Claude Code Architecture Study
Introduction: This project is a learning and research repository focused on CLI Agent architecture. All materials are compiled entirely from publicly available online references and discussions, with a particular focus on public information regarding the highly popular CLI Agent
claude-code. Our intention is to help developers better understand and utilize Agent technologies. We will continue to share more insights and practical discussions on Agent architecture in the future. Thank you for your support!
Disclaimer: All content in this repository is provided strictly for technical research, study, and educational exchange among enthusiasts. Commercial use is strictly prohibited. No individual, organization, or entity may use this content for commercial purposes, profit-making activities, illegal activities, or any other unauthorized scenarios. If any content infringes upon your legal rights, intellectual property, or other interests, please contact us and we will verify and remove it immediately.
Language: English | 中文 | 한국어 | 日本語
Table of Contents
- Deep Analysis Reports (
docs/) — Telemetry, codenames, undercover mode, remote control, future roadmap - Directory Reference — Code structure tree
- Architecture Overview — Entry → Query Engine → Tools/Services/State
- Tool System & Permissions — 40+ tools, permission flow, sub-agents
- The 12 Progressive Harness Mechanisms — How Claude Code layers production features on the agent loop
Deep Analysis Reports (docs/)
Deep analysis reports compiled from publicly available online references and community discussions on Claude Code v2.1.88. Quadrilingual (EN/JA/KO/ZH).
docs/
├── en/ # English
│ ├── [01-telemetry-and-privacy.md] # Telemetry & Privacy — what's collected, why you can't opt out
│ ├── [02-hidden-features-and-codenames.md] # Codenames (Capybara/Tengu/Numbat), feature flags, internal vs external
│ ├── [03-undercover-mode.md] # Undercover Mode — hiding AI authorship in open-source repos
│ ├── [04-remote-control-and-killswitches.md]# Remote Control — managed settings, killswitches, model overrides
│ └── [05-future-roadmap.md] # Future Roadmap — Numbat, KAIROS, voice mode, unreleased tools
│
├── ja/ # 日本語
│ ├── [01-テレメトリとプライバシー.md] # テレメトリとプライバシー — 収集項目、無効化不可の理由
│ ├── [02-隠し機能とコードネーム.md] # 隠し機能 — モデルコードネーム、feature flag、内部/外部ユーザーの違い
│ ├── [03-アンダーカバーモード.md] # アンダーカバーモード — オープンソースでのAI著作隠匿
│ ├── [04-リモート制御とキルスイッチ.md] # リモート制御 — 管理設定、キルスイッチ、モデルオーバーライド
│ └── [05-今後のロードマップ.md] # 今後のロードマップ — Numbat、KAIROS、音声モード、未公開ツール
│
├── ko/ # 한국어
│ ├── [01-텔레메트리와-프라이버시.md] # 텔레메트리 및 프라이버시 — 수집 항목, 비활성화 불가 이유
│ ├── [02-숨겨진-기능과-코드네임.md] # 숨겨진 기능 — 모델 코드네임, feature flag, 내부/외부 사용자 차이
│ ├── [03-언더커버-모드.md] # 언더커버 모드 — 오픈소스에서 AI 저작 은폐
│ ├── [04-원격-제어와-킬스위치.md] # 원격 제어 — 관리 설정, 킬스위치, 모델 오버라이드
│ └── [05-향후-로드맵.md] # 향후 로드맵 — Numbat, KAIROS, 음성 모드, 미공개 도구
│
└── zh/ # 中文
├── [01-遥测与隐私分析.md] # 遥测与隐私 — 收集了什么，为什么无法退出
├── [02-隐藏功能与模型代号.md] # 隐藏功能 — 模型代号，feature flag，内外用户差异
├── [03-卧底模式分析.md] # 卧底模式 — 在开源项目中隐藏 AI 身份
├── [04-远程控制与紧急开关.md] # 远程控制 — 托管设置，紧急开关，模型覆盖
└── [05-未来路线图.md] # 未来路线图 — Numbat，KAIROS，语音模式，未上线工具
Click any filename above to jump to the full report.
