ContextWeave Interactive Architecture Diagram Generator

Problems Solved

In software architecture design, quickly generating clear diagrams is crucial for team collaboration. Manual drawing is time-consuming and prone to oversights, while existing tools often require users to handle complex interfaces or scripts, adding cognitive load. The ContextWeave skill addresses this by automatically generating architecture diagrams from natural language descriptions, allowing developers to focus on design logic without unnecessary tool-related overhead.

How It Works

The skill operates based on three core invariants:

• Dereference Everything: All references (e.g., file paths, jargon) must be explicitly converted to plain text, as the cloud backend cannot access local environments. For instance, if a user mentions "reference /path/to/file", the file content must first be read and flattened into plain text within the request.
• Argument, Not Display: Diagram structures must serve logical arguments, with each relationship restatable as a clear statement, such as "Component A depends on Component B", prohibiting positional proximity as a substitute for defined relations.
• One Diagram, One Theme: Each diagram answers only one core question, avoiding clutter in a single view; complex requirements trigger multi-view handling.

The execution follows six steps:
1. Parse Materials: Identify the core problem and focus from user input, reading only relevant files and supplementing context as needed.
2. Write a One-Sentence Key Point: Clarify the theme the diagram should express, e.g., "Show component interactions in the order processing flow."
3. Determine Presentation and Colors: Use parameters like --diagram_style (topology, logic, hybrid, mindmap) and --morphology (container, flow, editorial) to set the layout.
4. Write Request File: Create .cw_skill/requests/request_.md with self-contained drawing intent.
5. Execute Script: Run the local script to interact with the cloud backend; the script saves .cw and downloads SVG/HTML outputs.
6. Reply in Fixed Format: Return a JSON object with fields such as script, input_file, status, etc.

Key parameters include:
- --diagram_style: Determines if the diagram focuses on component relationships (topology), step flows (logic), hybrid, or mindmaps.
- --morphology: Controls composition paradigm, such as regional grouping (container), flow emphasis (flow), or editorial text.
- --base_palette and --accent_targets: For setting overall color schemes and local highlights.

Boundaries and Notes

The skill has clear boundaries and usage constraints:

• Dereferencing Mandatory: Local file contents must be read and flattened into plain text; direct path references are not allowed. Jargon must be expanded with minimal information sets (role, hierarchy, action, upstream/downstream).
• Supported Diagram Types: Applicable for architecture diagrams, flowcharts, mindmaps, etc., but does not support pixel-precise layouts, hand-drawn illustrations, or complex statistical charts (e.g., intricate line graphs).
• Security and Authorization: Prior to networking, data categories and purposes must be briefly stated and agreed upon; data is only sent to the official service https://pptx.chenxitech.site, prohibiting directory traversal or API key requests.
• Multi-View Handling: When a single diagram is insufficient (e.g., needing to show global, module, and detail views), multi-view triggers and confirmations are required; unauthorized splitting is not permitted.
• Anti-Pattern Avoidance: Avoid errors like using undefined jargon as node labels or expressing relations via proximity, ensuring structure serves logical argumentation.

By adhering to these rules, the skill efficiently generates structured diagrams to help teams visualize complex systems, but its scope must be respected to prevent misuse.