Two tools for complex Claude Code workflows: doc2md + orchestration protocol

[neoncapy]

Sharing two open-source tools that complement the resources in this repo:

1. doc2md — Document-to-Markdown Pipeline

Problem: Anthropic's copyright filter blocks PDF reading in Claude Code.

Solution: A 15,000-line Python pipeline that converts PDF/DOCX/PPTX to high-fidelity Markdown with:

• Multi-extractor support (pymupdf, pdfplumber, MinerU)
• Per-image classification (8 heuristics)
• Multi-stage QC (loops until zero issues)
• Claude Code integration via PreToolUse hook and SKILL.md

https://github.com/orangefineblue/doc2md

2. claude-code-orchestration-protocol — Zero-Read Orchestrator

Problem: Context rot degrades output quality in long Claude Code sessions.

Solution: A /orchestrate slash command that turns the main session into a pure router:

• 100-line hard read limit for main session
• QC loops until genuinely zero issues (all severities)
• Anti-polling protocol (prevents 50-150K token dumps)
• Progressive documentation with handover architecture
• 5-tier token budget management

https://github.com/orangefineblue/claude-code-orchestration-protocol

Both MIT licensed. They work well together for batch document processing where context management matters.