The whole hunk
from line 15, old and new numbered
/
lines
from line 15
1515Extensions plug into different parts of the agentic loop:
1616
1717* **[CLAUDE.md](/docs/en/memory)** adds persistent context Claude sees every session
18* **[Output styles](/docs/en/output-styles)** set Claude's role, tone, and response format for every response in a session
1819* **[Skills](/docs/en/skills)** add reusable knowledge and invocable workflows
1920* **[Code intelligence](/docs/en/tools-reference#lsp-tool-behavior)** connects Claude to a language server for symbol-level navigation and live type errors
2021* **[MCP](/docs/en/mcp)** connects Claude to external services and tools
from line 31
3031
3132Features range from always-on context that Claude sees every session, to on-demand capabilities you or Claude can invoke, to background automation that runs on specific events. The table below shows what's available and when each one makes sense.
3233
33| Feature | What it does | When to use it | Example |
34| -------------------------------------------------------------- | ---------------------------------------------------------------------------- | ----------------------------------------------------------------------------- | ------------------------------------------------------------------------------------ |
35| **CLAUDE.md** | Persistent context loaded every conversation | Project conventions, "always do X" rules | "Use pnpm, not npm. Run tests before committing." |
36| **Skill** | Instructions, knowledge, and workflows Claude can use | Reusable content, reference docs, repeatable tasks | `/deploy` runs your deployment checklist; API docs skill with endpoint patterns |
37| **Subagent** | Isolated execution context that returns summarized results | Context isolation, parallel tasks, specialized workers | Research task that reads many files but returns only key findings |
38| **[Dynamic workflow](/docs/en/workflows)** | Script Claude writes that runs many subagents in the background | Work that outgrows a handful of subagents, or findings you want cross-checked | Audit a whole codebase, with a second set of agents verifying each finding |
39| **[Cross-session messaging](/docs/en/cross-session-messaging)** | Claude delivers a message from one of your sessions to another | Sessions you run yourself that need each other's findings mid-task | One session warns another that a change it made breaks what the other is building on |
40| **[Code intelligence](/docs/en/tools-reference#lsp-tool-behavior)** | Language-server navigation and diagnostics | Typed languages, large codebases where grep is slow or imprecise | Jump to a symbol's definition instead of reading the whole file |
41| **MCP** | Connect to external services | External data or actions | Query your database, post to Slack, control a browser |
42| **Hook** | Script, HTTP request, MCP tool call, prompt, or subagent triggered by events | Automation that must run on every matching event | Run ESLint after every file edit |
43| **[Artifact](/docs/en/artifacts)** | Publish session output as a private, interactive web page | Output you want to see or share visually rather than as terminal text | An incident timeline that updates as Claude investigates |
34| Feature | What it does | When to use it | Example |
35| -------------------------------------------------------------- | ---------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------- |
36| **CLAUDE.md** | Persistent context loaded every conversation | Project conventions, "always do X" rules | "Use pnpm, not npm. Run tests before committing." |
37| **[Output style](/docs/en/output-styles)** | Instructions that set Claude's role, tone, and response format for a whole session | A voice, length, or format you want in every response, or Claude working as something other than a software engineer | The built-in Concise style for shorter responses; a custom style that answers every question with a diagram first |
38| **Skill** | Instructions, knowledge, and workflows Claude can use | Reusable content, reference docs, repeatable tasks | `/deploy` runs your deployment checklist; API docs skill with endpoint patterns |
39| **Subagent** | Isolated execution context that returns summarized results | Context isolation, parallel tasks, specialized workers | Research task that reads many files but returns only key findings |
40| **[Dynamic workflow](/docs/en/workflows)** | Script Claude writes that runs many subagents in the background | Work that outgrows a handful of subagents, or findings you want cross-checked | Audit a whole codebase, with a second set of agents verifying each finding |
41| **[Cross-session messaging](/docs/en/cross-session-messaging)** | Claude delivers a message from one of your sessions to another | Sessions you run yourself that need each other's findings mid-task | One session warns another that a change it made breaks what the other is building on |
42| **[Code intelligence](/docs/en/tools-reference#lsp-tool-behavior)** | Language-server navigation and diagnostics | Typed languages, large codebases where grep is slow or imprecise | Jump to a symbol's definition instead of reading the whole file |
43| **MCP** | Connect to external services | External data or actions | Query your database, post to Slack, control a browser |
44| **Hook** | Script, HTTP request, MCP tool call, prompt, or subagent triggered by events | Automation that must run on every matching event | Run ESLint after every file edit |
45| **[Artifact](/docs/en/artifacts)** | Publish session output as a private, interactive web page | Output you want to see or share visually rather than as terminal text | An incident timeline that updates as Claude investigates |
4446
4547**[Plugins](/docs/en/plugins)** are the packaging layer. A plugin bundles skills, hooks, subagents, and MCP servers into a single installable unit. Plugin skills are namespaced (like `/my-plugin:review`) so multiple plugins can coexist. Use plugins when you want to reuse the same setup across multiple repositories or distribute to others via a **[marketplace](/docs/en/plugin-marketplaces)**.
