The System Prompt

The system prompt is the text that Claude Code sends to the model before your first message. It tells the model how to behave and what the session looks like. You can add to it, replace it, or switch to a shorter version. This page explains how the prompt is built. The exact captured texts are on the Captured System Prompts page.

Ways to Change It

Sources: CLI Reference, Output Styles, and Environment Variables.

Item	Status	What it does
--system-prompt <text>	Documented: An official docs page names this item.	Replaces the whole default prompt.
--system-prompt-file <file>	Documented: An official docs page names this item.	Replaces the prompt with the text of a file. Hidden in claude --help.
--append-system-prompt <text>	Documented: An official docs page names this item.	Adds text to the end of the prompt.
--append-system-prompt-file <file>	Documented: An official docs page names this item.	Adds the text of a file to the end. Hidden in claude --help.
--append-subagent-system-prompt, --append-subagent-system-prompt-file	Documented: An official docs page names this item.	Add text to every subagent prompt, in print mode only. Hidden in claude --help.
--system-prompt-snapshot on|off	Documented: An official docs page names this item.	on records the prompt on the first request and reuses it. off rebuilds it for every request. Needs v2.1.257 or later.
--exclude-dynamic-system-prompt-sections	Documented: An official docs page names this item.	Moves the working folder, the environment, the memory paths, and the git status into the first user message.
__SYSTEM_PROMPT_DYNAMIC_BOUNDARY__	Documented: An official docs page names this item.	A line in replacement text. Text above it stays cached. Text below it can change. Needs v2.1.275 or later.
CLAUDE_CODE_SIMPLE_SYSTEM_PROMPT	Documented: An official docs page names this item.	0 chooses the full prompt. 1 chooses the short prompt. See Short or Full Prompt.
CLAUDE_CODE_SIMPLE and --bare	Documented: An official docs page names this item.	Use a minimal prompt. See Bare Mode.
Output style	Documented: An official docs page names this item.	Adds style text. The frontmatter key keep-coding-instructions keeps the coding section.
prompt.section and prompt.compose	Documented: An official docs page names this item. Differs: The official docs and the program disagree. The row says how. prompt.section does not fire for shared sections such as lean_body	Mod events that rewrite one section or the whole list. See the Mod API pages in Extend.

How Claude Code Builds It

Claude Code makes a list of sections. Each section has a name and a text, or no text when its condition is false. A shared section is the same for every user of a model. A session section can change with the session. Claude Code sends the shared sections first, so that the shared part can stay in the cache.

The request has three system blocks: the attribution line, the identity line, and the body. See Request and Cache for the cache markers.

Short or Full Prompt

A model gets the short prompt (one lean_body section, about 5.7k characters) or the full prompt (six longer sections, about 26.9k characters). The official docs and the real request captures agree here. Only the old wiki was wrong. Captures of 2.1.311 show the following. See Captured System Prompts for the texts.

Case	Result	Status
Opus 5.5, Sonnet 5.5 by default	The short prompt.	Documented: An official docs page names this item.
Haiku 5.5 by default	The short prompt plus 5 extra paragraphs (about 2.8k characters).	Undocumented: Only the program shows this item. It can change without notice.
CLAUDE_CODE_SIMPLE_SYSTEM_PROMPT=0	The full prompt. This is the only way to get it on these models.	Documented: An official docs page names this item.
CLAUDE_CODE_SIMPLE_SYSTEM_PROMPT=1	The short prompt. On the three 5.5 models it changes nothing, because they already get it.	Documented: An official docs page names this item.

Names That Claude Code Does Not Recognize

Since 2.1.312, a model name that Claude Code does not recognize gets the short prompt, and CLAUDE_CODE_SIMPLE_SYSTEM_PROMPT=0 keeps the full prompt. This covers custom deployment names and inference profile ARNs. The test in the program has no check for the provider. The program decides in this order:

1. CLAUDE_CODE_SIMPLE_SYSTEM_PROMPT set to a true value gives the short prompt. A false value, such as 0, gives the full prompt.
2. If the variable is unset, an unknown name gives the short prompt. A known name has claude-3- or haiku or sonnet in it, or is claude-opus-4-0, -4-1, -4-5, -4-6, or -4-7. A server-side catalog flag lean_prompt for the model comes first.
3. For a known name, the gate tengu_velvet_tide gives the short prompt when on. Otherwise the server setting simple_system_prompt for the model decides.

