Diagram Rendering Skill (Internal Monologue)#

You are an autonomous solver agent capable of transforming natural language into technical visuals. You prioritize Local Rendering for speed and privacy, using the Cloud API only as a fallback.

1. Activation Triggers#

Invoke this skill when the user asks to:

• "Draw," "map out," or "diagram" a system.
• "Visualize" an architecture or cloud infrastructure.
• "Create a flowchart" or "sequence diagram."
• "Show me an ERD" or "database schema."

2. Decision Reasoning#

• D2: Use for complex architecture, nested containers, and high-fidelity layouts.
• Mermaid: Use for standard flowcharts, sequence diagrams, and simple logic paths.
• Graphviz (DOT): Use for mathematical graphs, network topologies, and relational ERDs.

3. Template-Assisted Generation#

Before generating from scratch, check ./templates/.

• These are "Gold Standard" examples.
• You can inject data into templates using the --inject '{"key": "value"}' flag in local_render.py.

4. Persistent Execution & Self-Correction#

Your rendering attempts are stored in ./runs/<run_id>/.

1. Attempt Render: Call ./.venv/bin/python3 scripts/local_render.py <engine> "<code>".
2. Handle Failure: If the output is an error, open the metadata.json in the specific run directory.
3. Parse Stderr: Read the compiler's stderr. It will pinpoint the exact line and character of your syntax error.
4. Recursive Fix: Use the error feedback to fix your code and re-run. Do not give up until the status is "success".

5. Visual Verification (Vision Agents)#

If you have a Vision Language Model (VLM) capability:

• Inspect the generated PNG/SVG in the run folder.
• Compare the visual output against the logical intent of the prompt.
• If the layout is confusing or logically incorrect, refine the code and re-render.