| # | Topic | Key Findings |
|---|---|---|
| 01 | Telemetry & Privacy | Two analytics sinks (1P, Datadog). Environment fingerprint, process metrics, repo hash on every event. No UI-exposed opt-out for 1st-party logging. OTEL_LOG_TOOL_DETAILS=1 enables full tool input capture. |
| 02 | Hidden Features & Codenames | Animal codenames (Capybara v8, Tengu, Fennec→Opus 4.6, Numbat next). Feature flags use random word pairs (tengu_frond_boric) to obscure purpose. Internal users get better prompts, verification agents, and effort anchors. Hidden commands: /btw, /stickers. |
| 03 | Undercover Mode | Official employees auto-enter undercover mode in public repos. Model instructed: "Do not blow your cover" — strip all AI attribution, write commits "as a human developer would." No force-OFF exists. Raises transparency questions for open-source communities. |
| 04 | Remote Control | Hourly polling of /api/claude_code/settings. Dangerous changes show blocking dialog — reject = app exits. 6+ killswitches (bypass permissions, fast mode, voice mode, analytics sink). GrowthBook flags can change any user's behavior without consent. |
| 05 | Future Roadmap | Numbat codename confirmed. Opus 4.7 / Sonnet 4.8 in development. KAIROS = fully autonomous agent mode with <tick> heartbeats, push notifications, PR subscriptions. Voice mode (push-to-talk) ready but gated. 17 unreleased tools found. |
Copyright & Disclaimer
This repository is provided strictly for technical research and educational purposes.
Commercial use is strictly prohibited.
If you are the copyright owner and believe this repository content infringes your rights,
please contact the repository owner for immediate removal.
Stats
| Item | Count |
|---|---|
| Files (.ts/.tsx) | ~1,884 |
| Lines | ~512,664 |
| Largest single file | query.ts (~785KB) |
| Built-in tools | ~40+ |
| Slash commands | ~80+ |
| Dependencies (node_modules) | ~192 packages |
| Runtime | Bun (compiled to Node.js >= 18 bundle) |
The Agent Pattern
THE CORE LOOP
=============
User --> messages[] --> Claude API --> response
|
stop_reason == "tool_use"?
/ \
yes no
| |
execute tools return text
append tool_result
loop back -----------------> messages[]
That is the minimal agent loop. Claude Code wraps this loop
with a production-grade harness: permissions, streaming,
concurrency, compaction, sub-agents, persistence, and MCP.
Directory Reference
src/
├── main.tsx # REPL bootstrap, 4,683 lines
├── QueryEngine.ts # SDK/headless query lifecycle engine
├── query.ts # Main agent loop (785KB, largest file)
├── Tool.ts # Tool interface + buildTool factory
├── Task.ts # Task types, IDs, state base
├── tools.ts # Tool registry, presets, filtering
├── commands.ts # Slash command definitions
├── context.ts # User input context
├── cost-tracker.ts # API cost accumulation
├── setup.ts # First-run setup flow
│
├── bridge/ # Claude Desktop / remote bridge
│ ├── bridgeMain.ts # Session lifecycle manager
│ ├── bridgeApi.ts # HTTP client
│ ├── bridgeConfig.ts # Connection config
│ ├── bridgeMessaging.ts # Message relay
│ ├── sessionRunner.ts # Process spawning
│ ├── jwtUtils.ts # JWT refresh
│ ├── workSecret.ts # Auth tokens
