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/primer/agent-eval --skill changesets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。
正在显示 SKILL.md
| name | changesets |
| description | Determine whether agent-eval changes need a changeset and write consumer-focused changesets for public API changes. |
Add a changeset when a pull request changes the public API or user-visible
behavior of the published @primer/agent-eval package. Public API includes:
Do not add a changeset for scenarios, experiments, tests, internal refactors, or
repository tooling that do not affect consumers of @primer/agent-eval.
pnpm changeset from the repository root.@primer/agent-eval.patch for a backwards-compatible fix to existing behavior.minor for a backwards-compatible public API addition or enhancement.major for a breaking public API change that requires consumers to update
their code or configuration.Describe the public API or behavior that was added, changed, fixed, or removed. For breaking changes, state what consumers need to change. Do not describe implementation details, internal refactors, file changes, or the development process unless they directly affect consumers.
Prefer:
Add support for filtering scenarios by multiple tags.
Avoid:
Refactor the scenario loader and update its internal filtering helper.
Use when designing, testing, or refining an LLM-facing resource such as a skill, repository instruction, prompt, custom agent, documentation bundle, or MCP setup. Creates treatment-blind agent-eval scenarios, compares resource treatments against control, and iterates in quick or confidence-building mode.
Use when creating or reviewing agent evaluation scenarios and experiments, especially when comparing a skill, instruction, agent, or tool treatment against a control. Covers causal isolation, prompt leakage, blank-slate fixtures, grader design, browser-based UI evaluation, and deciding whether a requirement belongs in an eval or deterministic tooling.
基于 SOC 职业分类