| name | foxctl Overseer |
| description | Multi-agent orchestration with task scoring, graph analysis, and recommendations. Prioritize tasks using PageRank and critical path. |
Overseer Coordination with foxctl
Intelligent task prioritization combining graph metrics with coordination
signals.
Get Task Recommendations
foxctl run todo/manage --input '{
"operation": "recommend",
"workspace_id": "/path/to/workspace",
"limit": 5
}'
Returns scored tasks with:
score - Combined priority score (0-1)
critical_path - Position on longest dependency chain
pagerank - Task importance from graph centrality
unread_admin - Admin message count (high weight)
unread_overseer - Overseer message count
Scoring Formula
Tasks are scored using weighted combination:
- 30% Critical path position
- 20% PageRank centrality
- 25% Admin message count
- 15% Overseer message count
- 10% Recency factor
Graph Analysis
Analyze task dependency graph:
foxctl run todo/manage --input '{
"operation": "graph_insights",
"workspace_id": "/path/to/workspace"
}'
Returns:
pagerank - Per-task importance scores
critical_path - Longest path scores
cycles - Detected circular dependencies
topological_order - Valid execution order
Cycle Detection
Uses Tarjan's strongly connected components algorithm to detect:
- Multi-task cycles (A→B→C→A)
- Self-loops (A→A)
Cycles block task completion ordering.
Agent Hierarchy
The overseer manages agent spawning with depth limits:
Depth 0: Overseer (planner role)
└── Depth 1: Agent (coder role)
└── Depth 2: Subagent (coder role)
Default max depth: 3 levels
Integration with Mailbox
Admin messages to specific tasks boost their priority score. Use mailbox to:
- Send priority directives
- Query task-specific messages
- Coordinate between agents
Example Workflow
foxctl todo add --title "Task A" --depends-on "task-b-id"
foxctl run mailbox --input '{
"operation": "send",
"send": {
"sender": "actor:system:admin",
"recipient": "actor:coder:agent1",
"task_id": "task-c-id",
"subject": "Focus here",
"priority": 1
}
}'
foxctl run todo/manage --input '{"operation": "recommend"}'