What problem does it solve?
After implementing a feature in Qwen Code that mirrors Codex or Claude Code, you need evidence that the behavior actually matches the reference agent. This Skill runs both agents under the same scenario, captures HTTP and terminal traces, and compares request bodies, tool schemas, and outputs so parity gaps are found systematically instead of by guesswork.
Core Features & Use Cases
- Paired Trace Capture: Runs the reference agent and Qwen Code in separate capture directories with the same scenario, including before/after snapshots of reference-agent local state.
- Trace Normalization and Diffing: Normalizes mitm JSONL traces into stable JSON (dropping timestamps, IDs, and auth headers) and diffs request bodies, message roles, tool schemas, and required fields.
- Prioritized Triage Loop: Ranks differences by severity (missing tools and wrong schemas first, wording last) and drives a patch-rerun-compare iteration loop until must-match fields align.
- Use Case: You implemented a slash command in Qwen Code modeled on Claude Code. Run both agents through the same prompt, normalize the traces, and confirm the tool schemas and request structure match before shipping.
Quick Start
Ask the agent to run the agent-reproduce-align workflow comparing Qwen Code against codex for the feature you just implemented, using the paired capture script and trace comparison.
