Settings Snippet Keys

Settings Snippet Keys

This page tells maintainers why each key of the snippet is set. It gives each key and env entry of the Settings snippet. Each row gives the value, what the key does in Claude Code, and why dotclaude sets it. The snippet sets a key only when the Claude Code default does not give the result that dotclaude needs.

The snippet merges the settings entries of the slices in src/plugins/dotclaude/slices/. The prompt slice sets most keys and the env entries. The trim slice sets the env entries that disable the built-in parts, and the keys autoMemoryEnabled to syncClaudeAiPlugins. The statusline slice sets statusLine and subagentStatusLine. The styles slice sets outputStyle. A test checks the keys of the snippet against schemas/claude-code-settings.json. So a key that the schema lacks fails the test.

Environment Entries

Entry	Value	What it does in Claude Code	Why dotclaude sets it
CLAUDE_CODE_AUTO_COMPACT_WINDOW	300000	Sets the window that automatic compaction works against. Default: the window of the model.	The window that /context shows and warns against.
CLAUDE_CODE_AUTO_CONNECT_IDE	false	Default: the IDE connects by itself.	The global config key autoConnectIde is not set by the snippet, so the variable does the work.
CLAUDE_CODE_DISABLE_ADVISOR_TOOL	1	Removes the advisor tool and its instructions. Default: the tool can be on.	Each tool line and its instructions add tokens to each request.
CLAUDE_CODE_DISABLE_CFC_PROMPT	1	Removes the Chrome part of the system prompt.	The part is not used.
CLAUDE_CODE_DISABLE_CRON	1	Removes the Cron tools and /loop. Default: on.	No scheduled runs.
CLAUDE_CODE_DISABLE_EXPLORE_PLAN_AGENTS	1	Removes the built-in agents Explore and Plan. Default: on.	Each agent adds a line to each request.
CLAUDE_CODE_DISABLE_FAST_MODE	1	Keeps fast mode off. Default: fast mode can be on.	Fast mode stays off.
CLAUDE_CODE_DISABLE_FEEDBACK_SURVEY	1	Stops the session quality survey. Default: the survey can show.	No survey in a session.
CLAUDE_CODE_DISABLE_TERMINAL_TITLE	1	Stops the terminal title and its small-model request. Default: on.	No title and no request for one.
CLAUDE_CODE_FORK_SUBAGENT	0	Default: a subagent forks the main context.	A subagent starts with its own prompt.
CLAUDE_CODE_IDE_SKIP_AUTO_INSTALL	1	Stops the IDE extension from installing itself. Default: it installs.	No install without the user.
CLAUDE_CODE_MAX_CONCURRENT_SUBAGENTS	5	Caps the subagents that run in parallel. Default: no cap.	A hard cap on parallel subagents.
CLAUDE_CODE_MAX_CONTEXT_TOKENS	300000	Sets the context window of a session. Default: the window of the model.	The window that dotclaude sets for a session.
CLAUDE_CODE_SIMPLE_SYSTEM_PROMPT	1	Uses the short system prompt. Default: the long prompt.	The dotclaude:Concise style carries the tone and the format, and the hooks module puts the working rules in the system prompt.
CLAUDE_CODE_SUBAGENT_MODEL	claude-sonnet-5-5	Sets the model of a subagent that has no model of its own. Default: the main model.	A subagent with no model of its own runs on Sonnet 5.5.
CLAUDE_CODE_TODO_REMINDER_MODE	off	Default: a recurring reminder about the task list.	No recurring reminder.
DISABLE_COMPACT	1	Disables compaction and /compact. Default: compaction is on.	No compaction.
DISABLE_ERROR_REPORTING	1	Default: error reports are sent.	No error reports.
DISABLE_FEEDBACK_COMMAND	1	Removes the /feedback command. Default: the command is there.	No /feedback command.

The snippet never sets CLAUDE_CODE_EFFORT_LEVEL, CLAUDE_CODE_SUBAGENT_MODEL_FORCE, or CLAUDE_CODE_SUBAGENT_PROMPT_CACHE_TTL. The first two defeat the model and effort of the agent files (Decisions: router and agents). The third overrides the subagentPromptCacheTtl setting, which the snippet does not set either. See Off switches.

