Prompt Frequency Analyzer

Run a prompt N times against Gemini with Google Search grounding enabled. Collect and report the frequency of search queries Gemini generates across all runs.

Usage

CODEBLOCK0

Run from the skill directory. Resolve scripts/analyze.py relative to this SKILL.md.

Options

• - --runs N — Number of times to run the prompt (default: 10)
• INLINECODE2 — Gemini model to use (default: gemini-2.5-pro)
• INLINECODE3 — Max parallel API calls (default: 5; keep ≤5 to avoid rate limits)
• INLINECODE4 — Output format (default: text)

Output

Reports for each unique search query:

• - Frequency percentage (how many runs used that query)
• Raw count
• Top web sources referenced

Notes

• - Gemini API key must be in GEMINI_API_KEY env var (stored in macOS Keychain under nano-banana-pro)
• Each run is independent — Gemini may use different search queries each time
• Retries failed requests up to 3 times with exponential backoff
• Use --output json for programmatic consumption