drive-markdown-converter

Drive → Markdown Converter

Converts Google Docs and PDFs in a source Drive folder to Markdown and uploads results to a target folder. Core constraint: raw file content NEVER lands in the calling context. All I/O runs inside an isolated artifact or subagent.

---

Input Parameters

Collect these before choosing a mode:

Parameter	Required	Description
source_folder_id	✅	Google Drive folder ID to read from
target_folder_id	✅	Google Drive folder ID to write to
file_types	optional	docs, pdfs, or both (default: both)
naming_convention	optional	e.g. {original_name}.md (default)
dry_run	optional	List files without converting (default: false)

Folder IDs are the alphanumeric string in the Drive URL after /folders/.

---

Mode Selection

User wants interactive UI / one-off run?  →  MODE A: Artifact-Generator
User wants pipeline integration / JANUS?  →  MODE C: Subagent-Prompt
No preference stated?                     →  Ask with ask_user_input_v0

---

MODE A — Artifact-Generator

→ Read references/artifact-mode.md

Generates a React artifact that:

1. Accepts folder IDs as input
2. Calls Anthropic API with Google Drive MCP (https://drivemcp.googleapis.com/mcp/v1)
3. Subagent reads, converts, uploads — content stays in subagent context
4. Artifact displays only progress + result summary (file names, byte counts, errors)

When to use: User runs this manually, wants to see progress, single conversion job.

---

MODE C — Subagent-Prompt

→ Read references/subagent-mode.md

Generates a complete Anthropic API call payload (system prompt + user message) that can be embedded in a pipeline, cron job, or JANUS Executor node.

When to use: Automated pipeline, llm-wiki-agent ingest, JANUS orchestration, repeated scheduled runs.

---

Content Isolation Guarantee

Both modes enforce this contract:

Caller context  ←  {status, file_list, error_summary}  only
Isolated layer  ←  raw file content, conversion, upload

Never return raw Markdown content to the caller. If the user explicitly asks to see a converted file's content, that is a separate deliberate action outside this skill's default behaviour.

---

Conversion Strategy by File Type

Type	Drive MIME	Export Method	Markdown Quality
Google Doc	application/vnd.google-apps.document	Drive export as text/plain + post-process headers	High
PDF (text-based)	application/pdf	Download → extract text via pdfminer	Medium-High
PDF (scanned)	application/pdf	Download → OCR note + raw text fallback	Low (flag to user)

Google Docs export via the Drive API export endpoint preserves heading structure better than downloading as DOCX first.

---

Pre-Deploy Checklist

• Both reference files present in references/
• User has Google Drive MCP connected (check claude.ai connectors)
• Source and target folder IDs confirmed before generating output
• Mode selected (A or C)