Document Intelligence Skill

Purpose

Convert unstructured documents into LLM-ready Markdown using MarkItDown, with deterministic defaults and governance-compliant wrappers.

References:

• •core/GLOBAL_RULES.md
• •core/LOGGING_STANDARD.md
• •core/TESTING_STANDARD.md
• •core/STRUCTURED_OUTPUT_STANDARD.md

When to use

• •You need a standard document-to-Markdown conversion path.
• •You need deterministic conversion that does not require an LLM by default.
• •You need optional visual/LLM interpretation mode behind an explicit flag.

Hard MUST rules

1. •Use binary streams only (io.BytesIO + convert_stream).
2. •Default conversion mode MUST NOT require an LLM.
3. •Visual/LLM interpretation mode MUST be explicit and optional.
4. •All wrappers MUST write debug logs to file.
5. •All typed outputs MUST use pydantic.BaseModel.
6. •Tests MUST use stubs/fakes for LLM and framework integrations.

Workflow

1. •Validate request payload into a Pydantic request model.
2. •Build a binary stream from content_bytes.
3. •Instantiate MarkItDown in deterministic mode by default.
4. •Enable optional visual mode only when requested and explicitly wired with an llm_client.
5. •Convert with convert_stream.
6. •Return a Pydantic result model and emit debug log events.

Output checklist

• • Binary-stream conversion implemented.
• • Pydantic input/output models present.
• • Debug file logging enabled.
• • No-LLM default behavior documented and tested.
• • Optional visual mode documented and stub-tested.
• • Markdown structure contract test included.