Settings Keys

Key	Value	What it does in Claude Code	Why dotclaude sets it
$schema	The URL of schemas/claude-code-settings.json on the main branch	Points an editor at the JSON Schema of the settings.	An editor can check the keys of the file.
outputStyle	dotclaude:Concise	Selects the output style for assistant responses. The style is selected, not forced, so the outputStyle of the user can change it. Default: the style of the user.	The styles slice ships the concise style. It keeps the coding instructions, because the prompt slice changes the coding sections.
model	claude-opus-5-5	Sets the main model. Default: the model of the plan.	The main model.
availableModels	Fable 5.1, Opus 5.5, Sonnet 5.5, and Haiku 5.5	Lists the models that the user can select. Default: all models.	claude-haiku-4-5 is not in the roster. For a CVP account see the CVP procedure.
enforceAvailableModels	true	When true, Default resolves to a model of availableModels.	Default stays on a model of the list.
maxEffortLevel	max	Sets the highest effort level, and anything above it is clamped. Default: unset.	No cap in practice. Each model keeps its effort, and dotclaude never rewrites it.
enableArtifact	false	Disables the Artifact tool. Project and local settings can only disable it. Default: on when the feature is available.	Removes the Artifact tool from each request.
enableWorkflows	false	Disables the Workflows feature. Default: by plan.	Removes the Workflow tool from each request.
promptCacheTtl	5m	Sets the cache lifetime of the main conversation. Default: automatic, 1 hour on a subscription within its limits.	The lifetime is the same on each account.
disableBundledSkills	true	Removes the bundled skills and workflows. Built-in slash commands stay typable. Default: the skills are on.	Removes the bundled skills from the listing.
statusLine	type command, the command bun --no-env-file "<plugin folder>/bin/dotclaude-statusline", padding 0, refreshInterval 1, and hideVimModeIndicator true	Sets the status line command, and hides the built-in vim mode indicator. Default: unset.	The three-column status line. refreshInterval keeps the countdowns and the blinking glyph current in an idle session. The line shows vim.mode, so the built-in indicator would repeat it. See Status line.
subagentStatusLine	The command bun --no-env-file "<plugin folder>/bin/dotclaude-statusline" --subagent	Sets the status line of each subagent. Default: the default rows.	One row for each running subagent.
autoCompactEnabled	false	Disables automatic compaction. Default: on.	No automatic compaction.
autoMemoryEnabled	false	Stops the auto-memory directory of the project from being read or written. Default: on.	No auto memory.
includeGitInstructions	false	Excludes the built-in commit and PR workflow text from the system prompt. Default: included.	The git text adds tokens to each request, and attribution gives the commit line.
promptSuggestionEnabled	false	Disables prompt suggestions. Default: on.	No background request for suggestions.
awaySummaryEnabled	false	Disables the session recap that shows after 5 minutes away. Default: on.	No background request for a recap.
bashOutputMaxChars	4000	Sets how many characters of a successful Bash output go inline. Output past this goes to a file. Default: 30000, and values clamp to 4000 to 128000.	The lowest value. The bound BASH_OUTPUT_MAX_CHARS in src/kernel/budget.ts holds the same value.
spinnerTipsEnabled	false	Hides the tips in the spinner. Default: shown.	No tips.
feedbackSurveyRate	0	Sets the chance that the quality survey shows, from 0 to 1. Default: unset.	No quality survey.
feedbackDrafts	off	Disables the SendFeedback tool. Default: notify.	Removes the tool.
fastMode	false	Disables fast mode. Default: off.	Fast mode stays off, with the env entry CLAUDE_CODE_DISABLE_FAST_MODE.
disableAgentView	true	Disables the agent view, claude agents, --bg, and the background daemon. Default: false.	No agent view and no background daemon.
disableRemoteControl	true	Disables Remote Control. Default: false.	No Remote Control.
syncClaudeAiSkills	false	Stops the sync of the skills that you enable on claude.ai. Read in user or managed settings only. Default: on when the account has synced skills.	No anthropic-skills:* skill.
syncClaudeAiPlugins	false	Stops the sync of the plugins that you enable on claude.ai. Read in user or managed settings only. Default: on.	No plugin sync from claude.ai.
disableClaudeAiConnectors	true	Stops the claude.ai MCP connectors from being fetched. A true in any settings source wins. Default: false.	No claude.ai connector server.
enabledPlugins	Nine built-in plugins set to false	Enables or disables each plugin, by its plugin-id@marketplace-id. Default: each plugin on by its own default.	See Off switches.
skillOverrides	plugin-authoring set to name-only	Asks Claude Code to list the skill by name only, not re-checked on 2.1.296 (Off switches). Default: unset.	A shorter listing line for plugin-authoring.
attribution	The commit and PR lines	Sets the attribution text of commits and PRs. Default: the Claude Code lines.	The attribution that the commit and PR text carries.
sandbox	On, a retry outside the sandbox allowed, a list of allowed domains	Limits the files and the network of Bash. Default: off.	The sandbox limits Bash. git fetch, git pull, git push, and gh run outside it, and they still go through the permission flow.
permissions	disableBypassPermissionsMode set to disable, and the deny, ask, and allow lists	Sets the rules for each tool call. deny blocks, ask asks, and allow runs without a prompt. Default: unset.	Risky commands ask, and read-only commands run. The deny list holds the groups below.