4648
from line 53
5153| Trigger | Add |
5254| :------------------------------------------------------------------------------- | :--------------------------------------------------------------------------------------------- |
5355| Claude gets a convention or command wrong twice | Add it to [CLAUDE.md](/docs/en/memory) |
56| You keep asking Claude to be shorter, explain more, or answer in the same format | Set an [output style](/docs/en/output-styles) |
5457| You keep typing the same prompt to start a task | Save it as a user-invocable [skill](/docs/en/skills) |
5558| You paste the same playbook or multi-step procedure into chat for the third time | Capture it as a [skill](/docs/en/skills) |
5659| You keep copying data from a browser tab Claude can't see | Connect that system as an [MCP server](/docs/en/mcp) |
from line 106
103106 **Rule of thumb:** Keep CLAUDE.md under 200 lines. If it's growing, move reference content to skills or split into [`.claude/rules/`](/docs/en/memory#organize-rules-with-claude/rules/) files.
104107 </Tab>
105108
109 <Tab title="CLAUDE.md vs Output style">
110 Both give Claude standing instructions. CLAUDE.md carries what Claude should know, and an output style sets how Claude responds.
111
112 | Aspect | CLAUDE.md | Output style |
113 | ------------- | ----------------------------------------------- | --------------------------------------------------------------------------------------------------- |
114 | **Holds** | Facts and rules about your project | A role, tone, and response format |
115 | **Switching** | Always loaded | One active at a time; [switch styles](/docs/en/output-styles#change-your-output-style) whenever you want |
116 | **Best for** | Build commands, conventions, "never do X" rules | Shorter responses, explanations alongside code, a non-engineering role |
117
118 **Put it in CLAUDE.md** if it's true of the project whatever style you're in: coding conventions, build commands, project structure.
119
120 **Use an output style** if it's about the response itself and you might want it off again: length, format, how much Claude explains, or a different role such as a writing assistant. Claude Code includes [built-in styles](/docs/en/output-styles#built-in-output-styles), and you can write your own.
121
122 **They combine.** CLAUDE.md stays loaded whichever style you pick. Claude follows both as instructions, so neither is enforced. For anything that has to happen every time, use a [hook](/docs/en/hooks-guide).
123 </Tab>
124
106125 <Tab title="CLAUDE.md vs Rules vs Skills">
107126 All three store instructions, but they load differently:
108127
from line 218
199218
200219Each feature has a different loading strategy and context cost:
201220
202| Feature | When it loads | What loads | Context cost |
203| --------------------- | ------------------------------ | -------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------- |
204| **CLAUDE.md** | Session start | Full content | Every request |
205| **Skills** | Session start + when used | Descriptions at start, full content when used | Low (descriptions every request)\* |
206| **MCP servers** | Session start | Tool names; full schemas on demand | Low until a tool is used |
207| **Code intelligence** | After file edits and on demand | Diagnostics after edits; symbol locations on lookup | Low; reduces file reads elsewhere |
208| **Subagents** | When spawned | Fresh context with specified skills, or the parent conversation for a [fork](/docs/en/sub-agents#fork-the-current-conversation) | Isolated from main session |
209| **Hooks** | On trigger | Nothing (runs externally) | Zero, unless hook returns additional context |
221| Feature | When it loads | What loads | Context cost |
222| --------------------- | ----------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------- |
223| **CLAUDE.md** | Session start | Full content | Every request |
224| **Output styles** | Session start, and again when you switch styles | The active style's full instructions; nothing for the Default style | Every request |
225| **Skills** | Session start + when used | Descriptions at start, full content when used | Low (descriptions every request)\* |
226| **MCP servers** | Session start | Tool names; full schemas on demand | Low until a tool is used |
227| **Code intelligence** | After file edits and on demand | Diagnostics after edits; symbol locations on lookup | Low; reduces file reads elsewhere |
228| **Subagents** | When spawned | Fresh context with specified skills, or the parent conversation for a [fork](/docs/en/sub-agents#fork-the-current-conversation) | Isolated from main session |
229| **Hooks** | On trigger | Nothing (runs externally) | Zero, unless hook returns additional context |
210230
211231\*By default, skill descriptions load at session start so Claude can decide when to use them. Set `disable-model-invocation: true` in a skill's frontmatter to hide it from Claude entirely until you invoke it manually. For a skill you didn't write, set [`skillOverrides`](/docs/en/skills#override-skill-visibility-from-settings) in settings to do the same without editing its file.
212232