FlowForge

Installation
SKILL.md

FlowForge — Draw.io Diagram Skill

Generate professional diagrams as draw.io XML files. Supports flowcharts, architecture diagrams, comparisons, mind maps, timelines, and more.

Usage

/FlowForge "OAuth 2.0 authorization code flow"
/FlowForge path/to/design-doc.md
/FlowForge "CI/CD pipeline" --type loop
/FlowForge "Monolith vs Microservices" --type compare --theme morandi
/FlowForge "System Architecture" --lang en

Workflow

Step 1: Understand Requirements

If given a document:

  1. Read the full document
  2. Identify 1–3 places that benefit from a diagram
  3. For each, list:
    • Insert position (after which section)
    • Diagram topic (one sentence)
    • Recommended diagram type (see Diagram Type Reference below)
  4. Wait for user confirmation

If given a concept directly:

  1. Confirm the core message to convey
  2. Choose diagram type
  3. Proceed to Step 2

Step 2: Confirm Theme

If the user hasn't specified --theme, state:

Default theme: tech-blue (科技蓝灰). Other options: morandi, mint, terracotta, indigo. Want a different one?

Proceed after user confirms. For full theme definitions, see themes.md.

Step 3: Design Sketch (ASCII)

Produce an ASCII sketch showing:

  • All nodes with labels
  • Arrow directions and labels
  • Grouping/regions
  • Color role assignments (e.g., [primary], [accent], [neutral])
  • Approximate dimensions and layout direction

Example sketch:

Direction: left-to-right | Nodes: 4 | Type: flow

[primary] Code Commit  →  [process] Build  →  [process] Test  →  [accent] Deploy
                                                    ↓ (fail)
                                              [error] Alert

STOP — Wait for user confirmation or revision before proceeding to XML generation.

Step 4: Generate draw.io XML

  1. Select layout algorithm for the chosen type (see Layout System below)
  2. Compute all coordinates using the formulas and constants
  3. Apply theme colors from themes.md
  4. Use element templates from xml-reference.md
  5. Reference complete examples in examples.md for pattern matching
  6. Fix any issues found

Step 5: Save & Deliver

Save the .drawio file. Default path:

./diagrams/{diagram-name}.drawio

Or user-specified path. After saving, report:

Iteration

If the user wants changes after viewing:

  1. Read the current .drawio file
  2. Modify the XML per feedback
  3. Save as {name}_v2.drawio (preserve previous version)

Layout System

Global Constants

CANVAS_PAD    = 40       # padding from content to canvas edge
NODE_W        = 160      # default node width
NODE_H        = 50       # single-label node height
NODE_H_RICH   = 90       # node with title + bullet list
TITLE_H       = 28       # diagram title height
GAP_H         = 60       # horizontal gap between nodes
GAP_V         = 50       # vertical gap between nodes
DIAMOND_W     = 120      # decision diamond width
DIAMOND_H     = 80       # decision diamond height
GROUP_PAD     = 20       # padding inside dashed group boxes
FONT_TITLE    = 18       # diagram title font size
FONT_NODE     = 13       # node title font size
FONT_BODY     = 10       # node body/list font size
FONT_ARROW    = 9        # arrow label font size
STROKE_NODE   = 1.2      # node border stroke width
STROKE_ARROW  = 0.8      # arrow stroke width
ARC_SIZE      = 10       # rounded rect arc size
ARROW_SIZE    = 5        # arrowhead size

Adjust constants when needed:

  • If nodes > 6 in a flow, reduce NODE_W to 140 or GAP_H to 40
  • If labels are long, increase NODE_W up to 200
  • Rich nodes (title + list): use NODE_H_RICH instead of NODE_H

Layout Algorithms by Type

flow — Linear Flow (left-to-right)

x[i] = CANVAS_PAD + i × (NODE_W + GAP_H)
y    = CANVAS_PAD + TITLE_H + GAP_V

Canvas width  = CANVAS_PAD × 2 + n × NODE_W + (n-1) × GAP_H
Canvas height = CANVAS_PAD × 2 + TITLE_H + GAP_V + NODE_H

