Match Loop by John Perry

Two-agent iterative vibe-coding loop for OpenClaw. Use when the user wants one sub-agent to generate/build/code an app or artifact and a second analyst agent...

This skill appears coherent for running a generator ↔ analyst visual QA loop. Before using it, be aware that the loop expects agents to: run project commands (install dependencies, start dev servers), open browsers and take screenshots, and potentially invoke OS-level UI automation (AppleScript/Hammerspoon/macbot) if needed. Those actions require filesystem, process, and UI automation privileges on the host and could execute code produced by the generator agent — which is normal for this workflow but carries the usual risk of running untrusted code. If you plan to allow autonomous runs, consider sandboxing the agent, restricting network access, or reviewing generated code before execution; do not provide secrets or credentials to the skill unless explicitly required and justified.

Type: OpenClaw Skill Name: match-loop Version: 1.0.0 The `match-loop` skill defines an iterative workflow where one agent generates code and another performs visual QA. It explicitly instructs agents to execute shell commands (starting dev servers), automate browsers (Playwright), and use desktop control tools (macbot, Hammerspoon, AppleScript) as described in `SKILL.md` and `references/loop-patterns.md`. While these high-risk capabilities are aligned with the stated purpose of "vibe-coding" and UI refinement, they involve the execution of and interaction with potentially untrusted, auto-generated code, which poses an inherent security risk without evidence of intentional malice.