Skip to content

Context & Input

Understand the distinction between input and context files, supported document types, and how content is aggregated.

Input vs Context

WireFlow distinguishes between input (primary documents to process) and context (supporting materials):

Aspect Input (-in) Context (-cx)
Purpose Documents to analyze/transform Background information
Position After context in prompt Before input in prompt
Typical use PDFs to summarize, data to analyze Reference docs, prior outputs
Citations Yes (with --enable-citations) Yes

Example: 

# Summarize a report (input) using style guide (context)
wfw run summarize -in report.pdf -cx style-guide.md

# Analyze data (input) with reference materials (context)
wfw run analyze -in results.csv -cx methodology.md -cx prior-analysis.md

Supported Document Types

Text Files

All text-based files (.md, .txt, .py, .js, .csv, etc.) are processed directly with syntax highlighting based on extension.

wfw run analysis -cx data.csv -cx script.py -cx notes.md

PDF Documents

  • Processed using Claude API (text + visual analysis)
  • Maximum size: 32MB per PDF
  • Supports citations with document indices
wfw run analysis -in report.pdf -cx references.pdf

Microsoft Office Files

  • Supported: .docx, .pptx
  • Auto-converted to PDF via LibreOffice
  • Cached in .workflow/cache/office/
  • Citations use original filename
wfw run summary -in presentation.pptx -cx notes.docx

Requirements: LibreOffice installed (libreoffice or soffice in PATH)

Images

  • Native: .jpg, .jpeg, .png, .gif, .webp
  • Converted: .heic → JPEG, .tiff → PNG, .svg → PNG
  • Processed using Claude Vision API
  • Auto-resize if larger than 1568px on longest edge
  • Maximum size: 5MB per image
  • NOT citable (no document indices)
wfw run analyze-diagram -in flowchart.svg -cx photo.heic

Requirements: - HEIC: sips (macOS) or ImageMagick - TIFF: sips (macOS) or ImageMagick - SVG: rsvg-convert (librsvg) or Inkscape

Mixing Types

wfw run research \
    -in "papers/*.pdf" \
    -cx notes.docx \
    -cx diagram.png \
    -cx references.md

Context Aggregation

Content is assembled in this order:

Run mode: 1. System prompts (SYSTEM_PROMPTS) 2. Project description (project.txt) 3. Context files (config + CLI) 4. Dependencies (--depends-on outputs) 5. Input files (config + CLI) 6. Images 7. Task prompt (task.txt)

Task mode: 1. System prompts 2. Project description (if in project) 3. Context files (CLI) 4. Input files (CLI) 5. Images 6. Task prompt (inline or template)

Path Resolution

Config File Paths

Paths in config files are relative to project root. Glob patterns expand when the config is sourced:

# In .workflow/config or .workflow/run/<name>/config
CONTEXT=(notes/analysis.md refs/paper.pdf data/*.csv)
INPUT=(main-report.pdf reports/*.pdf)

CLI Paths

CLI paths are relative to current directory:

cd /project/subdir
wfw run analysis -cx local-notes.md
# Looks for: /project/subdir/local-notes.md

Glob Patterns

# Single directory
-cx "data/*.csv"

# Recursive
-cx "notes/**/*.md"

# Multiple directories
-cx "data/{exp1,exp2}/*"

# In config (globs expand at source time from project root)
CONTEXT=(data/*.csv notes/**/*.md)
INPUT=(reports/**/*.pdf)

Caching

WireFlow caches converted files to speed up repeated runs.

Office Conversion Cache

Location: .workflow/cache/office/

  • PDF conversions are cached per source file
  • Cache invalidated when source file changes (by mtime)
  • Shared across all workflows in the project

Image Conversion Cache

Location: .workflow/cache/images/

  • Resized/converted images cached per source
  • HEIC→JPEG, TIFF→PNG, SVG→PNG conversions cached
  • Original dimensions preserved in cache filename

Cache Management

# Clear project cache
rm -rf .workflow/cache/

# View cache contents
ls -la .workflow/cache/office/
ls -la .workflow/cache/images/

Configuration Variables

Variable Description Scope
CONTEXT Context file paths (globs expand at source time) Project, Workflow
INPUT Input file paths (globs expand at source time) Workflow
ENABLE_CITATIONS Enable source citations All

Best Practices

Input vs Context: - Use -in for documents you're actively processing - Use -cx for reference materials, prior outputs, style guides - Dependencies (--depends-on) automatically become context

File organization: - Keep source documents in predictable locations - Use glob patterns for dynamic file sets: CONTEXT=(data/*.csv) - Use explicit paths for fixed sets: INPUT=(report.pdf summary.md)

Performance: - First run with Office/image files is slower (conversion) - Subsequent runs use cache - Large images auto-resize to reduce API costs

Continue to Execution Modes →