Claude Code

Claude Code has the richest configuration surface of any supported tool. It supports all 8 entity types and all scope tiers (enterprise, project, and user) — the only exception is local settings files, which are intentionally excluded from dotai management.

Supported features

• Conditional rules with intelligent selection
• Skills (all 11 frontmatter fields)
• Sub-agents (all 14+ frontmatter fields)
• MCP servers (with headers and OAuth support)
• Lifecycle hooks (17 of 20 events, with command/prompt/agent types)
• Per-tool permissions with allow, deny, and ask decisions
• Enterprise, project, and user scope

Generated files

dotai source	Generated file
Rules (alwaysApply: true, project scope)	CLAUDE.md
Rules (local scope)	CLAUDE.local.md
Rules (alwaysApply: false or appliesTo set)	.claude/rules/*.md
Rules with outputDir	<outputDir>/CLAUDE.md, <outputDir>/.claude/rules/*.md
Skills	.claude/skills/<name>/SKILL.md
Agents	.claude/agents/*.md
Servers	.mcp.json
Hooks	.claude/settings.json (hooks)
Permissions	.claude/settings.json (permissions)
Ignore patterns	.claude/settings.json (deny rules)

Entity details

Rules

Rule output depends on scope and activation mode:

• Project scope + alwaysApply: true — concatenated into CLAUDE.md at the project root (or <outputDir>/CLAUDE.md if outputDir is set).
• Local scope — concatenated into CLAUDE.local.md (not checked into version control).
• Project scope + appliesTo globs — written as individual .claude/rules/<name>.md files with a YAML frontmatter block containing paths: for intelligent selection.
• Project scope + alwaysApply: false (no appliesTo) — written to .claude/rules/ but loads unconditionally (a warning is emitted since this has no effect vs alwaysApply: true).

The override field only affects Codex output (emitting to AGENTS.override.md); it has no effect on Claude Code.

Agents

Each agent is written as a markdown file under .claude/agents/<name>.md with YAML frontmatter. Claude Code supports all agent frontmatter fields:

Frontmatter field	Source field	Notes
name	name	Agent identifier
description	description	When to delegate to this agent
model	model	Translated via alias map (e.g. claude-sonnet-4-6 → sonnet)
modelReasoningEffort	modelReasoningEffort	low, medium, or high
tools	tools	Comma-separated list of allowed tool names
disallowedTools	disallowedTools or readonly	readonly: true maps to Write, Edit, NotebookEdit
permissionMode	permissionMode	default, acceptEdits, dontAsk, bypassPermissions, plan
maxTurns	maxTurns	Maximum agentic turns
skills	skills	Comma-separated list of skill names
memory	memory	user, project, or local
background	background	true to run in background
isolation	isolation	worktree for isolated git worktree
hooks	hooks	Serialized as nested YAML in frontmatter
mcpServers	mcpServers	Serialized as nested YAML in frontmatter

The model field uses an alias map: claude-sonnet-4-6 and claude-sonnet-4-5 become sonnet, claude-opus-4-6 and claude-opus-4-5 become opus, claude-haiku-4-5 variants become haiku.

Skills

Each skill is written to .claude/skills/<name>/SKILL.md with frontmatter. Claude Code supports all skill frontmatter fields:

Frontmatter field	Source field	Notes
name	name	Skill identifier (used for /skill-name invocation)
description	description	When to auto-invoke this skill
disable-model-invocation	disableAutoInvocation	true to prevent automatic invocation
argument-hint	argumentHint	Hint shown in slash command menu (e.g. <file-path>)
user-invocable	userInvocable	false to hide from user-facing slash commands
allowed-tools	allowedTools	Comma-separated list (e.g. Read, Grep, Glob)
model	model	Model override for skill execution
context	context	fork to run in isolated context
agent	agent	Agent to delegate to when invoked
hooks	hooks	Serialized as nested YAML in frontmatter

Note that allowed-tools uses comma-separated format, not array bracket syntax.

MCP servers

MCP servers are written to .mcp.json under the mcpServers key. Each entry is keyed by server name:

• stdio transport — command, args, and env fields.
• http transport — type: "http", url, and optionally headers and oauth fields.
• sse transport — type: "sse", url, headers, oauth. A deprecation warning is emitted recommending migration to HTTP (Streamable HTTP).

{
  "mcpServers": {
    "my-api": {
      "type": "http",
      "url": "https://api.example.com/mcp",
      "headers": {
        "Authorization": "Bearer ${API_TOKEN}"
      },
      "oauth": {
        "clientId": "my-app",
        "callbackPort": 8080
      }
    }
  }
}

Hooks

Claude Code supports 17 of the 20 lifecycle events. Events are translated from dotai’s camelCase to Claude Code’s PascalCase names:

dotai event	Claude Code event
preToolUse	PreToolUse
postToolUse	PostToolUse
sessionStart	SessionStart
sessionEnd	SessionEnd
userPromptSubmitted	UserPromptSubmit
agentStop	Stop
subagentStop	SubagentStop
permissionRequest	PermissionRequest
postToolUseFailure	PostToolUseFailure
notification	Notification
subagentStart	SubagentStart
teammateIdle	TeammateIdle
taskCompleted	TaskCompleted
configChange	ConfigChange
worktreeCreate	WorktreeCreate
worktreeRemove	WorktreeRemove
preCompact	PreCompact

Events not supported by Claude Code: preFileEdit, postFileEdit, errorOccurred.

Hooks are written to .claude/settings.json under the hooks key. Claude Code supports all three hook types (command, prompt, agent) and all optional fields (timeout, statusMessage, once, async, model).

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Bash",
        "hooks": [
          {
            "type": "prompt",
            "prompt": "Review this command for safety.",
            "model": "haiku",
            "timeout": 10000
          }
        ]
      }
    ]
  }
}

Permissions

Permissions are written to .claude/settings.json under the permissions key. Rules can be scoped per tool and include argument pattern matching. Each rule carries an allow, deny, or ask decision. Ignore patterns are expressed as deny rules on the Read and Edit tools with glob path matchers.