Prompt To Drawio

Use this skill to run a CLI-only version of Next AI Draw.io capabilities.

Capability Set

1. 1. Prompt -> .drawio generation.
2. Prompt-driven edit on existing diagrams (ID-based operations).
3. Optional image export (png, svg, pdf, jpg).
4. Context ingestion from local files (txt/md/json/csv/pdf/image) and URLs.
5. Shape library documentation injection (aws4, gcp2, azure2, kubernetes, etc.).
6. Visual quality validation with retry loops.
7. History snapshot backup before edits.

Runtime Script

Primary entrypoint:

• - INLINECODE10

Subcommands:

• - INLINECODE11
• INLINECODE12
• INLINECODE13
• INLINECODE14
• INLINECODE15

Compatibility:

• - Legacy usage without subcommand is treated as generate.

Runtime Modes

Mode A (default, recommended): in-session LLM mode

• - If the assistant is directly executing this skill inside a Codex/LLM session, generation/edit can be done by the session model itself.
• In this mode, DRAWIO_LLM_API_KEY is not required by the skill workflow.

Mode B: standalone CLI mode

• - If you run prompt_to_drawio.py generate|edit|validate as an independent subprocess that must call a model API by itself, then API key is required.
• Supported env key names: DRAWIO_LLM_API_KEY (preferred) or OPENAI_API_KEY.
• Default key-loading rule: script auto-loads nearest .env from current working directory upward (first match), then reads key variables from it.
• Override dotenv path with DRAWIO_DOTENV_FILE=/absolute/path/.env.
• Precedence: CLI flag --api-key/--validation-api-key > existing process env > auto-loaded .env.
• New runtime controls:

Optional env overrides for standalone CLI mode:
CODEBLOCK0

Google OpenAI-Compatible Example

CODEBLOCK1

Model naming notes:

• - Recommended baseline: gemini-3-pro-preview (or latest model shown by provider model listing).
• Avoid stale/nonexistent names such as gemini-3-pro unless your provider explicitly lists it.

Workflows

1) Create Diagram

CODEBLOCK2

2) Create Diagram from Image/PDF/URL Context

CODEBLOCK3

3) Edit Existing Diagram

CODEBLOCK4

4) Export Only

CODEBLOCK5

5) Validation Only

CODEBLOCK6

6) Shape Library Discovery

CODEBLOCK7

Output Contract

Always surface generated file paths exactly as printed by script:

• - INLINECODE33
• INLINECODE34 (if requested and successful)
• INLINECODE35 (for edits)
• INLINECODE36 block when validation runs

Diagnostics:

• - On startup, script prints effective runtime config summary to stderr (dotenv source, effective model/base-url, key presence mask).
• On JSON parse failures, raw model response is dumped to a temp file and the error includes that path.

Do not claim export success unless target file exists.