- Updated code-standards.md with v4.3.2 subsystem organization, delegation patterns, and symlinking standards - Enhanced codebase-summary.md to document delegation system, selective .claude/ symlinking, and shell completion - Revised project-overview-pdr.md with v4.3.2 capabilities, feature requirements, and architectural evolution - Added comprehensive project-roadmap.md documenting version history and future plans - Updated system-architecture.md with delegation architecture, symlinking strategy, and diagnostics infrastructure - Added repomix-output.xml to .gitignore
20 KiB
CCS Codebase Summary (v4.3.2)
Overview
CCS (Claude Code Switch) v4.3.2 is a lightweight CLI wrapper enabling instant profile switching between Claude Sonnet 4.5, GLM 4.6, GLMT (GLM with Thinking), and Kimi for Coding models. Version 4.x introduces AI-powered delegation, selective .claude/ directory symlinking, stream-JSON output, and enhanced shell completion.
Version Evolution
v4.3.2 Architecture (Current)
- Total LOC: ~8,477 lines (JavaScript only)
- Key Features: AI delegation, stream-JSON output, shell completion, doctor diagnostics, sync command
- New Components: delegation/, utils/claude-symlink-manager.js, utils/delegation-validator.js, utils/update-checker.js
- Architecture: Modular design with clear separation: auth/, delegation/, glmt/, management/, utils/
Evolution Summary
- v2.x: Vault-based credential encryption (~1,700 LOC)
- v3.0: Vault removal, login-per-profile (~1,100 LOC, 40% reduction)
- v4.0-4.3.2: Delegation system, .claude/ sharing, stream-JSON (~8,477 LOC including tests/utils)
Core Components (v4.3.2)
1. Main Entry Point (bin/ccs.js - ~800 lines)
Role: Central orchestrator with delegation routing
Key Functions:
execClaude(claudeCli, args, envVars): Unified spawn logic (Windows shell detection)handleVersionCommand(): Version display with delegation statushandleHelpCommand(): Comprehensive help with delegation examplesexecClaudeWithProxy(claudeCli, profile, args): GLMT proxy lifecyclemain(): Profile routing + delegation detection (-p flag)
v4.x Enhancements:
- Delegation detection:
-pflag routes to DelegationHandler - Stream-JSON output support for real-time tool tracking
- Shell completion installation (
--shell-completion) - Update checking and CCS sync commands
- Enhanced version display with API key validation
Architecture Flow:
// Delegation path (v4.0+)
if (args.includes('-p') || args.includes('--prompt')) {
const { DelegationHandler } = require('./delegation/delegation-handler');
const handler = new DelegationHandler();
await handler.route(args);
}
// Settings profile (glm, kimi, glmt)
const expandedSettingsPath = getSettingsPath(profileInfo.name);
if (profileInfo.name === 'glmt') {
await execClaudeWithProxy(claudeCli, 'glmt', remainingArgs);
} else {
execClaude(claudeCli, ['--settings', expandedSettingsPath, ...remainingArgs]);
}
// Account profile (work, personal)
const instancePath = instanceMgr.ensureInstance(profileInfo.name);
registry.touchProfile(profileInfo.name);
const envVars = { CLAUDE_CONFIG_DIR: instancePath };
execClaude(claudeCli, remainingArgs, envVars);
2. Delegation System (bin/delegation/ - ~1,200 lines)
New in v4.0: Complete delegation subsystem
Components:
- delegation-handler.js (~300 lines): Routes
-pcommands, validates profiles - headless-executor.js (~400 lines): Executes Claude CLI in headless mode with stream-JSON
- session-manager.js (~200 lines): Manages delegation session persistence (continue support)
- result-formatter.js (~150 lines): Formats execution results with cost/duration
- settings-parser.js (~150 lines): Parses profile settings for validation
Key Features:
- Stream-JSON output: Real-time tool visibility (
--output-format stream-json --verbose) - Session continuation:
ccs glm:continue -p "follow-up"resumes last session - Tool tracking: Shows file paths, commands, patterns as they execute
- Signal handling: Ctrl+C kills child processes properly
- Cost tracking: USD cost display per delegation
- 13 Claude Code tools supported: Bash, Read, Write, Edit, Glob, Grep, NotebookEdit, SlashCommand, TodoWrite, etc.