│ └── capacityWake.ts # Capacity-based wakeup
│
├── cli/ # CLI infrastructure
│ ├── handlers/ # Command handlers
│ └── transports/ # I/O transports (stdio, structured)
│
├── commands/ # ~80 slash commands
│ ├── agents/ # Agent management
│ ├── compact/ # Context compaction
│ ├── config/ # Settings management
│ ├── help/ # Help display
│ ├── login/ # Authentication
│ ├── mcp/ # MCP server management
│ ├── memory/ # Memory system
│ ├── plan/ # Plan mode
│ ├── resume/ # Session resume
│ ├── review/ # Code review
│ └── ... # 70+ more commands
│
├── components/ # React/Ink terminal UI
│ ├── design-system/ # Reusable UI primitives
│ ├── messages/ # Message rendering
│ ├── permissions/ # Permission dialogs
│ ├── PromptInput/ # Input field + suggestions
│ ├── LogoV2/ # Branding + welcome screen
│ ├── Settings/ # Settings panels
│ ├── Spinner.tsx # Loading indicators
│ └── ... # 40+ component groups
│
├── entrypoints/ # Application entry points
│ ├── cli.tsx # CLI main (version, help, daemon)
│ ├── sdk/ # Agent SDK (types, sessions)
│ └── mcp.ts # MCP server entry
│
├── hooks/ # React hooks
│ ├── useCanUseTool.tsx # Permission checking
│ ├── useReplBridge.tsx # Bridge connection
│ ├── notifs/ # Notification hooks
│ └── toolPermission/ # Tool permission handlers
│
├── services/ # Business logic layer
│ ├── api/ # Claude API client
│ │ ├── claude.ts # Streaming API calls
│ │ ├── errors.ts # Error categorization
│ │ └── withRetry.ts # Retry logic
│ ├── analytics/ # Telemetry + GrowthBook
│ ├── compact/ # Context compression
│ ├── mcp/ # MCP connection management
│ ├── tools/ # Tool execution engine
│ │ ├── StreamingToolExecutor.ts # Parallel tool runner
│ │ └── toolOrchestration.ts # Batch orchestration
│ ├── plugins/ # Plugin loader
│ └── settingsSync/ # Cross-device settings
│
├── state/ # Application state
│ ├── AppStateStore.ts # Store definition
│ └── AppState.tsx # React provider + hooks
│
├── tasks/ # Task implementations
│ ├── LocalShellTask/ # Bash command execution
│ ├── LocalAgentTask/ # Sub-agent execution
│ ├── RemoteAgentTask/ # Remote agent via bridge
│ ├── InProcessTeammateTask/ # In-process teammate
│ └── DreamTask/ # Background thinking
│
├── tools/ # 40+ tool implementations
│ ├── AgentTool/ # Sub-agent spawning + fork
│ ├── BashTool/ # Shell command execution
│ ├── FileReadTool/ # File reading (PDF, image, etc)
│ ├── FileEditTool/ # String-replace editing
│ ├── FileWriteTool/ # Full file creation
│ ├── GlobTool/ # File pattern search
│ ├── GrepTool/ # Content search (ripgrep)
│ ├── WebFetchTool/ # HTTP fetching
│ ├── WebSearchTool/ # Web search
│ ├── MCPTool/ # MCP tool wrapper
│ ├── SkillTool/ # Skill invocation
│ ├── AskUserQuestionTool/ # User interaction
│ └── ... # 30+ more tools
│
├── types/ # Type definitions
│ ├── message.ts # Message discriminated unions
│ ├── permissions.ts # Permission types
│ ├── tools.ts # Tool progress types
│ └── ids.ts # Branded ID types
│
├── utils/ # Utilities (largest directory)
│ ├── permissions/ # Permission rule engine
│ ├── messages/ # Message formatting
│ ├── model/ # Model selection logic
│ ├── settings/ # Settings management
│ ├── sandbox/ # Sandbox runtime adapter
