ebook-writer: Skill for Claude Code

Reference Analyzer Skill

Purpose

Parse and extract key content from reference materials in various formats (.md, .pdf, .txt, .docx).

Capabilities

1. File type detection and appropriate parsing
2. Key content extraction (concepts, methodologies, insights)
3. Structured output for integration with research report
4. Bounded chunk-file output to avoid injecting full references into agent context

Scripts

• scripts/parse_references.py — Parse reference files and extract text content

Usage

.venv/bin/python3 .claude/skills/reference-analyzer/scripts/parse_references.py <file_path>
.venv/bin/python3 .claude/skills/reference-analyzer/scripts/parse_references.py <file_path> \
  --output-dir output/research/reference_chunks/<source_name>/

Output: JSON to stdout

{
  "filename": "reference.pdf",
  "content_type": "pdf",
  "extracted_text": "...",
  "word_count": 1500,
  "status": "success"
}

Prefer --output-dir in the ebook pipeline. In that mode stdout contains only status, word count, chunk count, and manifest path; extracted text is written to bounded chunk JSON files under output/research/reference_chunks/.

Chunked stdout example:

{
  "filename": "reference.pdf",
  "content_type": "pdf",
  "word_count": 1500,
  "chunk_count": 3,
  "manifest_path": "output/research/reference_chunks/reference_pdf/manifest.json",
  "status": "success"
}

When to Use

• Researcher Agent processes files in input/references/
• When a domain plugin specifies reference materials

Failure Handling

• PDF parsing failure → attempt textutil conversion (macOS)
• DOCX parsing failure → attempt raw XML extraction
• If all parsing methods fail → return status "failed" with error message
• Never block the pipeline on a single reference file failure