菜单
Goal-driven UI test suite for Birdhouse. Runs an agent through real user flows in an isolated server using browser automation, producing timestamped artifacts (screenshots, logs, report).
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | ui-test-suite |
| description | Goal-driven UI test suite for Birdhouse. Runs an agent through real user flows in an isolated server using browser automation, producing timestamped artifacts (screenshots, logs, report). |
| tags | ["testing","birdhouse"] |
| trigger_phrases | ["UI test suite"] |
| metadata | {"internal":true} |
Traditional UI tests break when you rename a button or restructure a page. This suite takes a different approach: give an agent user goals, a fresh isolated environment, and browser automation tools, then let it figure out how to achieve those goals while observing what is clear, confusing, or broken along the way.
This makes tests resilient to UI changes (the agent adapts like a real user) and useful for UX review instead of only producing brittle pass/fail assertions. A good run should tell you both whether the flow worked and where a new user would struggle.
Each test case is a spec, not a script. It describes what the user is trying to accomplish, what success looks like, and what deserves scrutiny. The agent should adapt to the UI it finds rather than expecting exact copy, selectors, or button positions.
If you are running or authoring this suite, read these files in this order before doing anything else:
test-cases/ — the specific user journey to execute, for example test-cases/ftue-walk-through.mdDo not assume the test case contains runner setup. Environment isolation, temp directories, browser cleanup, and credentials handling belong in the runbook, not in the test case.
tmp/ui-test-runs/ containing screenshots, logs, and the written report for that runRun, write, and manage Birdhouse agentic tests. Each test exercises Birdhouse end-to-end using a real browser and a real sandbox or explicitly requested isolated OpenCode instance. Use when asked to run agentic tests, run a specific test, or add a new agentic test.
Add isolated Birdhouse web testing to an existing worktree-based agent workflow. Teaches agents to delegate browser validation to child agents, capture screenshots and MP4s, and clean up isolated runs when Cody asks.
Re-apply the Birdhouse opencode fork commits onto a new upstream tag one commit at a time. Load this skill when rebasing Birdhouse's custom opencode patches onto a fresh upstream release.
Compute Fibonacci numbers using recursive agent delegation where each agent spawns children for sub-problems. Use when you want to run a fibonacci test, recursive fibonacci test, or fibonacci test.
Birdhouse frontend architecture patterns for SolidJS and Corvu. Load when working on modals, dialogs, popovers, typeaheads, reactive component trees, buttons and navigation, or any component layering in birdhouse-workspace. Cross-references the generic `solidjs-best-practices` and `corvu-best-practices` skills.
Search message content across Birdhouse agents to find agents by what they did or said. Use when asked to find an agent, search agents, look up what an agent did, or locate a past conversation.