│ ├── hooks/ # Hook execution
│ ├── memory/ # Memory system utils
│ ├── git/ # Git operations
│ ├── github/ # GitHub API
│ ├── bash/ # Bash execution helpers
│ ├── swarm/ # Multi-agent swarm
│ ├── telemetry/ # Telemetry reporting
│ └── ... # 30+ more util groups
│
└── vendor/ # Native module source stubs
├── audio-capture-src/ # Audio input
├── image-processor-src/ # Image processing
├── modifiers-napi-src/ # Native modifiers
└── url-handler-src/ # URL handling
Architecture Overview
┌─────────────────────────────────────────────────────────────────────┐
│ ENTRY LAYER │
│ cli.tsx ──> main.tsx ──> REPL.tsx (interactive) │
│ └──> QueryEngine.ts (headless/SDK) │
└──────────────────────────────┬──────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────────┐
│ QUERY ENGINE │
│ submitMessage(prompt) ──> AsyncGenerator<SDKMessage> │
│ │ │
│ ├── fetchSystemPromptParts() ──> assemble system prompt │
│ ├── processUserInput() ──> handle /commands │
│ ├── query() ──> main agent loop │
│ │ ├── StreamingToolExecutor ──> parallel tool execution │
│ │ ├── autoCompact() ──> context compression │
│ │ └── runTools() ──> tool orchestration │
│ └── yield SDKMessage ──> stream to consumer │
└──────────────────────────────┬──────────────────────────────────────┘
│
┌────────────────┼────────────────┐
▼ ▼ ▼
┌──────────────────┐ ┌─────────────────┐ ┌──────────────────┐
│ TOOL SYSTEM │ │ SERVICE LAYER │ │ STATE LAYER │
│ │ │ │ │ │
│ Tool Interface │ │ api/claude.ts │ │ AppState Store │
│ ├─ call() │ │ API client │ │ ├─ permissions │
│ ├─ validate() │ │ compact/ │ │ ├─ fileHistory │
│ ├─ checkPerms() │ │ auto-compact │ │ ├─ agents │
│ ├─ render() │ │ mcp/ │ │ └─ fastMode │
│ └─ prompt() │ │ MCP protocol │ │ │
│ │ │ analytics/ │ │ React Context │
│ 40+ Built-in: │ │ telemetry │ │ ├─ useAppState │
│ ├─ BashTool │ │ tools/ │ │ └─ useSetState │
│ ├─ FileRead │ │ executor │ │ │
│ ├─ FileEdit │ │ plugins/ │ └──────────────────┘
│ ├─ Glob/Grep │ │ loader │
│ ├─ AgentTool │ │ settingsSync/ │
│ ├─ WebFetch │ │ cross-device │
│ └─ MCPTool │ │ oauth/ │
│ │ │ auth flow │
└──────────────────┘ └─────────────────┘
│ │
▼ ▼
┌──────────────────┐ ┌─────────────────┐
│ TASK SYSTEM │ │ BRIDGE LAYER │
│ │ │ │
│ Task Types: │ │ bridgeMain.ts │
│ ├─ local_bash │ │ session mgmt │
│ ├─ local_agent │ │ bridgeApi.ts │
│ ├─ remote_agent │ │ HTTP client │
│ ├─ in_process │ │ workSecret.ts │
│ ├─ dream │ │ auth tokens │
│ └─ workflow │ │ sessionRunner │
│ │ │ process spawn │
│ ID: prefix+8chr │ └─────────────────┘
│ b=bash a=agent │
│ r=remote t=team │
└──────────────────┘
Data Flow: A Single Query Lifecycle
USER INPUT (prompt / slash command)
│
▼
processUserInput() ← parse /commands, build UserMessage
│
▼
fetchSystemPromptParts() ← tools → prompt sections, CLAUDE.md memory
│
▼
recordTranscript() ← persist user message to disk (JSONL)
│
▼
┌─→ normalizeMessagesForAPI() ← strip UI-only fields, compact if needed
│ │
│ ▼
│ Claude API (streaming) ← POST /v1/messages with tools + system prompt
│ │
│ ▼
│ stream events ← message_start → content_block_delta → message_stop
│ │
│ ├─ text block ──────────────→ yield to consumer (SDK / REPL)
│ │
│ └─ tool_use block?