Delegation Flow:
User: ccs glm -p "add tests"
↓
DelegationHandler: Parse args, validate profile
↓
HeadlessExecutor: Spawn Claude CLI with --output-format stream-json --verbose
↓
Stream parser: Extract [Tool] lines, format in real-time
↓
SessionManager: Save session ID for :continue
↓
ResultFormatter: Display cost, duration, exit code
3. Auth System (bin/auth/ - ~800 lines)
Role: Multi-account management (unchanged from v3.0 core)
Components:
- auth-commands.js (~400 lines): CLI handlers for auth subcommands
- profile-detector.js (~150 lines): Profile type routing (settings vs account)
- profile-registry.js (~250 lines): Metadata management (profiles.json)
v4.x Status: Core logic stable, focus shifted to delegation
Profile Creation Flow (v3.0+):
# Create profile (prompts login)
ccs auth create work # Opens Claude, auto-prompts OAuth
# Use directly (credentials in instance)
ccs work "task"
4. GLMT System (bin/glmt/ - ~900 lines)
Role: GLM with thinking mode via embedded proxy
Components (unchanged from v3.x):
- glmt-proxy.js (~400 lines): HTTP proxy on localhost:random
- glmt-transformer.js (~300 lines): Anthropic ↔ OpenAI format conversion
- reasoning-enforcer.js (~100 lines): Inject reasoning prompts
- locale-enforcer.js (~50 lines): Force English output
- delta-accumulator.js (~200 lines): Streaming state tracking
- sse-parser.js (~50 lines): SSE stream parser
Status: Stable experimental feature, not actively developed in v4.x
5. Management System (bin/management/ - ~600 lines)
Role: Diagnostics, recovery, instance management
Components:
- doctor.js (~250 lines): Health check diagnostics
- instance-manager.js (~220 lines): Instance lifecycle (v3.0 simplified)
- recovery-manager.js (~80 lines): Auto-recovery for missing configs
- shared-manager.js (~50 lines): Shared data symlinking (v3.1+)
v4.x Enhancements:
- doctor.js: Now checks delegation commands in
~/.ccs/.claude/commands/ccs/ - Validates .claude/ symlinks from v4.1
6. Utilities (bin/utils/ - ~1,500 lines)
New in v4.x: Expanded utility modules
Components:
- claude-detector.js (~70 lines): Claude CLI detection
- claude-dir-installer.js (~150 lines): Copy .claude/ from package (v4.1.1)
- claude-symlink-manager.js (~200 lines): Selective .claude/ symlinking (v4.1)
- config-manager.js (~80 lines): Settings config management
- delegation-validator.js (~100 lines): Validate delegation eligibility (v4.0)
- error-codes.js (~50 lines): Standard error codes
- error-manager.js (~200 lines): Error handling utilities
- helpers.js (~100 lines): TTY colors, path expansion
- progress-indicator.js (~150 lines): Spinner/progress display
- prompt.js (~100 lines): User input prompting
- shell-completion.js (~250 lines): Shell auto-completion installation (v4.1.4)
- update-checker.js (~100 lines): Version update notifications (v4.1)
Key Utilities:
- ClaudeDirInstaller: Copies
.claude/from npm package to~/.ccs/.claude/ - ClaudeSymlinkManager: Creates selective symlinks to
~/.claude/(Windows fallback to copy) - DelegationValidator: Validates profile readiness (API keys, settings)
7. CCS .claude/ Directory (/.claude/ - packaged with npm)
New in v4.1: Selective symlink approach
Structure:
.claude/
├── commands/ccs/ # Delegation slash commands
│ ├── glm.md # /ccs:glm "task"
│ ├── kimi.md # /ccs:kimi "task"
│ ├── glm/continue.md # /ccs:glm:continue "task"
│ └── kimi/continue.md# /ccs:kimi:continue "task"
├── skills/ccs-delegation/ # Auto-delegation skill
│ ├── SKILL.md # Skill definition
│ ├── CLAUDE.md.template # User CLAUDE.md snippet
│ └── references/troubleshooting.md
└── settings.local.json # Repomix permissions
Symlink Strategy (v4.1):
- Source:
~/.ccs/.claude/(copied from npm package) - Target:
