技能创作

人工智能 社区

技能解读

当前原文的自动摘译,尚未人工审校。

技能文件布局

屏幕聊天代理技能位于浏览器模块内mod/<author>/<repo>/ext/skills/...。回购拥有的第一方共享屏幕技能通常应位于app/L0/_all/mod/_core/skillset/ext/skills/...。描述一个模块的私有合约的特定于模块的技能可能存在于该拥有模块的下ext/skills/...树。组范围或仅限管理的屏幕技能可能存在于可读的自定义软件根目录下,例如app/L0/_admin/mod/_core/<module>/ext/skills/...。

目录规则

屏幕提示目录仅列出顶级技能ext/skills/*/SKILL.md。默认情况下不列出嵌套技能。目录和显式space.skills.load(...)调用评估当前文档的<x-context>技能合格之前的标签。metadata.when可能是true或一个{ tags: [...] }状况;metadata.when.tags在技能变为可目录加载之前需要所有列出的标签。

冲突规则

技能 ID 在可读模组中必须是唯一的。如果一项技能仅对较小范围的受众可见,仍然为其提供一个不会与共享冲突的顶级 ID_all这些用户也可以阅读的技能。目录中省略了冲突的 ID。加载冲突的 ID 失败并出现歧义错误。

技能内容规则

从包含以下内容的 frontmatter 开始name,description,以及可选的运行时拥有的metadata。框架引导程序已经导出了一个运行时上下文:data-runtime="browser"在正常的网络会话中或data-runtime="app"在打包的桌面运行时中,加上派生标签runtime-browser或runtime-app;打包的应用程序路由在回退到启动器运行时信息或前端配置之前从桌面桥派生。

维护规则

当镜像源合同发生更改时,请在同一会话中更新受影响的技能文件。当稳定的功能或工作流程发生变化时,更新下面的相关文档/mod/_core/documentation/docs/以及文档技能/mod/_core/documentation/ext/skills/documentation/SKILL.md在同一个会话中。不要让技能指导偏离了拥有AGENTS.md文件。特别是对于开发超级技能,请保留/mod/_core/skillset/ext/skills/development/AGENTS.md每当其镜像的框架、路由器、API、层或身份验证合约发生变化时,它都是最新的。

流狐档案 运行与权限由流狐静态检查推断,不代表作者声明或实际测试结果。
领域
人工智能
作者声明的智能体
未找到明确声明;不据此推断已兼容或已测试
静态检查
88 / 100 · 启发式扫描,不代表运行安全
运行环境证据
未执行;当前没有隔离运行测试证据
作者 / 版本 / 许可证
@agent0ai · 未声明许可证
上下文用量评级
低消耗
接入复杂程度
需简单配置
是否需要外部接口密钥
不需要
兼容的系统
未声明(默认跨平台)
运行环境要求
未检出明确要求
检测到的文件与系统行为
  • 只读
  • 允许写入 / 修改
  • 命令执行
网络行为
仅限本地
安装命令数
无(仅作为资料)

需要注意: 未声明工具权限范围;实际权限由运行环境决定。

作者说明文件(原文) SKILL.md

Use this skill when creating or updating skills for the onscreen chat agent.

## Skill File Layout

- Onscreen chat-agent skills live inside browser modules under `mod/<author>/<repo>/ext/skills/...`.
- Repo-owned first-party shared onscreen skills should normally live under `app/L0/_all/mod/_core/skillset/ext/skills/...`.
- Module-specific skills that describe one module's private contracts may live under that owning module's `ext/skills/...` tree.
- Group-scoped or admin-only onscreen skills may live under readable customware roots such as `app/L0/_admin/mod/_core/<module>/ext/skills/...`.
- A skill file is always named `SKILL.md`.
- The skill id is the path relative to `ext/skills/` with the trailing `/SKILL.md` removed.