If n > 5, use snake layout — wrap to next row flowing right-to-left:

Row 0 (L→R): nodes 0..4
Row 1 (R→L): nodes 5..9
Row 2 (L→R): nodes 10..14

y[row] = CANVAS_PAD + TITLE_H + GAP_V + row × (NODE_H + GAP_V)
x[i] in even row = CANVAS_PAD + col × (NODE_W + GAP_H)
x[i] in odd row  = CANVAS_PAD + (4-col) × (NODE_W + GAP_H)

flow-vertical — Linear Flow (top-to-bottom)

x    = CANVAS_PAD + (Canvas width / 2) - NODE_W / 2
y[i] = CANVAS_PAD + TITLE_H + GAP_V + i × (NODE_H + GAP_V)

compare — Left-Right Comparison

VS_W = 50   # width of "VS" badge

Left column x  = CANVAS_PAD
Right column x = CANVAS_PAD + NODE_W + GAP_H + VS_W + GAP_H
VS badge x     = CANVAS_PAD + NODE_W + GAP_H

Header y = CANVAS_PAD + TITLE_H + GAP_V
Row y[j] = Header y + NODE_H + GAP_V/2 + j × (NODE_H + GAP_V/2)

Each side has a header node (title) and content rows below.

layers — Layer Stack

LAYER_W = 400

x = CANVAS_PAD + (canvas center offset)
y[i] = CANVAS_PAD + TITLE_H + GAP_V + i × (NODE_H_RICH + GAP_V/3)

Canvas width  = CANVAS_PAD × 2 + LAYER_W
Canvas height = CANVAS_PAD × 2 + TITLE_H + GAP_V + n × NODE_H_RICH + (n-1) × GAP_V/3

Layers span full width. Top layer = highest abstraction.

loop — Cycle / Feedback Loop

Arrange N nodes in a rectangular path (clockwise):

For 4 nodes (most common):

Positions:
  [0] top-left:     (CANVAS_PAD, CANVAS_PAD + TITLE_H + GAP_V)
  [1] top-right:    (CANVAS_PAD + NODE_W + GAP_H × 2, same y)
  [2] bottom-right: (same x as [1], y + NODE_H + GAP_V)
  [3] bottom-left:  (same x as [0], same y as [2])

Arrows: 0→1 (top), 1→2 (right), 2→3 (bottom, R→L), 3→0 (left, bottom→top)

For N nodes: distribute evenly across 4 sides of a rectangle.

tree — Tree / Hierarchy

Level spacing = GAP_V × 1.5
Sibling spacing = NODE_W + GAP_H / 2

Root: centered at top
Children: evenly distributed below parent, centered under parent

Subtree width = max(sum of children subtree widths, NODE_W)
Parent x = leftmost child x + (rightmost child x + NODE_W - leftmost child x) / 2 - NODE_W / 2

hub — Hub and Spoke (Center Radiate)

Center node at canvas center. Spokes at equal angles.

Lookup table (use these positions instead of computing cosines):

N spokes Positions relative to center (dx, dy)
3 (0, -R), (R×0.87, R×0.5), (-R×0.87, R×0.5)
4 (0, -R), (R, 0), (0, R), (-R, 0)
5 (0, -R), (R×0.95, -R×0.31), (R×0.59, R×0.81), (-R×0.59, R×0.81), (-R×0.95, -R×0.31)
6 (0, -R), (R×0.87, -R×0.5), (R×0.87, R×0.5), (0, R), (-R×0.87, R×0.5), (-R×0.87, -R×0.5)