~/.claude/commands/ccs@,~/.claude/skills/ccs-delegation@ - Selective: Only CCS items symlinked, doesn't overwrite user's other commands/skills
- Windows: Falls back to copying if Developer Mode not enabled
Installation Flow (v4.1.1):
npm install -g @kaitranntt/ccs- Postinstall: ClaudeDirInstaller copies
.claude/→~/.ccs/.claude/ - Postinstall: ClaudeSymlinkManager creates selective symlinks →
~/.claude/ - User can now use
/ccs:glmand/ccs:kimicommands
File Structure (v4.3.2)
bin/
├── auth/ # Multi-account management (v3.0 core)
│ ├── auth-commands.js # CLI handlers (~400 lines)
│ ├── profile-detector.js # Profile routing (~150 lines)
│ └── profile-registry.js # Metadata management (~250 lines)
├── delegation/ # AI delegation system (v4.0+)
│ ├── delegation-handler.js # Route -p commands (~300 lines)
│ ├── headless-executor.js # Execute with stream-JSON (~400 lines)
│ ├── session-manager.js # Session persistence (~200 lines)
│ ├── result-formatter.js # Format results (~150 lines)
│ ├── settings-parser.js # Parse settings (~150 lines)
│ └── README.md # Delegation documentation
├── glmt/ # GLM thinking mode (v3.x)
│ ├── glmt-proxy.js # Embedded HTTP proxy (~400 lines)
│ ├── glmt-transformer.js # Format conversion (~300 lines)
│ ├── reasoning-enforcer.js # Reasoning prompts (~100 lines)
│ ├── locale-enforcer.js # English enforcement (~50 lines)
│ ├── delta-accumulator.js # Stream state (~200 lines)
│ └── sse-parser.js # SSE parser (~50 lines)
├── management/ # System management (v3.x+)
│ ├── doctor.js # Health diagnostics (~250 lines)
│ ├── instance-manager.js # Instance lifecycle (~220 lines)
│ ├── recovery-manager.js # Auto-recovery (~80 lines)
│ └── shared-manager.js # Shared symlinking (~50 lines)
├── utils/ # Utilities (expanded in v4.x)
│ ├── claude-detector.js # CLI detection (~70 lines)
│ ├── claude-dir-installer.js # .claude/ installer (v4.1.1, ~150 lines)
│ ├── claude-symlink-manager.js # Selective symlinks (v4.1, ~200 lines)
│ ├── config-manager.js # Config management (~80 lines)
│ ├── delegation-validator.js # Delegation validation (v4.0, ~100 lines)
│ ├── error-codes.js # Error codes (~50 lines)
│ ├── error-manager.js # Error handling (~200 lines)
│ ├── helpers.js # Utilities (~100 lines)
│ ├── progress-indicator.js # Progress display (~150 lines)
│ ├── prompt.js # User prompting (~100 lines)
│ ├── shell-completion.js # Shell completion (v4.1.4, ~250 lines)
│ └── update-checker.js # Update checker (v4.1, ~100 lines)
└── ccs.js # Main entry (~800 lines)
.claude/ # CCS-provided items (v4.1+)
├── commands/ccs/ # Delegation commands
│ ├── glm.md
│ ├── kimi.md
│ ├── glm/continue.md
│ └── kimi/continue.md
├── skills/ccs-delegation/ # Auto-delegation skill
│ ├── SKILL.md
│ ├── CLAUDE.md.template
│ └── references/troubleshooting.md
└── settings.local.json
scripts/
├── postinstall.js # Auto-config + migration
├── sync-version.js # Version management
├── check-executables.js # Validation
├── completion/ # Shell completions (v4.1.4)
│ ├── ccs.bash
│ ├── ccs.zsh
│ ├── ccs.fish
│ ├── ccs.ps1
│ └── README.md
└── worker.js # Cloudflare Worker (ccs.kaitran.ca)
tests/
├── unit/ # Unit tests
│ ├── delegation/ # Delegation tests (v4.0+)
│ └── glmt/ # GLMT tests (v3.x)
├── npm/ # npm package tests
├── integration/ # Integration tests
└── shared/ # Shared test utilities
~/.ccs/ # User installation
├── .claude/ # CCS items (copied from package)
│ ├── commands/ccs/
│ └── skills/ccs-delegation/
├── shared/ # Shared across profiles (v3.1+)
│ ├── commands@ → ~/.claude/commands/
│ ├── skills@ → ~/.claude/skills/
│ └── agents@ → ~/.claude/agents/
├── instances/ # Isolated Claude instances
│ └── work/
│ ├── commands@ → shared/commands/
│ ├── skills@ → shared/skills/
│ ├── agents@ → shared/agents/
│ ├── settings.json (if any)
│ ├── sessions/
│ └── ...
