Generate Stimulus Set

# Generate Stimulus Set

Generate product descriptions and save to CSV.

## Usage

/generate-stimulus-set [column-name] [word-limit]

Example: `/generate-stimulus-set 50_word 50`

## Parameters
- **column-name**: Name for CSV column AND stimulus set filename (e.g., `50_word`, `25_word`)
- **word-limit**: Maximum words for each description (e.g., `50`, `25`, `100`)

## Critical Rules

**NEVER create JSON files directly** - always use the Python export function (Step 5).

**Product names MUST come from CSV's `Study Name` column** - sub-agents only generate descriptions, never names.

## Process

**Step 1: Check if column exists**
```bash
python scripts/read_products.py check [column-name]

Returns: {"exists": true/false, "has_data": true/false, "count": N}

• If has_data is true: STOP and tell the user the column already exists. They must choose a different column name.
• If not, proceed to Step 2

Step 2: Read products

python scripts/read_products.py read

Returns: [{asin, name, price, title}, ...]

Step 3: Generate descriptions (via sub-agents in batches of 5) using this prompt:

Generate a product description for the similarity experiment.

Product: "[TITLE]"
ASIN: [ASIN]

Instructions:
1. Web search for key features (use snippets only, don't fetch pages)
2. Write under [WORD-LIMIT] words, factual, feature-focused, neutral tone
3. Return ONLY: [ASIN]: description text
4. Do NOT include product name, price, or any other fields
5. Do NOT return JSON or markdown formatting

Output format: [ASIN]: Your description here as plain text.

Example output:
B00CH9QWOU: Semi-automatic espresso machine with integrated burr grinder. Features 15-bar pump, PID temperature control, and steam wand for milk frothing.

Step 4: Write to CSV

python scripts/read_products.py write [column-name] '{"ASIN": "description", ...}'

Adds column [column-name] to CSV with descriptions.

Step 5: Export JSON

python scripts/read_products.py export [column-name] [word-limit]

Creates stimuli/[column-name].json:

{
  "products": [
    {"id": "ASIN", "name": "Study Name", "description": "...", "price": "$X", "image": "ASIN.png"}
  ],
  "metadata": {"created": "YYYY-MM-DD", "word_limit": [WORD-LIMIT]}
}

Step 6: Validate output

1. Read the exported JSON file
2. Run python scripts/read_products.py read to get CSV data
3. Compare each product's name in JSON against the name from CSV
4. Report any mismatches - if found, the JSON was not created correctly

This validation ensures names came from the CSV's Study Name column, not from sub-agent generation.

Parallelization

Launch sub-agents in synchronous parallel batches of 5-6:

1. Send a single message with 5-6 Task tool calls (no run_in_background)
2. Results return directly in the response
3. Collect results, then launch next batch
4. Repeat until all products are processed

Do NOT use run_in_background: true - output files may be empty.

Notes

• name comes from Study Name column
• price comes from current_price column
• Only products with Include = "X" are processed
• Images must exist as stimuli/[ASIN].png