Where R = 180 (radius). Spoke node position:

x[i] = cx + dx[i] - NODE_W / 2
y[i] = cy + dy[i] - NODE_H / 2

columns — Parallel Columns

Column x[i] = CANVAS_PAD + i × (NODE_W + GAP_H)
Header y     = CANVAS_PAD + TITLE_H + GAP_V
Item y[j]    = Header y + NODE_H + GAP_V/2 + j × (NODE_H + GAP_V/3)

Each column has a header node (colored) and vertically stacked item nodes below.

matrix — Comparison Matrix

CELL_W = 140
CELL_H = 45
HEADER_H = 40

Col header x[c] = CANVAS_PAD + HEADER_W + GAP_H/3 + c × (CELL_W + GAP_H/3)
Row header y[r] = CANVAS_PAD + TITLE_H + HEADER_H + GAP_V/3 + r × (CELL_H + GAP_V/3)
Cell (r,c): x = Col header x[c], y = Row header y[r]

funnel — Funnel

MAX_W = 360
MIN_W = 120

w[i] = MAX_W - i × ((MAX_W - MIN_W) / (n - 1))
x[i] = CANVAS_PAD + (MAX_W - w[i]) / 2
y[i] = CANVAS_PAD + TITLE_H + GAP_V + i × (NODE_H + GAP_V / 3)

Centered, decreasing-width bands stacked vertically.

timeline — Timeline

LINE_Y = canvas vertical center
Node spacing = NODE_W + GAP_H

Event nodes alternate above and below the timeline:
  Above: y = LINE_Y - NODE_H - GAP_V/2
  Below: y = LINE_Y + GAP_V/2
  x[i]  = CANVAS_PAD + i × (NODE_W + GAP_H)

Draw a horizontal line at LINE_Y spanning full width.
Draw vertical ticks from line to each node.

sequence — Sequence Diagram

LIFELINE_GAP = 180
MESSAGE_GAP  = 50
PARTICIPANT_Y = CANVAS_PAD + TITLE_H + GAP_V
PARTICIPANT_W = 140
PARTICIPANT_H = 40

Participant x[i] = CANVAS_PAD + i × LIFELINE_GAP
Lifeline: vertical dashed line from bottom of participant box downward

Message y[j] = PARTICIPANT_Y + PARTICIPANT_H + GAP_V + j × MESSAGE_GAP
Messages: horizontal arrows between lifelines

Diagram Type Reference

Core Types

Type Code Best For Example
Linear Flow flow Sequential steps A→B→C API call flow, data pipeline
Vertical Flow flow-vertical Top-down processes Decision process, approval chain
Comparison compare A vs B side by side Traditional vs AI approach
Layer Stack layers Tech stack, tiers System architecture layers
Cycle loop Iterative processes ML training loop, CI/CD
Tree tree Hierarchies, taxonomies Org chart, decision tree
Hub & Spoke hub Core concept + branches Product feature map

Extended Types

Type Code Best For
Parallel Columns columns 3+ parallel concepts
Matrix matrix Multi-dimension comparison
Funnel funnel Filtering, conversion
Timeline timeline Version evolution, history
Sequence sequence Component interactions

Selection Guide

Content Pattern Recommended Type
Sequential steps flow or flow-vertical
Two things compared compare
3+ parallel concepts columns or hub
Layered system layers
Iterative/cyclical process loop
One core, many branches hub or tree
Components communicating sequence
Change over time timeline
Multi-criteria evaluation matrix
Progressive filtering funnel

Quality Checks

Before delivering the .drawio file, verify ALL of the following. Fix any failures before saving.

Layout Checks

  1. No overlapping nodes — For every pair of nodes, their bounding boxes must not intersect. Verify: |x1 - x2| >= width OR |y1 - y2| >= height.
  2. Canvas large enough — No node extends beyond: max(x + w) + CANVAS_PAD <= CANVAS_W and max(y + h) + CANVAS_PAD <= CANVAS_H.
  3. Consistent spacing — Gaps between adjacent nodes should match GAP_H / GAP_V (±10px tolerance).
  4. Alignment — Nodes in the same row share the same y. Nodes in the same column share the same x.

