ai-pi-gg-gemma

name: ai-pi-gg-gemma description: Adversarial review via pi using Gemma 4-31B-IT (Google) - multi-provider coding agent dispatch version: 1.0.0 enforcement: strict triggers: - /ai-pi-gemma-4-31b-it - /pi-gemma - adversarial review gemma - review with gemma-4-31b-it workflow_steps: - Parse target (file path or description) - Invoke pi dispatch_single with gemma model - Parse JSON output for findings - Report score, issues, and synthesis allowed_tools: - Bash - Read

/ai-pi-gemma-4-31b-it — Adversarial Review via Pi + Gemma 4-31B-IT

Run adversarial code review using Gemma 4-31B-IT via the pi multi-provider coding agent.

Quick Start

pi --model google/gemma-4-31b-it \
  -p @P:/path/to/file.py \
  "Read the file at @{path} carefully. Return JSON with score (0-1), summary (1 sentence), and issues. Begin your response with the exact first line of the file, quoted. If you did not read the file, set score to 0.0."

Model Info

API Key Setup

pi requires NVIDIA_NIM_API_KEY at startup even for google provider. Export both keys:

NV_KEY=$(python -c "import json; d=json.load(open('$USERPROFILE/.pi/agent/auth.json')); print(d.get('nvidia',{}).get('key',''))")
export NVIDIA_NIM_API_KEY="$NV_KEY"
pi --model google/gemma-4-31b-it -p @P:/path/to/file.py "Read the file at @{path} carefully. Return JSON with score (0-1), summary (1 sentence), and issues. Begin your response with the exact first line of the file, quoted. If you did not read the file, set score to 0.0."

Verification

pi --model google/gemma-4-31b-it -p @P:/path/to/file.py "Say hello"

If you get 429 status code — rate limit hit. Wait 30s and retry.