX commands or must match the paper notation.
-
-## Unsupported or Discouraged Output
-
-Do not generate these forms in final skill output:
-
-```text
-\frac{a}{b}          # bare LaTeX
-\alpha + \beta       # bare LaTeX
-$x^2 + y^2$          # discouraged single-dollar inline syntax
-\[x^2 + y^2\]        # discouraged bracket block syntax
-```
-
-Use these instead:
-
-```text
-\(\frac{a}{b}\)
-\(\alpha + \beta\)
-\(x^2 + y^2\)
-$$x^2 + y^2$$
-```
-
-Notes:
-
-- The skill should not emit bare LaTeX, `$...$`, or `\[...\]`.
-- The current math helpers normalize `$...$` to `\(...\)` and `\[...\]` to `$$...$$` when encountered in imported text, but final emitted labels should still use the three supported syntaxes only.
-
-## Required UI Setting
-
-Enable this in draw.io before previewing or editing formulas:
-
-`Extras > Mathematical Typesetting`
-
-When the toggle is enabled, draw.io renders the label through MathJax. Disable it temporarily to inspect the raw source.
-
-## YAML Writing Rules
-
-In YAML, escape backslashes inside quoted strings:
-
-```yaml
-nodes:
-  - id: input
-    label: "Input: \\(X \\in \\mathbb{R}^{n \\times d}\\)"
-
-  - id: loss
-    label: "$$\\mathcal{L} = -\\sum_{i=1}^{n} y_i \\log(\\hat{y}_i)$$"
-    type: formula
-```
-
-Checklist:
-
-- Write `\\(` and `\\)` in YAML strings when using inline LaTeX.
-- Keep `$$` unchanged in YAML strings.
-- Escape every LaTeX backslash inside the YAML string.
-
-## XML `value` Attribute Rules
-
-Equations are stored in the `value` attribute of `mxCell` nodes and edge labels.
-
-Keep the math delimiters inside `value="..."` and escape XML attribute characters:
-
-- `&` -> `&amp;`
-- `<` -> `&lt;`
-- `>` -> `&gt;`
-- `"` -> `&quot;`
-
-Avoid hidden HTML formatting tags in math labels.
-
-### Inline XML Example
-
-```xml
-<mxCell id="2" value="Model: \(y = Wx + b\)" style="rounded=1;html=1;fillColor=#dae8fc;strokeColor=#6c8ebf" vertex="1" parent="1">
-  <mxGeometry x="120" y="120" width="220" height="70" as="geometry"/>
-</mxCell>
-```
-
-### Block XML Example
-
-```xml
-<mxCell id="3" value="$$\mathcal{L} = \sum_i (y_i-\hat y_i)^2$$" style="rounded=1;html=1;fillColor=#d5e8d4;strokeColor=#82b366" vertex="1" parent="1">
-  <mxGeometry x="120" y="220" width="320" height="90" as="geometry"/>
-</mxCell>
-```
-
-## Export Rules
-
-- Default math output is SVG-based.
-- For exported PDF files where selectable math matters, use `math-output=html`.
-- Vector export remains preferred for academic figures: PDF or SVG.
-
-## Validation Checklist
-
-Before finalizing any diagram with formulas, verify:
-
-- Every formula uses one of: `$$...$$`, `\(...\)`, or `` `...` ``.
-- `$$...$$` is reserved for standalone equations, not sentence-level inline text.
-- No bare backslash commands remain in labels.
-- No `$...$` or `\[...\]` remains in emitted YAML/XML.
-- `Extras > Mathematical Typesetting` is enabled before visual inspection.
-- YAML strings escape LaTeX backslashes correctly.
-- XML `value` content is HTML-free and XML-escaped.
-
-## Troubleshooting
-
-### Math does not render
-
-1. Enable `Extras > Mathematical Typesetting`.
-2. Check that the label uses `$$...$$`, `\(...\)`, or `` `...` ``.
-3. Click `</>` in the toolbar and remove hidden HTML tags if present.
-
-### Formula overflows or gets clipped
-
-- Increase the shape size.
-- In the Style panel, under Property -> Text Overflow, try `Block` or `Width`.