│ │
│ ▼
│ StreamingToolExecutor ← partition: concurrent-safe vs serial
│ │
│ ▼
│ canUseTool() ← permission check (hooks + rules + UI prompt)
│ │
│ ├─ DENY ────────────────→ append tool_result(error), continue loop
│ │
│ └─ ALLOW
│ │
│ ▼
│ tool.call() ← execute the tool (Bash, Read, Edit, etc.)
│ │
│ ▼
│ append tool_result ← push to messages[], recordTranscript()
│ │
└─────────┘ ← loop back to API call
│
▼ (stop_reason != "tool_use")
yield result message ← final text, usage, cost, session_id
Tool System Architecture
TOOL INTERFACE
==============
buildTool(definition) ──> Tool<Input, Output, Progress>
Every tool implements:
┌────────────────────────────────────────────────────────┐
│ LIFECYCLE │
│ ├── validateInput() → reject bad args early │
│ ├── checkPermissions() → tool-specific authz │
│ └── call() → execute and return result │
│ │
│ CAPABILITIES │
│ ├── isEnabled() → feature gate check │
│ ├── isConcurrencySafe() → can run in parallel? │
│ ├── isReadOnly() → no side effects? │
│ ├── isDestructive() → irreversible ops? │
│ └── interruptBehavior() → cancel or block on user? │
│ │
│ RENDERING (React/Ink) │
│ ├── renderToolUseMessage() → input display │
│ ├── renderToolResultMessage() → output display │
│ ├── renderToolUseProgressMessage() → spinner/status │
│ └── renderGroupedToolUse() → parallel tool groups │
│ │
│ AI FACING │
│ ├── prompt() → tool description for LLM │
│ ├── description() → dynamic description │
│ └── mapToolResultToAPI() → format for API response │
└────────────────────────────────────────────────────────┘
Complete Tool Inventory
FILE OPERATIONS SEARCH & DISCOVERY EXECUTION
═════════════════ ══════════════════════ ══════════
FileReadTool GlobTool BashTool
FileEditTool GrepTool PowerShellTool
FileWriteTool ToolSearchTool
NotebookEditTool INTERACTION
═══════════
WEB & NETWORK AGENT / TASK AskUserQuestionTool
════════════════ ══════════════════ BriefTool
WebFetchTool AgentTool
WebSearchTool SendMessageTool PLANNING & WORKFLOW
TeamCreateTool ════════════════════
MCP PROTOCOL TeamDeleteTool EnterPlanModeTool
══════════════ TaskCreateTool ExitPlanModeTool
MCPTool TaskGetTool EnterWorktreeTool
ListMcpResourcesTool TaskUpdateTool ExitWorktreeTool
ReadMcpResourceTool TaskListTool TodoWriteTool
TaskStopTool
TaskOutputTool SYSTEM
════════
SKILLS & EXTENSIONS ConfigTool
═════════════════════ SkillTool
SkillTool ScheduleCronTool
LSPTool SleepTool
TungstenTool
Permission System
TOOL CALL REQUEST
│
▼
┌─ validateInput() ──────────────────────────────────┐
│ reject invalid inputs before any permission check │
└────────────────────┬───────────────────────────────┘
│
▼
┌─ PreToolUse Hooks ─────────────────────────────────┐
│ user-defined shell commands (settings.json hooks) │
│ can: approve, deny, or modify input │
└────────────────────┬───────────────────────────────┘
│
▼
┌─ Permission Rules ─────────────────────────────────┐
│ alwaysAllowRules: match tool name/pattern → auto │
│ alwaysDenyRules: match tool name/pattern → deny │
│ alwaysAskRules: match tool name/pattern → ask │
│ Sources: settings, CLI args, session decisions │
└────────────────────┬───────────────────────────────┘
│
no rule match?
