// Enter and manage Deep Work sessions in Agent Hive. Use this skill when starting a focused work session on a project, generating session context, following the handoff protocol, or managing your responsibilities as an agent during a work session.
| name | deep-work-session |
| description | Enter and manage Deep Work sessions in Agent Hive. Use this skill when starting a focused work session on a project, generating session context, following the handoff protocol, or managing your responsibilities as an agent during a work session. |
A Deep Work session is a focused period where an AI agent works on a specific Agent Hive project. This skill guides you through the complete session lifecycle.
1. ENTER -> 2. CLAIM -> 3. WORK -> 4. UPDATE -> 5. HANDOFF
Run the dashboard:
make dashboard
# Or: uv run streamlit run src/dashboard.py
Select a project from the sidebar
Click "Generate Context"
Copy the context and paste to your AI agent
make session PROJECT=projects/your-project
# Or: ./scripts/start_session.sh projects/your-project
Read the AGENCY.md file directly and follow the protocol below.
When entering a Deep Work session, you MUST:
owner to your agent nameblocked: true if you need helpBefore starting work, verify and update the frontmatter:
# Check these are true:
status: active # Must be active
blocked: false # Must not be blocked
owner: null # Must be unclaimed
# Then set owner:
owner: "claude-sonnet-4"
last_updated: "2025-01-15T14:30:00Z"
Work on tasks in this order:
critical priorityhigh priorityUpdate the markdown content as you complete tasks:
## Tasks
- [x] Research existing solutions # Completed
- [x] Design architecture # Completed
- [ ] Implement core feature # In progress
- [ ] Write tests # Not started
Document your work with timestamped notes:
## Agent Notes
- **2025-01-15 15:45 - claude-sonnet-4**: Completed architecture design.
Chose event-driven pattern for scalability. Implementation ready to begin.
- **2025-01-15 14:30 - claude-sonnet-4**: Starting research phase.
Will evaluate 3 existing solutions.
If you encounter something you cannot resolve:
blocked: true
blocking_reason: "Need database credentials from DevOps team"
status: blocked # Optional: change status too
Add a note explaining:
- **2025-01-15 16:00 - claude-sonnet-4**: BLOCKED - Cannot proceed with
database integration. Need credentials from DevOps. Created ticket #123.
Before ending your session, you MUST complete these steps:
Mark everything you finished with [x]
last_updated: "2025-01-15T16:30:00Z"
Document what was accomplished and any context for the next agent:
- **2025-01-15 16:30 - claude-sonnet-4**: Session complete. Finished
research and design phases. Implementation is ready to begin - start
with `src/feature.py`. Note: the authentication module has a known
issue (see issue #45).
If completely done with the project:
owner: null
status: completed # If all tasks done
If handing off to another agent:
owner: null
status: active # Keep active for next agent
If you'll continue later:
owner: "claude-sonnet-4" # Keep ownership
# Add note about when you'll return
blocked: true # Only if external help needed
blocking_reason: "Waiting for API access approval"
The Dashboard generates this context package:
# DEEP WORK SESSION CONTEXT
# Project: project-id
# Generated: 2025-01-15T14:30:00
---
## YOUR ROLE
[Responsibilities and instructions]
---
## AGENCY.MD CONTENT
[Full AGENCY.md file with frontmatter and content]
---
## PROJECT FILE STRUCTURE
[Directory tree of project files]
---
## AVAILABLE COMMANDS
[What actions you can take]
---
## HANDOFF PROTOCOL
[Required steps before ending]
---
## BOOTSTRAP COMPLETE
[Signal to begin work]
owner to your agent namelast_updatedblocked if you hit a walllast_updated periodicallylast_updated timestampowner released or retained intentionallystatus updated if appropriateblocked/blocking_reason set if needed