-- Prefer a dedicated formula node instead of forcing long block math into a small process node.
-
-### Exported PDF has extra whitespace
-
-- Keep formula nodes away from the diagram boundary.
-- Use left or right alignment for formula labels when appropriate.
-
-## LaTeX Quick Reference
-
-### Greek Letters
-
-| Symbol | Code | Symbol | Code |
-|--------|------|--------|------|
-| α | `\alpha` | Γ | `\Gamma` |
-| β | `\beta` | Δ | `\Delta` |
-| γ | `\gamma` | Θ | `\Theta` |
-| λ | `\lambda` | Σ | `\Sigma` |
-| μ | `\mu` | Φ | `\Phi` |
-| π | `\pi` | Ω | `\Omega` |
-
-### Common Operators
-
-```latex
-\times
-\div
-\pm
-\leq
-\geq
-\neq
-\approx
-\infty
-\partial
-\nabla
-\rightarrow
-\Rightarrow
-```
-
-### Fractions, Powers, Sums
-
-```latex
-\frac{a}{b}
-\sqrt{x}
-\sqrt[n]{x}
-x^{n}
-x_{i}
-\sum_{i=1}^{n} x_i
-\prod_{i=1}^{n} x_i
-\int_{a}^{b} f(x) \, dx
-\lim_{x \to \infty} f(x)
-```
-
-### Matrices and Sets
-
-```latex
-\begin{bmatrix} a & b \\ c & d \end{bmatrix}
-\begin{pmatrix} x \\ y \\ z \end{pmatrix}
-\mathbb{R}
-\mathbb{N}
-\mathbb{Z}
-\mathbb{C}
-\mathcal{L}
-```
-
-## Academic Diagram Notes
-
-When formulas appear in paper figures:
-
-- Keep equation labels concise.
-- Use the same notation as the surrounding manuscript.
-- Define symbols in the caption or legend when needed.
-- For grayscale print, do not rely on color alone to distinguish formula-bearing nodes.
-
-### Example Academic Labels
-
-```yaml
-nodes:
-  - id: input
-    label: "Input: \\(x \\in \\mathbb{R}^d\\)"
-    type: terminal
-
-  - id: model
-    label: "Model: \\(y = Wx + b\\)"
-    type: process
-
-  - id: loss
-    label: "$$\\mathcal{L} = -\\sum_{i=1}^{n} y_i \\log(\\hat{y}_i)$$"
-    type: formula
-```
-
-### Recommended Export Workflow
-
-1. Draft the diagram with official formula delimiters only.
-2. Enable `Extras > Mathematical Typesetting`.
-3. Verify inline labels use `\(...\)` and standalone formulas use `$$...$$`.
-4. Export as PDF or SVG.
-5. For selectable math in PDF, use `math-output=html`.

package/template/.agents/skills/drawio/references/docs/mcp-tools.md

@@ -1,85 +0,0 @@
-# Live Backend Reference
-
-This document describes the optional live backends that can enhance the draw.io skill. The core skill is still offline-first: YAML spec -> CLI -> `.drawio` + sidecars.
-
-The important rule is: **reason in capabilities first, provider names second**. Workflows should ask "which live capabilities exist?" before asking "which provider is installed?".
-
-## Capability Vocabulary
-
-| Capability              | Meaning                                                  | Typical Use                                             |
-| ----------------------- | -------------------------------------------------------- | ------------------------------------------------------- |
-| `replace_diagram_xml`   | Open or replace a diagram using full XML                 | Browser preview, live session bootstrap, inline preview |
-| `read_diagram_xml`      | Read the latest XML back from the live editor            | Preserve manual edits before further changes            |
-| `patch_diagram_cells`   | Apply cell-level add/update/delete operations            | Incremental live editing                                |
-| `export_diagram_file`   | Export `.drawio`, `.png`, or `.svg` from the live editor | Live-session file export                                |
-| `search_shape_catalog`  | Search an official shape/style index                     | Stencil-heavy cloud/network/P&ID diagrams               |
-| `render_inline_preview` | Render the diagram inline in chat                        | Visual review without opening a new tab                 |
-
-## Provider Matrix
-
-| Provider                             | `replace_diagram_xml` | `read_diagram_xml` | `patch_diagram_cells` | `export_diagram_file` | `search_shape_catalog` | `render_inline_preview` |
-| ------------------------------------ | --------------------- | ------------------ | --------------------- | --------------------- | ---------------------- | ----------------------- |
-| Current next-ai live MCP             | Yes                   | Yes                | Yes                   | Yes                   | No                     | No                      |
-| Official Tool Server (`@drawio/mcp`) | Yes                   | No                 | No                    | No                    | No                     | No                      |
-| Official App Server (`mcp.draw.io`)  | Yes                   | No                 | No                    | No                    | Yes                    | Yes                     |
-
-## Current Routing Rules
-
-1. Default to offline-first, even when a live provider is available.
