ai-pi-nv-qwen-coder

name: ai-pi-nv-qwen-coder description: Adversarial review via pi using Qwen3 Coder 480B A35B (35B MoE, 262K context) - multi-provider coding agent dispatch version: 1.0.0 enforcement: strict triggers: - /ai-pi-qwen3-coder-480b - /pi-qwen3-coder - adversarial review qwen - review with qwen3-coder-480b workflow_steps: - Parse target (file path or description) - Invoke pi dispatch_single with qwen model - Parse JSON output for findings - Report score, issues, and synthesis allowed_tools: - Bash - Read

/ai-pi-qwen3-coder-480b — Adversarial Review via Pi + Qwen3 Coder 480B A35B

Run adversarial code review using Qwen3 Coder 480B A35B (35B MoE, 262K context) via the pi multi-provider coding agent.

Quick Start

pi --model nvidia-nim/qwen/qwen3-coder-480b-a35b-instruct \
  -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. Export from auth.json:

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 nvidia-nim/qwen/qwen3-coder-480b-a35b-instruct -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 nvidia-nim/qwen/qwen3-coder-480b-a35b-instruct -p @P:/path/to/file.py "Say hello"

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