Text Checks

  1. Text fits in node — Estimate: Chinese characters ≈ 14px each at font 13; English characters ≈ 8px each. If text width > node width - 20px padding, either widen the node, reduce font, or add <br> line breaks.
  2. Minimum readability — No font size below 9px. No node smaller than 80×30.

XML Checks

  1. All arrow targets exist — Every source="X" and target="Y" must match an id="X" / id="Y" defined as a vertex.
  2. Unique IDs — Every id attribute is unique across the entire XML.
  3. Correct parent — All content elements have parent="1" unless intentionally grouped.
  4. Valid XML — Well-formed, all attributes quoted, HTML entities escaped (&lt; &gt; &amp; &quot;).

Style Checks

  1. Theme consistency — All fill/stroke/font colors come from the selected theme. No ad-hoc hex values.
  2. Arrow routing — Arrows use orthogonal routing only (horizontal + vertical). No diagonal arrows. Return arrows route around the outside, never crossing through unrelated nodes.

Color Usage Principles

This is the most important section for visual quality.

The 4+6 Rule

Each theme has 4 base colors and 6 extended colors:

  • Base 4: primary, process, accent, neutral
  • Extended 6: success, warning, error, secondary, storage, group

Blue (primary + process) should always be the dominant color family, but the amount of non-blue accents should scale with diagram size.

Color Budget by Diagram Size

Node Count Target Colors Guideline
3-5 nodes 2-3 colors Blue dominates. At most 1 accent node.
6-8 nodes 3-4 colors Blue ~60%. Add 1-2 non-blue at semantic turning points.
9+ nodes 4-5 colors Blue ~50%. Distribute 2-3 non-blue for visual rhythm.

6+ colors almost always looks bad regardless of diagram size.

Color Assignment Strategy

  1. Blue dominatesprimary and process should cover the largest visual area and the biggest nodes.

  2. Accent is a scalpel, not a paintbrush — Use accent for at most 1-2 key highlights. Never on the largest node.

  3. Scale with length — A 4-node flow with 2 colors looks clean. A 9-node flow with only 2 colors is a monotonous wall. Add non-blue colors at natural semantic boundaries to create rhythm.

  4. 3-consecutive guideline — Avoid more than 3 adjacent nodes in the same color. If you spot 4+ consecutive blue nodes, promote the most semantically distinctive one to a non-blue color.

  5. Distribute spatially — Non-blue nodes should be spread across the diagram (top/middle/bottom), not clustered in one region.

Semantic Color Heuristics

When deciding which nodes get non-blue colors, use this table:

Node Type Recommended Color
User input / starting point primary
Standard processing step process
Decision / branching point warning or accent
Domain-specific / translation step secondary
Key transformation / highlight accent
Data source / external system neutral
Error path / fallback accent or error
Successful output / endpoint success or primary
Database / storage storage

Examples

Bad (rainbow — every node a different color):

[neutral] 数据源 → [process] 采集 → [warning] 检查 → [success] 清洗 → [error] 告警 → [storage] 入库

Bad (wall of blue — 9 nodes all the same):

[primary] 提问 → [process] 校验 → [process] 评估 → [process] 翻译 → [process] 执行 → [process] 解读 → [process] 可视化 → [process] 标注 → [primary] 返回

Good (short flow, 2-3 colors):

[neutral] 数据源 → [process] 采集 → [process] 检查 → [process] 清洗 → [primary] 入库
                                        ↓ (异常)
                                  [accent] 告警处理

Good (long flow, 4-5 colors with rhythm):

[primary] 提问 → [process] 校验 → [warning] 评估 → [secondary] 翻译 → [process] 执行 → [process] 解读 → [accent] 可视化 → [process] 标注 → [success] 返回

Design Principles

  1. Narrative titles — Use "How OAuth 2.0 Authorizes a User" not "OAuth 2.0 Diagram".
  2. Bilingual labels — Use the user's language. Technical abbreviations (API, LLM, CI/CD) stay in English regardless.
  3. Orthogonal arrows only — All arrows run horizontally or vertically, no diagonals. Always include edgeStyle=orthogonalEdgeStyle; in every arrow's style attribute.