The deny list of the snippet holds these groups.

Denied items	Reason
Read of secret files, rm -rf of /, /*, ~, and ~/*, mkfs, and dd to a device	Secrets and disk wipes stay out of reach of a tool call.
EnterPlanMode and ExitPlanMode	Removes two tool lines. The eval of 2026-10-08 showed no loss.
TaskCreate, TaskGet, TaskList, TaskStop, and TaskUpdate	Removes five tool lines. The eval of 2026-10-08 showed no loss.
Agent(general-purpose)	Removes an agent line, and a run of this agent replaced the dotclaude agents (Usage evidence).
Agent(claude-code-guide) and Agent(statusline-setup)	Removes two agent lines from each request.
mcp__computer-use	A guard for the case that you enable the opt-in server (Computer use).
ScheduleWakeup	Removes the tool. It has no setting (Prompt surface).
ReportFindings	Removes the tool. /code-review gives its findings as text without it (Prompt surface).

The status line entries need the absolute path of the plugin folder. A status line command receives no ${CLAUDE_PLUGIN_ROOT}.

Auto Mode Lists

The snippet does not hold the autoMode lists. Setup writes them to the user settings, because Claude Code reads autoMode only from there.

• allow: $defaults, and two entries for the worktree and the temp folder.
• soft_deny: $defaults, and the ask rules of the catalog.
• hard_deny: $defaults, and the floor and subagent rules of the catalog.
• environment: $defaults only.

Each list starts with $defaults, which takes the built-in list of Claude Code. The rules are in The destructive floor.

Keys That the Snippet Does Not Set

• cleanupPeriodDays: see Retention.
• enableAllProjectMcpServers: the value false equals the default.
• effortLevel: each model keeps its default.
• The keys of the never-set list and the items without a gate: see Off switches.
• managed-settings.json: dotclaude writes no managed file, see Off switches.

Mythos Detection

Setup puts claude-mythos-5-1 in place of claude-fable-5-1 in availableModels when the model catalog of your account offers it. Claude Code 2.1.296 caches the catalog that the API serves to the account in <config dir>/cache/model-catalog/<organizationUuid>-<hash>-<surface>.json. The organization comes from oauthAccount.organizationUuid in .claude.json. The bundle of 2.1.296 has its own check for a Mythos row of the catalog. The row is not disabled, and it is not in the deprecated section. Setup makes the same test. This was found by reading the bundle, and no doc covers this file. A missing file, another shape, or a new release that moves the file gives no Mythos, and setup keeps Fable. A list that already holds claude-mythos-5-1 keeps it. This read needs no model call, so it replaces the probe below for setup.

CVP Procedure

An account in the Cyber Verification Program (CVP) can call claude-mythos-5-1 in place of claude-fable-5-1. An answer alone grants nothing, so a probe call must show that Mythos 5.1 serves the account.

1. Run the probe. --settings allows Mythos 5.1 for this call, whatever your settings hold.

   claude -p "Reply with the one word: ok" \
     --model claude-mythos-5-1 \
     --settings '{"availableModels":["claude-mythos-5-1"]}' \
     --output-format json

2. Read the output. The call is served by Mythos 5.1 only when the exit code is 0 and modelUsage has one key, claude-mythos-5-1.

3. If so, put claude-mythos-5-1 in availableModels in place of claude-fable-5-1. If not, keep claude-fable-5-1.

A call for a model outside availableModels falls back to another model with no error. So is_error is not a test, and modelUsage is. A live probe on 2026-10-08 exited with 0 and the result ok, but modelUsage named only claude-opus-5-5. That account was not served by Mythos 5.1, so Fable 5.1 stays.

Retention

The snippet sets no cleanupPeriodDays, so the Claude Code default of 30 days applies. To keep transcripts longer, set a larger value in your own settings, such as 400. To keep none, start Claude Code with --no-session-persistence.

Issue 41458

Issue #41458 of Claude Code reports that a large cleanupPeriodDays value was not honored. So Claude Code can remove a transcript although the setting asks for a long retention. Copy each transcript that you need from ~/.claude/projects/ to a safe place. dotclaude does not remove a transcript. It does not copy one either.

Related Pages

• Settings Snippet
• Off Switches