-
Notifications
You must be signed in to change notification settings - Fork 0
Open
Description
Problem
Currently there is a single shared set of specialist agents (review-agent, test-agent, merge-agent) that handle work across all projects. This causes issues:
- Queueing conflicts - When multiple projects have work ready for review, they compete for the same review-agent
- Context pollution - Specialists accumulate context from multiple unrelated projects
- Scaling bottleneck - Can only process one review/test/merge at a time across all projects
- Project-specific configuration - Different projects may need different review criteria, test commands, or merge strategies
Proposed Solution
Each project should have its own set of specialists:
~/.panopticon/specialists/
├── myn/
│ ├── review-agent/
│ ├── test-agent/
│ └── merge-agent/
├── panopticon/
│ ├── review-agent/
│ ├── test-agent/
│ └── merge-agent/
└── househunt/
├── review-agent/
├── test-agent/
└── merge-agent/
Benefits
- Parallel processing - Multiple projects can have reviews/tests running simultaneously
- Clean context - Each specialist only knows about its project
- Custom prompts - Projects can override default specialist prompts
- Independent queues - Project A's backlog doesn't block Project B
Implementation Notes
- Specialist tmux sessions should be named
specialist-<project>-<type>(e.g.,specialist-myn-review-agent) - Session IDs and history should be stored per-project
- Fallback to global specialists if project-specific ones don't exist
- Consider lazy initialization (only create specialists when first needed)
Acceptance Criteria
- Specialists are created per-project
- Specialist state (session ID, history) is stored per-project
- Multiple projects can run reviews in parallel
- Project-specific prompt overrides are supported
- Backward compatible with existing global specialists during migration
Metadata
Metadata
Assignees
Labels
No labels