Vessel + Talent System¶

Deep dive into the modular agent architecture.

Employee Directory Structure¶

employees/00010/
├── profile.yaml          # Employee profile
├── vessel/               # Vessel DNA
│   ├── vessel.yaml       # Config: runner / hooks / limits / capabilities
│   └── prompt_sections/  # Prompt fragments
├── skills/               # Talent — skills
└── progress.log          # Working memory

vessel.yaml — The DNA¶

Field	Purpose
`runner`	Neural system — custom runner module and class
`hooks`	Lifecycle hooks — pre_task / post_task callbacks
`context`	Context injection — prompt sections, progress log, task history
`limits`	Execution limits — retry count, timeout, subtask depth
`capabilities`	Capability declarations — sandbox, file upload, WebSocket, image gen

Vessel Harness — 6 Connection Protocols¶

Harness	Responsibility
`ExecutionHarness`	Executor protocol (execute / is_ready)
`TaskHarness`	Task queue management (push / get_next / cancel)
`EventHarness`	Logging and event publishing
`StorageHarness`	Progress log and history persistence
`ContextHarness`	Prompt / context assembly
`LifecycleHarness`	Pre/post task hook invocation

Talent → Employee Flow¶

flowchart LR
    T["Talent Package<br>(pilot)"] -->|"onboarding"| V["Vessel<br>(mech)"]
    D["default_vessel.yaml"] -.->|"fallback"| V
    V --> E["Employee<br>(active unit)"]
    E --> EM["EmployeeManager<br>register + dispatch"]