The program now canonicalizes the name for this test with the option deterministic. The option skips the lookup that turns an application-inference-profile ARN into the model behind it. So an ARN no longer counts as the Sonnet or Haiku model that it fronts, unless modelOverrides maps it. Source: the program (chunk-9wgj206m.js lines 334 to 363, chunk-3xe0k172.js lines 24698 to 24717, and chunk-rww18cjm.js of 2.1.311, where the same test used no deterministic option).

Prompt by Model

We ran the default prompt for every model in the program's catalog, plus names that end in -eap and an unknown name (see Probes). The stub accepted every name. The real API may refuse retired models.

The models with the catalog flag lean_prompt get the short prompt: Haiku 5.5, Sonnet 5.5, Opus 4.8, Opus 5, Opus 5.5, Fable 5, Fable 5.1, and Mythos 5.1. claude-mythos-5 has no such flag. A name rule gives it the short prompt. The same rule applies to a name that ends in -eap and to an unknown name. Older models, and claude-sonnet-5, get the full prompt. Some models add sections to the short prompt. These sections follow the flags opus_5_prompt_bundle, opus_5_5_prompt_bundle, fable_5_mitigations, and fable_5_1_prompt_bundle.

Model	Prompt	System Characters
claude-opus-5-5, claude-sonnet-5-5, and their -eap names	Short	5,699
claude-haiku-5-5	Short plus 5 paragraphs on effort and turns	8,491
claude-opus-4-8	Short, with one different bullet in a system reminder	5,641
claude-opus-5	Short plus "Delivering work" and "Corrections" sections and a line to not use agents or workflows unless asked	9,079
claude-fable-5	Short plus "Communicating with the user", a comment rule, an identity paragraph, and an autonomy block	9,992
claude-fable-5-1	Short plus "Delivering work", "Writing for the user", an identity paragraph, and an autonomy block	11,771
claude-mythos-5	Short plus "Communicating with the user" and an autonomy block	9,414
claude-mythos-5-1	Like Fable 5.1	11,084
Unknown name	Short, without the total_tokens line	5,648
claude-sonnet-5	Full, with one different TaskCreate bullet	26,944
claude-haiku-4-5, claude-3-5-haiku, claude-3-5-sonnet, claude-3-7-sonnet, claude-sonnet-4-0, claude-sonnet-4-5, claude-sonnet-4-6, claude-opus-4-5, claude-opus-4-6, claude-opus-4-7	Full	27,046
claude-opus-4-0, claude-opus-4-1	Claude Code silently changes the request model to claude-opus-5-5, so they get the short prompt	5,699

The 3.5 and 3.7 Haiku and Sonnet models also get 25 tools and not 12, because they have no tool search. In 2.1.309 it was 24, before Monitor joined the tool list (probe print.models.matrix). The captures cover only Opus 5.5, Sonnet 5.5, Haiku 5.5, and the full prompt. The other models have no capture page.

The official docs say that Haiku 4.5, Sonnet 5, Opus 4.7, and earlier models get the full prompt, and newer models get the short one. The captures match this for Sonnet 5.5, Opus 5.5, and Haiku 5.5. The old wiki said Sonnet gets the full prompt. That is wrong.

Captured texts: lean prompt, Sonnet 5.5, Haiku 5.5, and the full prompt.

Where Each Part Goes in the Request

Not everything the model reads as instructions is in system. The captures show this split.

Part	Where it goes	Status
Billing header line	system[0]	Undocumented: Only the program shows this item. It can change without notice.
Identity line	system[1]	Undocumented: Only the program shows this item. It can change without notice.
Rules, memory section, appended text	system[2] (the body)	Undocumented: Only the program shows this item. It can change without notice.
Environment details, skills list, agent types, output style text, the date	A system-role entry in messages[]	Undocumented: Only the program shows this item. It can change without notice.
Git status, CLAUDE.md files, the git attribution reminder	The first user message	Undocumented: Only the program shows this item. It can change without notice.

The identity line is You are a Claude agent, built on Anthropic's Claude Agent SDK. in print mode. With --append-system-prompt, print mode sends You are Claude Code, Anthropic's official CLI for Claude, running within the Claude Agent SDK. The interactive line is You are Claude Code, Anthropic's official CLI for Claude. The program picks it when the session is interactive. It also picks it for Google Cloud Agent Platform (Vertex) sessions.

