iii-worker-catalog

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Worker Registry

The public worker surface is registry-first. This repo should not hardcode one static skill per worker because workers publish their own skills in

iii-hq/workers

and the registry. Use this skill to discover the right worker, install it, then load the worker-provided skill/docs before generating code.

公开的worker界面采用注册表优先的设计。本仓库不应为每个worker硬编码静态skill，因为workers会在

iii-hq/workers

和注册表中发布自己的skill。使用本skill来发现合适的worker，安装它，然后在生成代码前加载worker提供的skill/文档。

Flow

流程

Infer the needed capability from the task: agent harness, shell tools, sandboxing, database, storage, MCP, ACP, image transforms, etc.
Query the registry or local directory before naming a worker.
Install the selected worker with
```
iii worker add
```
.
Sync and verify the lockfile.
Load the worker's own skill markdown from
```
iii-directory
```
or the registry before using worker-specific functions, trigger schemas, config, or policy details.

从任务中推断所需能力：agent harness、shell工具、沙箱、数据库、存储、MCP、ACP、图像转换等。
在指定worker之前，先查询注册表或本地目录。
使用
```
iii worker add
```
命令安装选中的worker。
同步并验证锁文件。
在使用worker专属函数、触发器schema、配置或策略细节之前，先从
```
iii-directory
```
或注册表中加载worker自身的skill markdown文档。

Registry Discovery

注册表发现

Use the CLI when working locally:

bash

iii worker add iii-directory
iii worker list
iii worker status iii-directory --no-watch

Use

iii-directory

from an agent or worker:

typescript

const candidates = await iii.trigger({
  function_id: 'directory::registry::workers::list',
  payload: { search: 'harness' },
})

const info = await iii.trigger({
  function_id: 'directory::registry::workers::info',
  payload: { name: 'harness' },
})

Use live engine discovery after installation:

typescript

await iii.trigger({
  function_id: 'directory::engine::workers::list',
  payload: {},
})

await iii.trigger({
  function_id: 'directory::engine::functions::list',
  payload: { search: 'session' },
})

本地工作时使用CLI：

bash

iii worker add iii-directory
iii worker list
iii worker status iii-directory --no-watch

从agent或worker中使用

iii-directory

：

typescript

const candidates = await iii.trigger({
  function_id: 'directory::registry::workers::list',
  payload: { search: 'harness' },
})

const info = await iii.trigger({
  function_id: 'directory::registry::workers::info',
  payload: { name: 'harness' },
})

安装后使用实时引擎发现：

typescript

await iii.trigger({
  function_id: 'directory::engine::workers::list',
  payload: {},
})

await iii.trigger({
  function_id: 'directory::engine::functions::list',
  payload: { search: 'session' },
})

Install

安装

Install the worker the registry returns:

bash

iii worker add <worker>
iii worker sync
iii worker verify
iii worker start <worker>

Use

iii worker add

, not

cargo build

npm install

, or

pip install

, for normal users installing published workers. Source builds are contributor workflows.

安装注册表返回的worker：

bash

iii worker add <worker>
iii worker sync
iii worker verify
iii worker start <worker>

对于安装已发布workers的普通用户，请使用

iii worker add

，而非

cargo build

、

npm install

或

pip install

。源码构建是贡献者的工作流程。

Worker Skills

After installing

iii-directory

, download or read worker-provided skill markdown instead of relying on this repo to know each worker's API:

typescript

await iii.trigger({
  function_id: 'directory::skills::download',
  payload: { worker: 'harness', tag: 'latest' },
})

const index = await iii.trigger({
  function_id: 'directory::skills::index',
  payload: {},
})

const skill = await iii.trigger({
  function_id: 'directory::skills::get',
  payload: { id: 'harness/index.md' },
})

directory::skills::get

accepts

id

<id>.md

, or

iii://<id>

. The worker skill is the source of truth for function IDs, trigger config, call payloads, installation dependencies, examples, and security notes.

