dcurlewis/.claude/skills/doc

Process documents from the queue:

1. List files in Raw-Materials/Docs/
2. Identify the OLDEST file by date (ascending order)
3. Read ONLY that single file
4. Process and write the summary
5. Archive the processed document
6. STOP - do not process multiple files

PDF Visual Processing

When the source file is a PDF, use both text extraction and page-image rendering to ensure diagrams, charts, tables, and visual layout are captured alongside the text content.

Steps:

1. Create a temporary directory: /tmp/ai-context-pdf-pages/
2. Convert all pages to PNG images at 200 DPI using one of:
   • pdftoppm -png -r 200 "<source>.pdf" /tmp/ai-context-pdf-pages/page (poppler-utils)
   • Or via Python pymupdf (fitz) if pdftoppm is unavailable:

     import fitz
     doc = fitz.open("<source>.pdf")
     for i, page in enumerate(doc):
         pix = page.get_pixmap(dpi=200)
         pix.save(f"/tmp/ai-context-pdf-pages/page-{i+1}.png")
     

3. Read both the page images (for visual content) and the extracted text (for reliable searchable text).
4. When summarising, note any significant visual elements (diagrams, flow charts, architecture visuals, tables) that are not fully represented in the text extraction. Include a brief description of what each diagram conveys in the summary under a Visual Content section (only if visual elements exist).
5. Clean up: remove /tmp/ai-context-pdf-pages/ after processing is complete.

Install dependency if needed: pip install pymupdf or brew install poppler (for pdftoppm).

Context Integration

Before summarising, scan recent documents and meeting summaries from past 2 weeks in Curated-Context/ for:

• Related documents or discussions on this topic
• Previous versions or updates to this content
• Dependencies or impacts on other workstreams

Balanced Reporting

• Foundation: Report what is explicitly stated in the document
• Context enrichment: Note connections to recent work where relevant
• Professional analysis: Identify implications without dramatisation
• NO dramatic descriptors (crisis, critical, urgent) unless directly quoted
• Attribute all assessments to document or named authors: "The report describes this as urgent" NOT "This is urgent"

Summary Structure

1. Document Overview - Type, author, date, purpose if stated
2. Key Topics Covered - Factual list of what was discussed
3. Main Findings/Conclusions - Only those explicitly stated
4. Visual Content - (PDFs only, if applicable) Brief descriptions of diagrams, charts, or visual elements and what they convey. Omit this section if the document contains no significant visual content.
5. Action Items - Items requiring follow-up if mentioned
6. Key Data Points - Important metrics, dates, numbers if present
7. Direct Quotes - Include 2-3 relevant direct quotes that capture key insights
8. Context Notes - Connections to recent work or decisions (if relevant)
9. Insights - Professional observations about implications (1-2 max, grounded in facts)

Leadership Partner Guidelines

• Note if document updates or contradicts previous positions
• Identify impacts on current projects or decisions
• Highlight gaps or questions that need addressing
• Keep insights factual and actionable

Wikilinks

Apply Obsidian wikilinks to the output file per Guidelines/wikilink-guidelines.md.

Entity Updates

After writing the summary, update entity stubs for any new facts discovered in the document (per Guidelines/wikilink-guidelines.md § Stub Creation Rules, rule 6):

• People: Add timeline entries for role changes, project assignments, or significant decisions.
• Vendors: Add timeline entries for contract changes, evaluations, migration decisions.
• Projects: Add timeline entries for scope changes, milestones, status updates.
• Teams: Add timeline entries for reorgs, new responsibilities.

Use the document's date for timeline entries. Only update for confirmed facts, not proposals or speculation.

Output

1. Create summary file using naming format [YYYYMMDD]-Document-Title.md. Route to the appropriate directory based on content type:

   | Content type | Directory | Examples | |---|---|---| | Time-bound: decisions, project docs, strategic docs | Insights/YYYYMM/ | Strategy reviews, planning docs, proposals | | Evergreen: reference material, technical docs | Knowledge/YYYYMM/ | Architecture docs, team knowledge, org context | | Career/performance content | Growth/YYYYMM/ | Feedback, development plans, review materials |

   Extract year and month from the document date for the YYYYMM/ path; create the subdirectory if it does not exist. Include YAML frontmatter with authorship: ai (see Guidelines/wikilink-guidelines.md § Authorship).

2. Archive the processed document: move from Raw-Materials/Docs/ to Archive/Raw-Materials/Docs/

3. Inform the user to run the command again if there are more files to process