│
▼
┌─ Interactive Prompt ───────────────────────────────┐
│ User sees tool name + input │
│ Options: Allow Once / Allow Always / Deny │
└────────────────────┬───────────────────────────────┘
│
▼
┌─ checkPermissions() ───────────────────────────────┐
│ Tool-specific logic (e.g. path sandboxing) │
└────────────────────┬───────────────────────────────┘
│
APPROVED → tool.call()
Sub-Agent & Multi-Agent Architecture
MAIN AGENT
==========
│
┌───────────────┼───────────────┐
▼ ▼ ▼
┌──────────────┐ ┌──────────┐ ┌──────────────┐
│ FORK AGENT │ │ REMOTE │ │ IN-PROCESS │
│ │ │ AGENT │ │ TEAMMATE │
│ Fork process │ │ Bridge │ │ Same process │
│ Shared cache │ │ session │ │ Async context│
│ Fresh msgs[] │ │ Isolated │ │ Shared state │
└──────────────┘ └──────────┘ └──────────────┘
SPAWN MODES:
├─ default → in-process, shared conversation
├─ fork → child process, fresh messages[], shared file cache
├─ worktree → isolated git worktree + fork
└─ remote → bridge to Claude Code Remote / container
COMMUNICATION:
├─ SendMessageTool → agent-to-agent messages
├─ TaskCreate/Update → shared task board
└─ TeamCreate/Delete → team lifecycle management
SWARM MODE (feature-gated):
┌─────────────────────────────────────────────┐
│ Lead Agent │
│ ├── Teammate A ──> claims Task 1 │
│ ├── Teammate B ──> claims Task 2 │
│ └── Teammate C ──> claims Task 3 │
│ │
│ Shared: task board, message inbox │
│ Isolated: messages[], file cache, cwd │
└─────────────────────────────────────────────┘
Context Management (Compact System)
CONTEXT WINDOW BUDGET
═════════════════════
┌─────────────────────────────────────────────────────┐
│ System Prompt (tools, permissions, CLAUDE.md) │
│ ══════════════════════════════════════════════ │
│ │
│ Conversation History │
│ ┌─────────────────────────────────────────────┐ │
│ │ [compacted summary of older messages] │ │
│ │ ═══════════════════════════════════════════ │ │
│ │ [compact_boundary marker] │ │
│ │ ─────────────────────────────────────────── │ │
│ │ [recent messages — full fidelity] │ │
│ │ user → assistant → tool_use → tool_result │ │
│ └─────────────────────────────────────────────┘ │
│ │
│ Current Turn (user + assistant response) │
└─────────────────────────────────────────────────────┘
THREE COMPRESSION STRATEGIES:
├─ autoCompact → triggers when token count exceeds threshold
│ summarizes old messages via a compact API call
├─ snipCompact → removes zombie messages and stale markers
│ (HISTORY_SNIP feature flag)
└─ contextCollapse → restructures context for efficiency
(CONTEXT_COLLAPSE feature flag)
COMPACTION FLOW:
messages[] ──> getMessagesAfterCompactBoundary()
│
▼
older messages ──> Claude API (summarize) ──> compact summary
│
▼
[summary] + [compact_boundary] + [recent messages]
MCP (Model Context Protocol) Integration
┌─────────────────────────────────────────────────────────┐
│ MCP ARCHITECTURE │
│ │
│ MCPConnectionManager.tsx │
│ ├── Server Discovery (config from settings.json) │
│ │ ├── stdio → spawn child process │
│ │ ├── sse → HTTP EventSource │
│ │ ├── http → Streamable HTTP │
│ │ ├── ws → WebSocket │
│ │ └── sdk → in-process transport │
│ │ │
│ ├── Client Lifecycle │
│ │ ├── connect → initialize → list tools │
│ │ ├── tool calls via MCPTool wrapper │
│ │ └── disconnect / reconnect with backoff │
│ │ │
│ ├── Authentication │
│ │ ├── OAuth 2.0 flow (McpOAuthConfig) │
│ │ ├── Cross-App Access (XAA / SEP-990) │
│ │ └── API key via headers │
│ │ │
│ └── Tool Registration │
│ ├── mcp__<server>__<tool> naming convention │
│ ├── Dynamic schema from MCP server │
│ ├── Permission passthrough to Claude Code │
│ └── Resource listing (ListMcpResourcesTool) │
│ │
└─────────────────────────────────────────────────────────┘