-2. Use a live backend only when the user explicitly wants browser or inline refinement, or when a workflow requires a capability that is only available live.
-3. **Incremental live edit requires both `read_diagram_xml` and `patch_diagram_cells`.**
-   - If either capability is missing, do not attempt live edit.
-   - Instead: import `.drawio` -> generate `.spec.yaml` / `.arch.json` -> edit the YAML spec -> regenerate outputs.
-4. `search_shape_catalog` is optional guidance, not a hard dependency.
-   - If present, use it for AWS/Azure/GCP/Cisco/Kubernetes/P&ID/electrical diagrams when exact shapes matter.
-   - If absent, fall back to known design-system icons or semantic shapes instead of blocking the task.
-5. `render_inline_preview` improves review loops but does not replace edit-session capabilities.
-
-## Provisioning (`.mcp.json`)
-
-The base skill ships a tracked `.mcp.json` that provisions the "Current next-ai live MCP" provider:
-
-- It pins `@next-ai-drawio/mcp-server@0.4.13` and launches it through `npx --yes`, which fetches the package over the network at startup (version-pinned, but with no additional integrity hash).
-- Only MCP-aware hosts that treat the skill directory as a project root read this file. The offline authoring path (YAML -> CLI -> `.drawio` + sidecars) never reads it, and no workflow requires it.
-- To upgrade the provider, bump the pinned version in `.mcp.json`. To opt out, delete the file — every offline workflow keeps working without it.
-
-## Provider-Specific Tool Mapping
-
-### Current next-ai live MCP
-
-| Tool                 | Capability             |
-| -------------------- | ---------------------- |
-| `start_session`      | Session bootstrap only |
-| `create_new_diagram` | `replace_diagram_xml`  |
-| `get_diagram`        | `read_diagram_xml`     |
-| `edit_diagram`       | `patch_diagram_cells`  |
-| `export_diagram`     | `export_diagram_file`  |
-
-### Official Tool Server
-
-| Tool                  | Capability                                        |
-| --------------------- | ------------------------------------------------- |
-| `open_drawio_xml`     | `replace_diagram_xml`                             |
-| `open_drawio_csv`     | Convenience import path, not a live-edit contract |
-| `open_drawio_mermaid` | Convenience import path, not a live-edit contract |
-
-### Official App Server
-
-| Tool             | Capability                                     |
-| ---------------- | ---------------------------------------------- |
-| `create_diagram` | `replace_diagram_xml`, `render_inline_preview` |
-| `search_shapes`  | `search_shape_catalog`                         |
-
-## Practical Guidance
-
-- If the user says "open this in draw.io" or wants a quick browser handoff, `replace_diagram_xml` is enough.
-- If the user wants "change the current live diagram without losing my manual edits", require `read_diagram_xml + patch_diagram_cells`.
-- If the user wants cloud or network icons, check whether `search_shape_catalog` exists. If not, continue with documented icon mappings and semantic fallbacks.
-- If the user wants review inside the chat instead of a new tab, `render_inline_preview` helps, but it is still only a preview path.
-
-## Related
-
-- [Migration Readiness](migration-readiness.md)
-- [Stencil Library Guide](stencil-library-guide.md)
-- [Official XML Reference Mirror](../official/xml-reference.md)
-- [Official Style Reference Mirror](../official/style-reference.md)