Commands and Skills

You work with Context through a handful of conversational entries. Everything else — the lifecycle CLI commands, the Indexers and the workflow's own step Skills — runs behind them, chosen by the current Route rather than typed by you.

flowchart LR
  you["You, in plain language"] --> entry["Entry Skill<br/>/c4a:context and others"]
  entry --> cli["context CLI<br/>current Route"]
  cli --> step["Workflow step Skills"]
  cli --> idx["Indexer Skills"]
  step --> cli
  idx --> cli

Entries you invoke

Entry	Use it to	Example
/c4a:context	Create a Context knowledge workspace, continue one, and run production, review, update and build	"Build reviewed knowledge for ../billing-service for on-call engineers."
/c4a:context-plan	Plan a large project or a broad update in stages before production	"Plan knowledge for these three repositories and the design folder."
/c4a:context-inspect-search	Ask questions against knowledge packages or approved knowledge, with sources	"How are refunds retried? Cite the code."
/c4a:context-repo-content	Register or edit documentation and authored Skills in the same repository, without starting production	"Register docs/ and .agents/skills as repository content."

/c4a:context is the production entry for new and existing knowledge workspaces. The other three entries do not drive production themselves: a plan, an answer that points to a gap, or registered repository content is handed back to /c4a:context when knowledge should be produced. A conversation where Context is already active continues without invoking it again; ordinary coding or design conversations never start it on their own.

To create a new Indexer, ask in plain language — "Create an Indexer Skill for our RPC framework in ./skills". The context-indexer-create Skill picks this up; it is not a menu command.

How each host spells them

Host	How entries appear
Claude Code	Plugin commands /c4a:context, /c4a:context-plan, /c4a:context-inspect-search, /c4a:context-repo-content
Cursor	Commands /c4a-context, /c4a-context-plan, /c4a-context-inspect-search, /c4a-context-repo-content
Codex	Skills invoked by name: context, context-plan, context-inspect-search, context-repo-content
Local install into one repository (context plugin install --local)	The same entries without the c4a plugin namespace

Restart or refresh the host after installing or updating the plugin so it loads the current entries. See Installation.

Skills that run on their own

Skill	When it runs
context-code-indexer, context-markdown-indexer, context-note-indexer, context-sessions-indexer	The Agent selects them for the current sources and requirements during planning and writing. Installing one makes it available; it is not enabled for every knowledge workspace. See Community Indexers.
context-inspect-search	Also when a host or Bot configuration routes knowledge questions to it.
Workflow step Skills: work-production-stage, translate-site-locales, record-workspace-version, the Indexer lifecycle steps, and recover-workspace	Each runs when the current Route selects its step — writing a production stage, translating the site, recording a version, or recovering a stuck task.

Indexer Skills are installed without a plugin namespace, so they keep the same names in every host.

Every Skill in the Agent plugin, with its trigger and hand-off, is listed in the Skills reference.

What you never need to type

The lifecycle commands — context source add, context review …, context build and the rest — are run by the Agent with the exact revision the Route returns. You will see them in progress messages, but you describe goals instead. There is no context continue command: when the Agent starts, context entry and context status observe the knowledge workspace and return the next step. The full command list is in the CLI Commands reference.

Requests you will use often

You want to	Say something like	Read more
Start	"Create a Context knowledge workspace in ./kb for …; readers are …"	Quick Start
Add material	"Add ../docs/refunds to the sources and re-plan the affected part."	Working with Agents
Fix a page	"Revise the retry page: the limit is five, not three."	Revising a Page
Follow a change	"The billing repository moved to a new release; update the affected pages."	Updating Knowledge
Hand off review	Add a CONTEXT_RUN_POLICY block to the task	Managed Mode and Review Policy

