Problem
Current Coze Agent capabilities are scattered across multiple execution entry points:
- Skill: invoked via
skill_load
- Adapter: standalone scripts
- Schedule (Calendar): time-triggered tasks
- Heartbeat: periodic state checks
This fragmentation causes:
- No unified management interface for execution entry points
- Duplicated capabilities across mechanisms, hard to reuse
- No unified scheduling strategy (priority, timeout, retry)
- Users must learn each mechanism independently
Proposed Solution
1. Executor Hub - Unified Execution Framework
Design a single dispatch layer that all execution entry points register with:
ExecutorHub
├── SkillExecutor (skill_load → registry → execute)
├── AdapterExecutor (bash script → registry → execute)
├── ScheduleExecutor (calendar trigger → registry → execute)
└── HeartbeatExecutor (timer → registry → check)
2. Capability Registry
- Each capability registers its type, cost, timeout, and priority
- Auto-discovery of available capabilities per session
- Conflict detection when multiple executors can handle the same task
3. Smart Dispatch
- Cost-aware routing: choose lowest-cost executor first
- Priority queue: critical tasks get dedicated resources
- Timeout & retry: unified timeout handling with configurable retry
- De-duplication: prevent the same task from running on multiple executors
4. Unified Monitoring
- Single dashboard for all execution history
- Cost tracking per executor type
- Performance metrics (latency, success rate)
Impact
- Reduces user learning curve (one mental model instead of four)
- Improves execution reliability (unified retry/fallback)
- Enables cost optimization at the dispatch layer
- Makes the platform more maintainable as new executor types are added
Priority
P2 (Medium) - Improves developer experience and platform maintainability
Problem
Current Coze Agent capabilities are scattered across multiple execution entry points:
skill_loadThis fragmentation causes:
Proposed Solution
1. Executor Hub - Unified Execution Framework
Design a single dispatch layer that all execution entry points register with:
2. Capability Registry
3. Smart Dispatch
4. Unified Monitoring
Impact
Priority
P2 (Medium) - Improves developer experience and platform maintainability