Managed Deep Agents project structure

​

Project layout

my-agent/
├── agent.py                        # Core agent definition

├── instructions.md                 # Managed context
├── skills/
│   └── <name>/
│       └── SKILL.md

├── tools/                          # Application code
├── middleware/

├── channels/                       # Managed configuration
│   └── <name>.py
├── schedules/
│   └── <name>.py
├── sandbox/
│   └── __init__.py
├── identity.py
├── memory.py

├── pyproject.toml                  # Dependencies and secrets
├── .env

└── evals/                          # Harbor workspace
    ├── tasks/                      # Canonical Harbor tasks
    │   └── <task>/
    └── scaffold/                   # Optional task scaffolds
        └── <task>/

​

How MDA treats project files

• Managed context: instructions.md defines the system prompt. Each directory under skills/ contains task-specific instructions. MDA syncs both to Context Hub.
• Application code: Files under tools/ and middleware/ are ordinary project modules. Import them from the agent entry. Other local modules work the same way.
• Managed configuration: Root identity.py and memory.py, direct children of channels/ and schedules/, and sandbox/__init__.py enable their corresponding capabilities.
• Dependencies and secrets: Declare dependencies in pyproject.toml. MDA loads .env locally and forwards eligible values as deployment secrets, but never includes .env files in the build archive.
• Evals: Managed Deep Agents evals are Harbor evals. evals/tasks/ is the canonical Harbor task dataset. Author tasks there directly, or run mda evals init <name> to create an optional starter under evals/scaffold/. mda evals compile copies scaffolds into evals/tasks/ and packages the agent for Harbor. The evals/ directory is not included in the deployed agent build.