Examples:

- `mod/_core/skillset/ext/skills/browser-control/SKILL.md` -> `browser-control`
- `mod/_core/skillset/ext/skills/development/SKILL.md` -> `development`
- `mod/_core/skillset/ext/skills/development/modules-routing/SKILL.md` -> `development/modules-routing`

## Catalog Rules

- The onscreen prompt catalog lists only top-level skills from `ext/skills/*/SKILL.md`.
- Nested skills are not listed by default.
- Both the catalog and explicit `space.skills.load(...)` calls evaluate the current document's `<x-context>` tags before a skill is eligible.
- `metadata.when` may be `true` or a `{ tags: [...] }` condition; `metadata.when.tags` requires all listed tags before the skill becomes catalog-loadable.
- `metadata.placement` accepts `system`, `transient`, or `history`; ordinary skills default missing or invalid placement to `history`, but auto-loaded skills may not resolve to `history`, so missing or invalid placement and explicit `history` all fall back to `system` unless the skill explicitly sets `transient`.
- Only top-level `ext/skills/*/SKILL.md` skills can auto-load through prompt discovery. Nested skills stay explicit-load-only routing targets even if they define `metadata.loaded`.
- Auto-loaded skills appear after the catalog in the `auto loaded` prompt block when their effective placement is `system`, or in the transient channel when they explicitly set `metadata.placement: transient`.
- Routing skills should tell the agent which deeper skill ids to load next.
- `space.skills.load("<path>")` loads the full skill file on demand and applies the same placement rule.
- `history` placement keeps the loaded skill body in ordinary execution-output history.
- `system` placement stores the loaded skill in the current chat's runtime system-skill registry and execution reports `skill loaded to system message`.
- `transient` placement stores the loaded skill in the current chat's runtime transient-skill registry and execution reports `skill loaded to transient area`.
- Auto-loaded skills cannot resolve to `history`; if they also set `metadata.loaded`, then `metadata.placement: history` is treated as `system`.
- A plain top-level `await space.skills.load("<path>")` is enough to apply the placement; use `return` only if you also want the execution result value explicitly.

## Conflict Rules

- Skill ids must be unique across readable mods.
- If a skill is visible only to a narrower audience, still give it a top-level id that will not collide with shared `_all` skills that those users can also read.
- Conflicting ids are omitted from the catalog.
- Loading a conflicting id fails with an ambiguity error.

## Skill Content Rules

- Start with frontmatter containing `name`, `description`, and optional runtime-owned `metadata`.
- Framework bootstrap already exports exactly one runtime context: `data-runtime="browser"` in normal web sessions or `data-runtime="app"` in the packaged desktop runtime, plus the derived tag `runtime-browser` or `runtime-app`; packaged app routes derive that from the desktop bridge before falling back to launcher runtime info or frontend config.
- Import `/mod/_core/framework/js/context.js` when code needs to read the current live `<x-context>` set directly; use `getTags(...)` for tag collection, `getAttributeValues("data-runtime", ...)` for raw runtime reads, and the other helpers for broader context inspection.
- Use `<x-context>` tags in mounted DOM when a module needs to expose additional live skill-filter state such as `onscreen`, `admin`, `route:spaces`, or `space:open`.
- Use `metadata.when.tags` when the skill should exist only in those live contexts.
- Use `metadata.loaded` only when the skill should be auto-injected without an explicit `space.skills.load(...)` call.
- Use `metadata.placement: system` when the skill body is durable instruction, `metadata.placement: transient` when it should live in the mutable transient block, and let the default `history` placement stand only for ordinary non-auto-loaded skills that should behave like normal conversation context.
- Prompt-facing skill text is token-budgeted. Keep wording terse, avoid unnecessary markdown or filler, and measure before or after changes with the local tokenizer when you edit auto-loaded or catalog-facing skill text.
- When a skill needs reusable browser logic, prefer a small JS helper stored inside that skill's own folder and imported from a stable `/mod/<author>/<repo>/ext/skills/...` path instead of pasting a long inline script into `SKILL.md`.
- Keep the top-level router skill directive and concise.
- Keep nested skills focused on one stable area.
- Prefer exact file paths, runtime names, and examples over vague guidance.
- If a skill subtree becomes complex, add an `AGENTS.md` file inside that subtree and keep it current.

## Maintenance Rules

- When a mirrored source contract changes, update the affected skill files in the same session.
- When a stable feature or workflow changes, update the relevant docs under `/mod/_core/documentation/docs/` and the documentation skill at `/mod/_core/documentation/ext/skills/documentation/SKILL.md` in the same session.
- Do not let skill guidance drift away from the owning `AGENTS.md` files.
- For the development super-skill specifically, keep `/mod/_core/skillset/ext/skills/development/AGENTS.md` current whenever the framework, router, API, layer, or auth contracts it mirrors change.

讨论

基于 GitHub Discussions。登录 GitHub 即可参与讨论、点赞、订阅更新。