skills.rest
QwenLM
QwenLMOFFICIAL
Explore all skills by this author →
Website𝕏@Alibaba_QwenGitHub

agent-reproduce-feature

Capture reference agent HTTP traffic, prompts, and terminal output to reproduce features in Qwen Code.

27.5k|3.0k|Updated Jun 26, 2025
Search Tags:#mitmproxy#traffic capture#tool schemas#Tmux#agent CLI#qwen code#feature reproduction
One-click install
npx skills add https://github.com/QwenLM/qwen-code --skill agent-reproduce-feature
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: agent-reproduce-feature
Source: https://github.com/QwenLM/qwen-code/tree/main/.qwen/skills/agent-reproduce-feature
Command: npx skills add https://github.com/QwenLM/qwen-code --skill agent-reproduce-feature

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires mitmproxy, and includes scripts (resource) and references (resource) components.

What problem does it solve?

Reimplementing a feature from another AI coding agent (Codex or Claude Code) in Qwen Code requires knowing its exact prompts, tool schemas, and terminal behavior, which are not publicly documented. This Skill provides a structured capture-and-implement workflow that turns observed reference-agent behavior into a concrete implementation task.

Core Features & Use Cases

  • HTTP Traffic Capture: Run a nested reference agent through mitmproxy to record redacted request/response bodies, including messages, tool schemas, and model settings.
  • Terminal and State Capture: Record interactive TUI sessions via tmux and snapshot/diff local agent state (~/.codex or ~/.claude) before and after a scenario.
  • Guided Implementation Workflow: Extract behavioral facts from traces, implement the smallest compatible behavior in the target repo, and add focused verification tests.
  • Use Case: You want Qwen Code to support a slash command that exists in Codex. Use this Skill to run Codex under capture, extract the command's request shape and terminal output, then implement and test the matching behavior.

Quick Start

Ask the agent to reproduce a specific Codex or Claude Code feature in Qwen Code, naming the feature and which reference agent to capture from.

Frequently Asked Questions about agent-reproduce-feature

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I capture HTTP requests from Codex or Claude Code?▼

Run the reference agent through the run_with_mitm.sh script, which starts a local mitmdump proxy and records redacted request and response bodies to http.jsonl. It sets standard proxy and CA certificate environment variables for Node, Python, and curl-based CLIs.

How do I reproduce a Claude Code feature in another agent CLI?▼

Select claude-code as the reference agent, discover its local launch command, then capture its HTTP traffic, terminal output, and state changes while exercising the feature. Extract the prompts, tool schemas, and visible behavior from the trace and implement the smallest matching behavior in the target repo.

Does the capture workflow work with interactive TUI features?▼

Yes, the run_tmux_capture.sh script launches the reference agent in a detached tmux session and captures pane contents for slash-command and TUI-dependent features. The session stays alive so you can send keystrokes and capture again.

Why is my http.jsonl capture file empty?▼

An empty capture usually means the CLI ignores the proxy variables or TLS trust is not configured. Verify the mitmproxy CA certificate exists at ~/.mitmproxy/mitmproxy-ca-cert.pem and confirm the CLI supports HTTP_PROXY and NODE_EXTRA_CA_CERTS before interpreting missing traffic as product behavior.

Are captured API keys and tokens stored safely?▼

The capture scripts redact authorization headers, bearer tokens, API keys, and private keys before writing records, and output files are created with owner-only permissions. You should still review state diffs and traces before copying any excerpt into tracked files or commits.

Related Skills

Explore More Software Engineering Skills→
alexfazioalexfazio
159

cc-trace

Intercept and inspect Claude Code API traffic using mitmproxy.

Community
Advanced
cncorpcncorp

langfuse-prompt-iterator

Automate test-driven prompt iteration and debugging for Langfuse prompts.

Official
Advanced
itallstartedwithaideaitallstartedwithaidea
31

secret-protection

Scan .env files, code, and diffs for exposed secrets and redact runtime outputs.

Community
Advanced
liuzhengdongfortestliuzhengdongfortest
1.1k

cs-libdoc

Generates API references from source code for libraries and frameworks.

Community
Advanced