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。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/microsoft/amplifier-agent --skill e2e-crusty-probe命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Build software on amplifier-agent, the Microsoft agent engine that other software runs on. Use when (1) adding an AI agent, agent loop, or chat backend to an app, service, CLI, or bot, (2) integrating amplifier-agent from TypeScript/Node, Python, HTTP, or a shell script, (3) choosing between the TypeScript SDK, Python SDK, in-process library, HTTP face, and raw CLI contract, (4) debugging a host that spawns the engine: approval_unconfigured, protocol_version_mismatch, binary_not_found, lost session continuity, or stdout/stderr parsing. Triggers on "amplifier-agent", "amplifier agent", "amplifier-agent-ts", "amplifier_agent_py", "spawnAgent", "spawn_agent_sync", "serve chat-completions", "amplifier_agent_lib", "add an agent to my app".
Start an amplifier-agent release: scope which artifacts need cutting, sweep the diff for anything that does not belong in a public release, set versions, write the changelog, run the local gate, then open the release PR. This is the FIRST of two release skills. It stops at an open PR. After the PR merges, run amplifier-agent-finish-release-process to tag, publish, and update downstreams.
Triage and fix an amplifier-agent bug, from report to verified fix, using E2E-test-driven development. Covers intake, ruling out non-code causes, root-cause pinning, the coverage-gap gate, red/green regression TDD, scoped verification, and the evaluation path for quality regressions where contracts hold but output got worse. Stops before release.
正在显示 SKILL.md
基于 SOC 职业分类
| name | e2e-crusty-probe |
| description | E2E test reviewer lens. Gives a short grounded engineering critique of the target. |
| user-invocable | true |
A minimal reviewer-lens persona used only by the amplifier-agent e2e skill-invocation tests.
Give a short, grounded engineering critique of the following target, focusing on hidden costs and likely failure modes. Keep it to a few sentences.
Target: $ARGUMENTS