安装

iii-directory

后，下载或读取worker提供的skill markdown文档，而非依赖本仓库了解每个worker的API：

typescript

await iii.trigger({
  function_id: 'directory::skills::download',
  payload: { worker: 'harness', tag: 'latest' },
})

const index = await iii.trigger({
  function_id: 'directory::skills::index',
  payload: {},
})

const skill = await iii.trigger({
  function_id: 'directory::skills::get',
  payload: { id: 'harness/index.md' },
})

directory::skills::get

接受

id

、

<id>.md

或

iii://<id>

格式的参数。worker skill是函数ID、触发器配置、调用负载、安装依赖、示例和安全说明的唯一可信来源。

Choosing Workers

选择Workers

Examples of capability-to-registry searches:

Agent runtime or durable chat: search
```
harness
```
, then read the harness worker skill.
Host commands or filesystem tools: search
```
shell
```
, then read the shell worker skill.
Isolation for untrusted code: search
```
sandbox
```
, then read the sandbox worker skill.
SQL access: search
```
database
```
, then read the database worker skill.
Object storage: search
```
storage
```
, then read the storage worker skill.
External protocol bridges: search
```
mcp
```
or
```
acp
```
, then read that worker's skill.

Do not encode these worker APIs here. New workers can appear without a change to this repo.

能力到注册表搜索的示例：

Agent运行时或持久化聊天：搜索
```
harness
```
，然后查看harness worker的skill文档。
主机命令或文件系统工具：搜索
```
shell
```
，然后查看shell worker的skill文档。
不可信代码隔离：搜索
```
sandbox
```
，然后查看sandbox worker的skill文档。
SQL访问：搜索
```
database
```
，然后查看database worker的skill文档。
对象存储：搜索
```
storage
```
，然后查看storage worker的skill文档。
外部协议桥：搜索
```
mcp
```
或
```
acp
```
，然后查看对应worker的skill文档。

请勿在此处编码这些worker的API。新workers无需修改本仓库即可添加。

Troubleshooting

故障排查

Use

iii worker list

iii worker status <name> --no-watch

, and

iii worker logs <name>

for lifecycle debugging.

Use

directory::engine::workers::info

directory::engine::functions::list

, and

directory::engine::triggers::list

to inspect what is actually connected.

Use
```
iii worker sync --frozen
```
in CI and
```
iii worker verify
```
before committing worker config changes.
If a function is missing, confirm the worker is installed, synced, started, and visible through engine discovery before changing code.

使用

iii worker list

、

iii worker status <name> --no-watch

和

iii worker logs <name>

进行生命周期调试。

使用

directory::engine::workers::info

、

directory::engine::functions::list

和

directory::engine::triggers::list

检查实际连接的内容。

在CI环境中使用
```
iii worker sync --frozen
```
，并在提交worker配置更改前使用
```
iii worker verify
```
。
如果某个函数缺失，请先确认worker已安装、同步、启动并可通过引擎发现，再修改代码。

When to Use

使用场景

Use this skill when choosing, installing, discovering, or troubleshooting any published iii worker.
Use it when the task asks for a worker-backed capability and the exact worker/API should come from the registry or worker-provided skill.

当你需要选择、安装、发现或排查任何已发布的iii worker时，使用本skill。
当任务需要由worker支持的能力，且具体的worker/API应来自注册表或worker提供的skill文档时，使用本skill。

Boundaries

边界限制

Do not tell users to install published workers by building from source.
Do not keep one static local skill per worker in this repo.
Do not invent worker names, function IDs, trigger schemas, or payloads; check the registry and worker skill first.
Do not generate removed service APIs, adapter-extension APIs, or removed stream APIs.

不要告知用户通过源码构建来安装已发布的workers。
不要在本仓库中为每个worker保留静态本地skill文档。
不要自行创建worker名称、函数ID、触发器schema或负载；请先检查注册表和worker skill文档。
不要生成已移除的服务API、适配器扩展API或已移除的流API。