Bridge Layer (Claude Desktop / Remote)
Claude Desktop / Web / Cowork Claude Code CLI
══════════════════════════ ═════════════════
┌───────────────────┐ ┌──────────────────┐
│ Bridge Client │ ←─ HTTP ──→ │ bridgeMain.ts │
│ (Desktop App) │ │ │
└───────────────────┘ │ Session Manager │
│ ├── spawn CLI │
PROTOCOL: │ ├── poll status │
├─ JWT authentication │ ├── relay msgs │
├─ Work secret exchange │ └── capacityWake │
├─ Session lifecycle │ │
│ ├── create │ Backoff: │
│ ├── run │ ├─ conn: 2s→2m │
│ └─ stop │ └─ gen: 500ms→30s│
└─ Token refresh scheduler └──────────────────┘
Session Persistence
SESSION STORAGE
══════════════
~/.claude/projects/<hash>/sessions/
└── <session-id>.jsonl ← append-only log
├── {"type":"user",...}
├── {"type":"assistant",...}
├── {"type":"progress",...}
└── {"type":"system","subtype":"compact_boundary",...}
RESUME FLOW:
getLastSessionLog() ──> parse JSONL ──> rebuild messages[]
│
├── --continue → last session in cwd
├── --resume <id> → specific session
└── --fork-session → new ID, copy history
PERSISTENCE STRATEGY:
├─ User messages → await write (blocking, for crash recovery)
├─ Assistant msgs → fire-and-forget (order-preserving queue)
├─ Progress → inline write (dedup on next query)
└─ Flush → on result yield / cowork eager flush
Feature Flag System
DEAD CODE ELIMINATION (Bun compile-time)
══════════════════════════════════════════
feature('FLAG_NAME') ──→ true → included in bundle
──→ false → stripped from bundle
FLAGS (observed in source):
├─ COORDINATOR_MODE → multi-agent coordinator
├─ HISTORY_SNIP → aggressive history trimming
├─ CONTEXT_COLLAPSE → context restructuring
├─ DAEMON → background daemon workers
├─ AGENT_TRIGGERS → cron/remote triggers
├─ AGENT_TRIGGERS_REMOTE → remote trigger support
├─ MONITOR_TOOL → MCP monitoring tool
├─ WEB_BROWSER_TOOL → browser automation
├─ VOICE_MODE → voice input/output
├─ TEMPLATES → job classifier
├─ EXPERIMENTAL_SKILL_SEARCH → skill discovery
├─ KAIROS → push notifications, file sends
├─ PROACTIVE → sleep tool, proactive behavior
├─ OVERFLOW_TEST_TOOL → testing tool
├─ TERMINAL_PANEL → terminal capture
├─ WORKFLOW_SCRIPTS → workflow tool
├─ CHICAGO_MCP → computer use MCP
├─ DUMP_SYSTEM_PROMPT → prompt extraction (ant-only)
├─ UDS_INBOX → peer discovery
├─ ABLATION_BASELINE → experiment ablation
└─ UPGRADE_NOTICE → upgrade notifications
RUNTIME GATES:
├─ process.env.USER_TYPE === 'ant' → internal features
└─ GrowthBook feature flags → A/B experiments at runtime
State Management
┌──────────────────────────────────────────────────────────┐
│ AppState Store │
│ │
│ AppState { │
│ toolPermissionContext: { │
│ mode: PermissionMode, ← default/plan/etc │
│ additionalWorkingDirectories, │
│ alwaysAllowRules, ← auto-approve │
│ alwaysDenyRules, ← auto-reject │
│ alwaysAskRules, ← always prompt │
│ isBypassPermissionsModeAvailable │
│ }, │
│ fileHistory: FileHistoryState, ← undo snapshots │
│ attribution: AttributionState, ← commit tracking │
│ verbose: boolean, │
│ mainLoopModel: string, ← active model │
│ fastMode: FastModeState, │
│ speculation: SpeculationState │
│ } │
│ │
│ React Integration: │
│ ├── AppStateProvider → creates store via createContext │
│ ├── useAppState(sel) → selector-based subscriptions │
│ └── useSetAppState() → immer-style updater function │
└──────────────────────────────────────────────────────────┘
The 12 Progressive Harness Mechanisms
This architecture demonstrates 12 layered mechanisms that a production AI agent harness needs beyond the basic loop. Each builds on the previous:
s01 THE LOOP "One loop & Bash is all you need"
query.ts: the while-true loop that calls Claude API,
checks stop_reason, executes tools, appends results.
