Context Ingestion

Context Ingestion

Scans the project folder, validates structure, fetches the GitHub repository, and generates an inventory of all available materials.

Input

User provides path to project folder (or current directory if already there).

Workflow

[Receive project path]
     │
     ▼
[Validate Folder Structure] ─── Check required folders exist
     │
     ▼
[Parse config.md] ─── Extract GitHub URL, constraints
     │
     ▼
[Clone GitHub Repository] ─── Fetch code for analysis
     │
     ▼
[Inventory Materials] ─── List all available files
     │
     ▼
[Extract IRB Content] ─── If irb/ exists, generate notes/irb-summary.md
     │
     ▼
[Generate inventory.md] ─── Structured summary

Step 1: Validate Folder Structure

Check that required folders exist:

# Required structure
project/
├── papers/       # Must exist (can be empty)
├── data/         # Must exist (can be empty)
├── figures/      # Must exist (can be empty)
├── irb/          # Optional - IRB protocol and regulatory documents
└── config.md     # Must exist

Validation:

cd /path/to/project

# Check required folders
[ -d "papers" ] || echo "ERROR: papers/ folder missing"
[ -d "data" ] || echo "ERROR: data/ folder missing"
[ -d "figures" ] || echo "ERROR: figures/ folder missing"
[ -f "config.md" ] || echo "ERROR: config.md missing"

If validation fails, inform user what's missing and provide the expected structure template.

Step 2: Parse config.md

Extract configuration values:

# Expected config.md format

## GitHub Repository
url: https://github.com/username/repo-name
branch: main
access: private

## Constraints
word_limit: 3500
target_journal: Radiology: Artificial Intelligence
citation_style: AMA

## Additional Notes
[Free text notes]

Parse and store:

• github_url: Repository URL
• github_branch: Branch to clone (default: main)
• github_access: public or private
• word_limit: Target word count
• target_journal: Journal name for formatting
• citation_style: AMA, Vancouver, APA, etc.

Step 3: Clone GitHub Repository

For public repositories:

git clone --depth 1 --branch main https://github.com/username/repo-name.git code/

For private repositories, user must have GitHub CLI authenticated:

gh repo clone username/repo-name code/ -- --depth 1 --branch main

If clone fails:

1. Check if gh is authenticated: gh auth status
2. Provide instructions: "Run gh auth login to authenticate"
3. Allow user to proceed without code (Methods section will be limited)

Store cloned repo at: project/code/

Step 4: Inventory Materials

Scan each folder and catalog contents:

Papers Inventory

ls -la papers/*.pdf 2>/dev/null | wc -l  # Count PDFs

For each PDF, extract basic info:

• Filename
• File size
• (Attempt to extract title from first page if possible)

Data Inventory

ls -la data/*.csv data/*.xlsx 2>/dev/null

For each data file:

• Filename
• File size
• Row/column count (for CSVs)
• Sheet names (for Excel)

Preview CSV structure:

head -5 data/results.csv

Figures Inventory

ls -la figures/*.png figures/*.jpg figures/*.svg 2>/dev/null

For each figure:

• Filename
• Dimensions (if determinable)
• File size

Code Inventory

If GitHub clone succeeded:

find code/ -name "*.py" -o -name "*.ipynb" -o -name "*.R" | head -20

Identify:

• Primary language (Python, R, etc.)
• Notebook files (.ipynb)
• Key script files
• Requirements/dependencies file

IRB Inventory (Optional)

If irb/ folder exists, scan for regulatory documents:

ls -la irb/*.pdf irb/*.docx irb/*.md 2>/dev/null

Supported formats:

• .md - Read directly with Read tool
• .pdf - Read with Claude's native PDF capability
• .docx - Extract text using document-skills:docx skill

Step 5: Extract IRB Content

Skip this step if irb/ folder does not exist or is empty.

For each document in irb/:

1. Read the document content using appropriate method for format
2. Extract comprehensive study information
3. Generate notes/irb-summary.md

IRB Summary Template

Create notes/irb-summary.md:

# IRB Document Summary

**Source**: [filename]
**Extracted**: [timestamp]

## Study Identification
- **Protocol Title**: [extracted or "[not found]"]
- **IRB Approval Number**: [extracted or "[not found]"]
- **Principal Investigator**: [extracted or "[not found]"]
- **Approval Date**: [extracted or "[not found]"]

## Study Design
- **Study Type**: [interventional/observational/retrospective/etc.]
- **Design**: [RCT, cohort, case-control, cross-sectional, etc.]
- **Duration**: [study period]

## Population
- **Target Population**: [description]
- **Inclusion Criteria**:
  - [criterion 1]
  - [criterion 2]
  - ...
- **Exclusion Criteria**:
  - [criterion 1]
  - [criterion 2]
  - ...
- **Sample Size**: [N with justification if provided]

## Procedures & Interventions
- [Procedure 1]
- [Procedure 2]
- ...

## Endpoints
- **Primary**: [endpoint]
- **Secondary**: [endpoints]

## Statistical Considerations
- **Power Analysis**: [if provided or "[not found]"]
- **Planned Analyses**: [if provided or "[not found]"]

## Notes
[Any additional relevant context, caveats, or sections that were unclear]

Mark fields as [not found] if not present in the document.

Step 6: Generate inventory.md

Create structured inventory document:

# Project Inventory

Generated: [timestamp]
Project: [folder name]

## Configuration

- **GitHub**: [url] (branch: [branch])
- **Target Journal**: [journal]
- **Word Limit**: [limit]
- **Citation Style**: [style]

## Papers ([count] files)

| Filename | Size | Notes |
|----------|------|-------|
| smith-2023.pdf | 1.2 MB | |
| jones-2022.pdf | 0.8 MB | |

## Data ([count] files)

| Filename | Size | Rows | Columns | Preview |
|----------|------|------|---------|---------|
| results.csv | 45 KB | 156 | 12 | patient_id, age, sex, ... |
| demographics.csv | 12 KB | 156 | 8 | patient_id, age, sex, ... |

## Figures ([count] files)

| Filename | Dimensions | Size |
|----------|------------|------|
| figure1.png | 1200x800 | 340 KB |
| figure2.png | 1000x600 | 210 KB |

## Code Repository

- **URL**: [github url]
- **Language**: Python
- **Key Files**:
  - `analysis.ipynb` - Main analysis notebook
  - `preprocessing.py` - Data preprocessing
  - `models.py` - ML models
- **Dependencies**: pandas, scikit-learn, matplotlib, ...

## IRB Documents

| Filename | Format | Status |
|----------|--------|--------|
| protocol.pdf | PDF | ✓ Extracted to notes/irb-summary.md |

*Or: "No IRB documents provided"*

## Summary

| Category | Count | Status |
|----------|-------|--------|
| Papers | [n] | ✓ Ready |
| Data files | [n] | ✓ Ready |
| Figures | [n] | ✓ Ready |
| Code repo | 1 | ✓ Cloned |
| IRB documents | [n] | ✓ Extracted / Not provided |

## Missing/Warnings

- [List any issues found]

Output

Save to: project/inventory.md

Create notes directory structure:

mkdir -p notes/papers notes/search notes/references notes/papers-library drafts

Return to parent skill with inventory summary.