Gotchas — Common Failure Points

These are the most common mistakes when generating draw.io XML. Check for these first when debugging.

  1. Rich node text overflow — The <b>Title</b> inside a rich node uses a separate text cell. If you put the title in the value attribute of the background rect AND in a text cell, the title renders twice. Use value="" on the background rect.

  2. Arrow ignores exit/entry points — If you omit exitX/exitY/entryX/entryY from the arrow style, draw.io auto-routes, which often looks wrong. Always specify explicit exit/entry points.

  3. HTML entities in value attributesvalue attributes use HTML, so < must be &lt;, > must be &gt;, & must be &amp;. Missing escapes break the entire XML. The most common miss: &bull; in bulleted lists — use &amp;bull; when inside an attribute.

  4. Dashed group box covers its children — If a dashed group box is defined AFTER its child nodes in the XML, it renders on top and hides them. Always define group boxes BEFORE their child elements.

  5. Canvas too smallpageWidth/pageHeight that are too small cause nodes to be cut off in the exported image. Always compute from layout formulas: max(x + w) + CANVAS_PAD.

  6. Forgetting parent="1" — Every content element must have parent="1". Missing this silently breaks the node.

  7. Diamond text overflow — Diamond/rhombus shapes have less usable area than rectangles. Keep labels to 2-3 short words. Use font size 11 instead of 13.

  8. Snake layout wrong arrow direction — In snake layouts (flow with >5 nodes), the wrap-around arrow must exit from the bottom of the last node in row N and enter the top of the first node in row N+1, not go sideways.

  9. Cycle arrow goes through center — In loop diagrams, the return arrow (last→first) must route along the outside of the rectangle, not diagonally through the center.

  10. Coordinate calculation off-by-one — When computing n items: there are n nodes but n-1 gaps. Canvas width = PAD×2 + n×NODE_W + (n-1)×GAP_H, not n×GAP_H.

  11. Diagonal arrows on branching/merging — When a centered node branches to two side-by-side targets, exit/entry points are not aligned vertically. Without edgeStyle=orthogonalEdgeStyle; in the style, draw.io renders these as diagonal lines. Always include it in every arrow.


Initial Setup

On first invocation, check if config.json exists in the skill directory. If not, ask the user:

  1. Default theme — Which color theme to use by default? (tech-blue / morandi / mint / terracotta / indigo)
  2. Output directory — Where to save .drawio files? (default: ./diagrams/)
  3. Language — Primary language for labels? (zh / en / auto-detect)

Save responses to config.json:

{
  "defaultTheme": "tech-blue",
  "outputDir": "./diagrams",
  "language": "auto"
}

On subsequent invocations, read config.json and apply defaults. User can override per-invocation with --theme, --lang flags.


Supporting Files

The skill directory contains these files. Read them when needed — don't load everything upfront.

File When to read
themes.md When applying colors — has all 5 themes with 10 semantic colors + 7 text colors each
xml-reference.md When writing XML — canvas boilerplate, 8 element templates, arrow direction table
examples.md When generating a diagram — 3 complete XML examples with coordinate calculations

| config.json | On every invocation — user preferences (created on first run, see Initial Setup) |

Example .drawio files (in examples/)

These are real, working .drawio files you can read as reference when generating similar diagram types:

  • examples/flow-cicd.drawio — Linear flow, 4 nodes, tech-blue theme
  • examples/compare-monolith-vs-micro.drawio — Comparison, 2 columns × 3 rows, morandi theme
  • examples/loop-ml-training.drawio — Cycle, 4 nodes clockwise, mint theme
Installs
14
GitHub Stars
33
First Seen
14 days ago