├── config.json # Settings-based profiles
├── profiles.json # Account-based profiles
├── delegation-sessions.json # Delegation session history (v4.0)
├── glm.settings.json
├── glmt.settings.json
├── kimi.settings.json
└── logs/ # Debug logs
~/.claude/ # User's Claude directory
├── commands/ccs@ → ~/.ccs/.claude/commands/ccs/ # Selective symlink (v4.1)
├── skills/ccs-delegation@ → ~/.ccs/.claude/skills/ccs-delegation/ # Selective symlink
└── (user's other commands/skills remain untouched)
Data Flow (v4.3.2)
Delegation Execution (v4.0+)
User: ccs glm -p "add tests to UserService"
↓
ccs.js: Detect -p flag → route to DelegationHandler
↓
DelegationHandler: Parse { profile: 'glm', prompt: 'add tests', options: {} }
↓
DelegationValidator: Check API key, settings validity
↓
HeadlessExecutor: Spawn Claude CLI with:
- --settings ~/.ccs/glm.settings.json
- --output-format stream-json
- --verbose
- --prompt "add tests to UserService"
↓
Stream parser: Extract [Tool] lines in real-time
- [Tool] Grep: searching for test patterns
- [Tool] Read: reading UserService.js
- [Tool] Write: creating UserService.test.js
↓
SessionManager: Save { sessionId, profile: 'glm', timestamp }
↓
ResultFormatter: Display summary
- Working Directory: /home/user/project
- Model: GLM-4.6
- Duration: 12.3s
- Cost: $0.0023
- Session ID: abc123 (use :continue to resume)
↓
Exit with Claude CLI exit code
Settings Profile Execution (glm, kimi, glmt)
User: ccs glm "command"
↓
ccs.js: Parse arguments, detect profile "glm"
↓
ProfileDetector: detectProfileType("glm") → {type: 'settings'}
↓
ConfigManager: getSettingsPath("glm") → "~/.ccs/glm.settings.json"
↓
ccs.js: execClaude(["--settings", path, "command"])
↓
Claude CLI: Execute with GLM API
Account Profile Execution (work, personal)
User: ccs work "command"
↓
ccs.js: Parse arguments, detect profile "work"
↓
ProfileDetector: detectProfileType("work") → {type: 'account'}
↓
InstanceManager: ensureInstance("work") → "~/.ccs/instances/work/"
↓
ProfileRegistry: touchProfile("work") → Update last_used
↓
ccs.js: execClaude(["command"], {CLAUDE_CONFIG_DIR: instancePath})
↓
Claude CLI: Read credentials from instance, execute
Key Features (v4.3.2)
1. AI-Powered Delegation (v4.0)
- Headless execution:
ccs glm -p "task"runs without interactive UI - Stream-JSON output: Real-time tool visibility (
[Tool] Write: file.js) - Session continuation:
ccs glm:continue -p "follow-up"resumes last session - Cost tracking: USD cost display per delegation
- Signal handling: Proper Ctrl+C cleanup
- 13 tools supported: Comprehensive Claude Code tool coverage
2. Selective .claude/ Symlinking (v4.1)
- Package-provided:
.claude/ships with npm, copied to~/.ccs/.claude/ - Selective symlinks: Only CCS items linked to
~/.claude/ - Non-invasive: Doesn't overwrite user's commands/skills
- Windows support: Falls back to copying if symlinks unavailable
- Auto-sync:
ccs syncre-creates symlinks
3. Enhanced Shell Completion (v4.1.4)
- 4 shells: bash, zsh, fish, PowerShell
- Color-coded: Commands vs descriptions
- Categorized: Model profiles, account profiles, flags
- Auto-install:
ccs --shell-completionorccs -sc
4. Comprehensive Diagnostics (v4.1+)
- ccs doctor: Health check for installation, configs, symlinks, delegation
- ccs sync: Re-sync delegation commands and skills
- ccs update: Check for updates (v4.1+)
5. GLMT Thinking Mode (v3.x, stable experimental)
- Embedded proxy: HTTP proxy on localhost:random
- Format conversion: Anthropic ↔ OpenAI