What Each Flag Does in the Captures

Flag or setting	Captured effect	Status
--system-prompt, --system-prompt-file, --agent	Replace only the body (system[2]). The identity line stays.	Documented: An official docs page names this item.
--append-system-prompt	Adds text to the end of the body and changes the identity line.	Documented: An official docs page names this item.
--exclude-dynamic-system-prompt-sections	Moves Memory into the first user message. Drops Session-specific guidance and the fast-mode sentence.	Documented: An official docs page names this item.
Output style with keep-coding-instructions: false	Removes "Doing tasks" from the full prompt only. It does nothing to the short prompt.	Documented: An official docs page names this item.
--permission-mode plan	Adds about 4.8k characters to the environment message.	Documented: An official docs page names this item.
--permission-mode acceptEdits, dontAsk, --brief	Do not change the text.	Undocumented: Only the program shows this item. It can change without notice.

Sections

No official docs page lists the sections, so every row below is

Undocumented: Only the program shows this item. It can change without notice.

unless the row says otherwise. The names matter because of this. A prompt.compose hook sees all of them. A prompt.section hook sees only the session sections.

Shared Sections

These come first. Claude Code uses either the short row or the full rows. From 2.1.302 to 2.1.307, no shared section (lean_body, intro, tone, doing_tasks, memory_static, and the others) fires prompt.section, so a prompt.section hook does not see them. Only a prompt.compose hook sees them.

Section	Status	When	Content
lean_body	Undocumented: Only the program shows this item. It can change without notice.	Short prompt	A list of harness rules: markdown output, permission modes, hooks as user feedback, and dedicated tools first.
intro	Undocumented: Only the program shows this item. It can change without notice.	Full prompt	Who the agent is and whom it works for.
system	Undocumented: Only the program shows this item. It can change without notice.	Full prompt	Output is shown to the user, tool results can hold prompt injection, and old messages get compressed.
doing_tasks	Undocumented: Only the program shows this item. It can change without notice.	Full prompt, and the output style is none or has keep-coding-instructions: true	Do only what the task needs, add no extra features, and write few comments.
actions	Undocumented: Only the program shows this item. It can change without notice.	Full prompt	Confirm actions that are hard to undo.
tools	Undocumented: Only the program shows this item. It can change without notice.	Full prompt	Use dedicated tools before Bash, and call independent tools in parallel.
tone	Undocumented: Only the program shows this item. It can change without notice.	Full prompt	No emojis, short answers, and file_path:line_number references.
memory_static	Undocumented: Only the program shows this item. It can change without notice.	Only with --exclude-dynamic-system-prompt-sections	The memory text, moved to the shared part.

Session Sections

They follow in this order. Claude Code omits a section that has no text. Session sections, such as memory, do fire prompt.section.

