How shslab-org/shs-code works
SHS-Code is a persistent autonomous coding agent: users submit a task through an interactive or service interface; the agent plans, acts through tools, and verifies results, retaining session and memory state. Its documented product branches include Team103 multi-agent execution, project intelligence, skills, MCP, integrations, and multiple model providers. The tree and README support these subsystem boundaries; sampled source confirms server-to-agent execution, engineer delegation, skill-tool management, and intelligence indexing, but not every runtime connection shown here.
Main entry points: shscode.py Coding agent · main.py HTTP server · selector.py Agent tools · cli.py CLI
Components 25
User interfaces
- CLI
app/cli.py - HTTP server
app/server/main.py - Messaging channels
app/messaging/gateway.py - Cron scheduler
app/cron.py - Live canvas
app/canvas/server.py
Agent runtime
- Coding agent
app/agent/shscode.py - Agent orchestrator
app/agent/orchestrator.py - Team103 roles
app/team103/scheduler.py - Model interface
app/llm/llm.py - Model providers
app/providers.py
Execution capabilities
- Agent tools
app/tool/selector.py - Verification
app/verification.py - Sandbox execution
app/sandbox/factory.py - Security controls
app/security/policy_rails.py - MCP client/server
app/mcp/client.py
Context and knowledge
- Context management
app/context/view.py - Persistent memory
app/memory/long_term.py - Skills engine
app/skills/skill_engine.py - Project intelligence
app/intelligence/manager.py
Persistence and integrations
- Conversation state
app/conversation/factory.py - Session database
app/db/session.py - Git providers
app/git_providers/provider.py - External integrations
app/integrations/resolver.py
Other components
- User
- Model APIs
Connections 27
- User → CLI: submits task
- User → HTTP server: submits task
- User → Messaging channels: sends message
- CLI → Coding agent: starts task
- HTTP server → Coding agent: runs task
- HTTP server → Agent orchestrator: runs multi-agent
- Cron scheduler → Coding agent: schedules task
- Messaging channels → Coding agent: routes messages
- Agent orchestrator → Team103 roles: coordinates roles
- Team103 roles → Coding agent: delegates implementation
- Coding agent → Agent tools: invokes tools
- Coding agent → Model interface: requests completion
- Model interface → Model providers: selects provider
- Model providers → Model APIs: calls model API
- Coding agent → Context management: builds context
- Context management → Persistent memory: uses memory
- Coding agent → Skills engine: loads skills
- Coding agent → Project intelligence: queries project
- Coding agent → Verification: verifies work
- Agent tools → Sandbox execution: runs code
- Coding agent → Security controls: checks risky actions
- Coding agent → MCP client/server: uses MCP tools
- Coding agent → Conversation state: uses conversation
- HTTP server → Session database: reads session data
- Agent tools → Git providers: manages repositories
- Coding agent → External integrations: uses integrations
- HTTP server → Live canvas: mounts canvas