s02 TOOL DISPATCH "Adding a tool = adding one handler"
Tool.ts + tools.ts: every tool registers into the dispatch
map. The loop stays identical. buildTool() factory provides
safe defaults.
s03 PLANNING "An agent without a plan drifts"
EnterPlanModeTool/ExitPlanModeTool + TodoWriteTool:
list steps first, then execute. Doubles completion rate.
s04 SUB-AGENTS "Break big tasks; clean context per subtask"
AgentTool + forkSubagent.ts: each child gets fresh messages[],
keeping the main conversation clean.
s05 KNOWLEDGE ON DEMAND "Load knowledge when you need it"
SkillTool + memdir/: inject via tool_result, not system prompt.
CLAUDE.md files loaded lazily per directory.
s06 CONTEXT COMPRESSION "Context fills up; make room"
services/compact/: three-layer strategy:
autoCompact (summarize) + snipCompact (trim) + contextCollapse
s07 PERSISTENT TASKS "Big goals → small tasks → disk"
TaskCreate/Update/Get/List: file-based task graph with
status tracking, dependencies, and persistence.
s08 BACKGROUND TASKS "Slow ops in background; agent keeps thinking"
DreamTask + LocalShellTask: daemon threads run commands,
inject notifications on completion.
s09 AGENT TEAMS "Too big for one → delegate to teammates"
TeamCreate/Delete + InProcessTeammateTask: persistent
teammates with async mailboxes.
s10 TEAM PROTOCOLS "Shared communication rules"
SendMessageTool: one request-response pattern drives
all negotiation between agents.
s11 AUTONOMOUS AGENTS "Teammates scan and claim tasks themselves"
coordinator/coordinatorMode.ts: idle cycle + auto-claim,
no need for lead to assign each task.
s12 WORKTREE ISOLATION "Each works in its own directory"
EnterWorktreeTool/ExitWorktreeTool: tasks manage goals,
worktrees manage directories, bound by ID.
Key Design Patterns
| Pattern | Where | Purpose |
|---|---|---|
| AsyncGenerator streaming | QueryEngine, query() |
Full-chain streaming from API to consumer |
| Builder + Factory | buildTool() |
Safe defaults for tool definitions |
| Branded Types | SystemPrompt, asSystemPrompt() |
Prevent string/array confusion |
| Feature Flags + DCE | feature() from bun:bundle |
Compile-time dead code elimination |
| Discriminated Unions | Message types |
Type-safe message handling |
| Observer + State Machine | StreamingToolExecutor |
Tool execution lifecycle tracking |
| Snapshot State | FileHistoryState |
Undo/redo for file operations |
| Ring Buffer | Error log | Bounded memory for long sessions |
| Fire-and-Forget Write | recordTranscript() |
Non-blocking persistence with ordering |
| Lazy Schema | lazySchema() |
Defer Zod schema evaluation for performance |
| Context Isolation | AsyncLocalStorage |
Per-agent context in shared process |
License
This repository content is for technical research and education only. All intellectual property rights belong to the original company. If there is any infringement, please contact us for removal.