Section	Status	When	Content
communication	Undocumented: Only the program shows this item. It can change without notice.	Always	How to write for the user, with text that depends on the model.
pronouns	Undocumented: Only the program shows this item. It can change without notice.	Always	Use they/them when pronouns are not stated.
action_caution	Undocumented: Only the program shows this item. It can change without notice.	Short prompt	Confirm actions that reach outside the session or are hard to undo.
fable_identity	Undocumented: Only the program shows this item. It can change without notice.	A Fable model, or the model in ANTHROPIC_DEFAULT_FABLE_MODEL	A paragraph about Fable 5.1 or Fable 5 and Mythos.
tool_param_json	Gated: Works only behind a flag or a server gate. An experiment, or a tool strictness setting	An experiment, or a tool strictness setting	One sentence: object and array values are one JSON value.
session_guidance	Undocumented: Only the program shows this item. It can change without notice.	When any of its items applies	The ! <command> tip, the Agent tool and forks, skills by /name, and ultrareview.
memory	Undocumented: Only the program shows this item. It can change without notice.	Not with --exclude-dynamic-system-prompt-sections	Instructions for auto memory.
surface:<kind>, system_messages	Gated: Works only behind a flag or a server gate. An experiment form of the prompt	An experiment form of the prompt	Where Claude Code runs, such as desktop, vscode, web, or terminal.
env_info_simple	Undocumented: Only the program shows this item. It can change without notice.	Normally	The latest model IDs, where Claude Code runs, and a note about /fast.
env_info_static	Undocumented: Only the program shows this item. It can change without notice.	With --exclude-dynamic-system-prompt-sections	The same text, without data of the machine.
bg-session	Gated: Works only behind a flag or a server gate. CLAUDE_CODE_SESSION_KIND=bg and CLAUDE_JOB_DIR set	CLAUDE_CODE_SESSION_KIND=bg and CLAUDE_JOB_DIR set	Rules for background jobs: use $CLAUDE_JOB_DIR/tmp, use EnterWorktree, and end with a report.
context_management	Undocumented: Only the program shows this item. It can change without notice.	Always	Old context gets summarized, so do not finish early.
brief	Undocumented: Only the program shows this item. It can change without notice.	--brief, or the SendUserMessage tool	Rules for that tool.
focus_mode	Undocumented: Only the program shows this item. It can change without notice.	The view mode is focus	The user sees only the final message.
act_dont_rederive	Gated: Works only behind a flag or a server gate. CLAUDE_CODE_ACT_DONT_REDERIVE, or an experiment that is on by default	CLAUDE_CODE_ACT_DONT_REDERIVE, or an experiment that is on by default	Act when you have enough information.
delivering_work_max	Gated: Works only behind a flag or a server gate. CLAUDE_CODE_BISON_CAIRN, or an experiment for some models	CLAUDE_CODE_BISON_CAIRN, or an experiment for some models	Do the full scope of the work.
overcorrection	Gated: Works only behind a flag or a server gate. CLAUDE_CODE_LARCH_CISTERN, or an experiment	CLAUDE_CODE_LARCH_CISTERN, or an experiment	Correct yourself only when it matters.
subagent_steer_delegation	Gated: Works only behind a flag or a server gate. The Agent tool exists and an experiment arm is on	The Agent tool exists and an experiment arm is on	Steers when to delegate.
opus5_reduced_delegation	Gated: Works only behind a flag or a server gate. Opus 5 prompt bundle, with an experiment that is on by default	Opus 5 prompt bundle, with an experiment that is on by default	Do not use the Agent tool, workflows, or deep research unless asked.
heron_brook, brook_heron	Gated: Works only behind a flag or a server gate. An experiment, or Haiku 5.5 early-stopping guidance	An experiment, or Haiku 5.5 early-stopping guidance	Text that the server sets.
willow_tern	Gated: Works only behind a flag or a server gate. CLAUDE_CODE_WILLOW_TERN, or an experiment	CLAUDE_CODE_WILLOW_TERN, or an experiment	Rules for writing to the user.
autonomy_append	Gated: Works only behind a flag or a server gate. Mythos 5, a Fable 5 mitigation model, or CLAUDE_CODE_AMBER_ASTROLABE	Mythos 5, a Fable 5 mitigation model, or CLAUDE_CODE_AMBER_ASTROLABE	Work without asking in the middle of a task.
endconv_deferred_hint	Gated: Works only behind a flag or a server gate. The EndConversation tool exists	The EndConversation tool exists	A hint for that tool.
total_tokens	Gated: Works only behind a flag or a server gate. Mode comes from CLAUDE_CODE_TOTAL_TOKENS_REMINDER	Not with CLAUDE_CODE_DISABLE_ATTACHMENTS or CLAUDE_CODE_SIMPLE, and the mode comes from CLAUDE_CODE_TOTAL_TOKENS_REMINDER	A reminder of the token budget.
user_pays	Gated: Works only behind a flag or a server gate. An experiment that is off by default	An experiment that is off by default	The user pays for each token.

Experiments come from GrowthBook, a flag service that Anthropic runs. You cannot set an experiment. But the env vars in the table switch some of the sections.

Bare Mode

--bare and CLAUDE_CODE_SIMPLE=1 give the same result: a two-line body (CWD: <path> and Date: <date>), 3 tools, and an empty environment message. Source: Run Claude Code Programmatically: Start Faster with Bare Mode for the flag. The body comes from the captures (

Undocumented: Only the program shows this item. It can change without notice.

). See the captured bare mode prompt and simple mode env var.

Subagent Prompt

A subagent gets its own agent text plus short notes. The default agent text starts with You are an agent for Claude Code, Anthropic's official CLI for Claude. The notes say to share absolute paths, to avoid emojis, to avoid a colon before tool calls, and not to write report files. Source: the program (

Undocumented: Only the program shows this item. It can change without notice.

). See the captured prompts for general-purpose, explore, plan, and custom subagents.

Example

You want the model to follow one more rule. You also want a mod to remove the memory section.

claude --append-system-prompt "Always use TypeScript"

Claude Code adds the text to the end of the default prompt. A mod can hook prompt.section with { name: "memory" } and return { text: null } to drop the memory section.

Related Pages

• Captured System Prompts
• Request and Cache
• Models
• Output Styles and Mod API pages in Configure and Extend.