命令与技能

你通过少数几个对话入口使用 Context。其余一切——生命周期 CLI 命令、Indexer 以及工作流自身的步骤 Skill——都在这些入口背后运行，由当前 Route 选择，而不需要你手动输入。

flowchart LR
  you["你，使用自然语言"] --> entry["入口 Skill<br/>/c4a:context 等"]
  entry --> cli["context CLI<br/>当前 Route"]
  cli --> step["工作流步骤 Skill"]
  cli --> idx["Indexer Skill"]
  step --> cli
  idx --> cli

你调用的入口

入口	用途	示例
/c4a:context	创建或继续 Context 知识工作区，执行生产、审核、更新和构建	“为 ../billing-service 构建经过审核的知识，读者是值班工程师。”
/c4a:context-plan	在生产前分阶段规划大型项目或大范围更新	“为这三个仓库和设计文档目录规划知识。”
/c4a:context-inspect-search	基于知识包或已批准的知识提问，并附带来源	“退款是如何重试的？请引用代码。”
/c4a:context-repo-content	登记或编辑同一仓库中的文档和自编 Skill，不启动生产	“把 docs/ 和 .agents/skills 登记为仓库内容。”

/c4a:context 是新建和已有知识工作区的生产入口。其他三个入口自身不推进生产：需要生产知识时，规划结果、指出知识缺口的回答或已登记的仓库内容都会交回 /c4a:context。Context 已经处于活动状态的对话无需再次调用即可继续；普通的编码或设计对话永远不会自行启动它。

要创建新的 Indexer，用自然语言提出即可——“在 ./skills 中为我们的 RPC 框架创建一个 Indexer Skill”。context-indexer-create Skill 会接手这个请求；它不是菜单命令。

各宿主中的写法

宿主	入口形式
Claude Code	插件命令 /c4a:context、/c4a:context-plan、/c4a:context-inspect-search、/c4a:context-repo-content
Cursor	命令 /c4a-context、/c4a-context-plan、/c4a-context-inspect-search、/c4a-context-repo-content
Codex	按名称调用的 Skill：context、context-plan、context-inspect-search、context-repo-content
只安装到单个仓库（context plugin install --local）	同样的入口，但没有 c4a 插件命名空间

安装或更新插件后，请重启或刷新宿主，让它加载当前的入口。参见安装。

自行运行的 Skill

Skill	何时运行
context-code-indexer、context-markdown-indexer、context-note-indexer、context-sessions-indexer	在规划和写作期间，由 Agent 根据当前来源和需求选用。安装某个 Indexer 只是让它可用，并不会在每个知识工作区中启用。参见社区 Indexer。
context-inspect-search	当宿主或 Bot 配置把知识问题路由给它时，也会运行。
工作流步骤 Skill：work-production-stage、translate-site-locales、record-workspace-version、Indexer 生命周期步骤，以及 recover-workspace	当前 Route 选中对应步骤时运行——写作某个生产阶段、翻译站点、记录版本，或恢复卡住的任务。

Indexer Skill 安装时不带插件命名空间，因此在每个宿主中名称都相同。

Agent 插件中每个 Skill 的触发条件与交接关系，见技能清单。

你永远不需要输入的内容

生命周期命令——context source add、context review …、context build 等——都由 Agent 按 Route 返回的确切修订执行。你会在进度消息中看到它们，但你只需描述目标。没有 context continue 命令：Agent 启动时，context entry 和 context status 会观察知识工作区并返回下一步。完整的命令列表见 CLI 命令参考。

常用请求

你想要	可以这样说	延伸阅读
开始	“在 ./kb 中为……创建一个 Context 知识工作区；读者是……”	快速开始
添加材料	“把 ../docs/refunds 加入来源，并重新规划受影响的部分。”	与 Agent 协作
修正页面	“修订重试页面：上限是五次，不是三次。”	修订页面
跟进变更	“billing 仓库发布了新版本；更新受影响的页面。”	更新知识
移交审核	在任务中添加一个 CONTEXT_RUN_POLICY 块	托管模式与审核策略