- Reasoning injection: Force English, thinking prompts
- Debug logging:
CCS_DEBUG_LOG=1writes to~/.ccs/logs/
Breaking Changes
v3.0 → v4.0
- No breaking changes: v4.0 purely additive (delegation features)
- New dependencies: None (all Node.js built-ins)
- Migration: Automatic via postinstall
v2.x → v3.0 (Historical)
- Command renamed:
ccs auth save→ccs auth create - Schema changed: Removed
vault,subscription,emailfields - Migration required: Users must recreate profiles
Performance Characteristics (v4.3.2)
Delegation Performance
- Headless spawn: ~20-30ms overhead
- Stream parsing: <5ms per tool call
- Session save: ~10ms (JSON write)
- Total overhead: ~35-45ms vs direct Claude CLI
Profile Activation (unchanged from v3.0)
- Instance validation: ~5ms
CLAUDE_CONFIG_DIRenv var: ~1ms- Claude CLI spawn: ~20-30ms
- Total overhead: ~26-36ms
.claude/ Symlinking (v4.1)
- Symlink creation: ~5-10ms per link
- Copy fallback (Windows): ~50-100ms total
- Sync command: ~100-200ms (full re-sync)
Security Model (v4.3.2)
Unchanged from v3.0
- No custom encryption (credentials managed by Claude CLI)
- Spawn with array arguments (no shell injection)
- Atomic file writes (temp + rename)
- Instance directory permissions (0700)
New in v4.x
- API key validation (checks for placeholder keys)
- Delegation eligibility checks (validates settings before execution)
- Signal handling (proper cleanup of child processes)
Testing Coverage (v4.3.2)
Unit Tests
tests/unit/delegation/: Delegation system tests (v4.0+)tests/unit/glmt/: GLMT transformer tests (v3.x)tests/shared/unit/: Utility function tests
Integration Tests
tests/npm/: End-to-end CLI functionalitytests/integration/: Cross-platform behavior, edge cases
Test Count
- Total: ~50 test files
- Coverage: >90% for critical paths
Dependencies (v4.3.2)
Production
cli-table3@^0.6.5: Table formatting (doctor command)ora@^5.4.1: Spinner display (progress indicators)
Development
mocha@^11.7.5: Test runner
Note: Minimal dependencies, all critical functionality uses Node.js built-ins
Future Extensibility
Extension Points (v4.x)
- New delegation profiles: Easy addition via DelegationValidator
- Custom result formatters: Pluggable ResultFormatter
- Session management: SQLite for better session queries
- MCP integration: Delegation via MCP tools
- Cost optimization: Model selection based on task complexity
Summary
CCS v4.3.2 Achievements:
- Delegation system: Complete AI-powered task routing with stream-JSON
- Selective symlinking: Non-invasive .claude/ directory sharing
- Shell completion: Enhanced UX with color-coded completions
- Diagnostics: Comprehensive health checking and auto-recovery
- Modular architecture: Clear separation of concerns (auth/, delegation/, glmt/, management/, utils/)
Design Principles Maintained:
- YAGNI: Only essential features implemented
- KISS: Simple, readable code without over-engineering
- DRY: Single source of truth for each concern
Code Quality:
- Total LOC: ~8,477 lines (bin/ JavaScript only)
- Test Coverage: >90% for critical paths
- Modularity: 7 subsystems (main, auth, delegation, glmt, management, utils, .claude/)
- Documentation: Comprehensive inline comments and external docs
v4.3.2 demonstrates successful feature expansion (delegation, symlinking, diagnostics) while maintaining core simplicity and zero breaking changes from v3.0. The modular architecture provides a sustainable foundation for future AI-powered development workflow enhancements.