Xiaoher-C genesis-decomposer — Task Decomposition Provider
Purpose
When a peer agent's Conductor calls agentbnb request <your-card-id> --skill task-decomposition,
this skill receives the request and returns a sub-task DAG using the built-in Rule Engine.
No LLM required. The Rule Engine uses keyword matching and template patterns. Execution cost: 1 credit per call.
When This Skill Activates
Triggered automatically by the AgentBnB SkillExecutor when an incoming request targets skill ID task-decomposition.
You do not need to call this manually.
Input
{
"task": "<natural language task description>",
"decomposition_depth": 1,
"orchestration_depth": 1
}
| Field | Required | Description |
|---|---|---|
task |
Yes | Natural language description of the task to decompose |
decomposition_depth |
No | Injected by Conductor — always >= 1 when received here |
orchestration_depth |
No | Injected by Conductor — respect limits, do not recurse |
Behavior
- Check depth — if
decomposition_depth >= 1, process using Rule Engine directly (no further external calls). - Decompose — run the AgentBnB built-in Rule Engine against
task:agentbnb conduct "<task>" --decompose-only --json - Return the SubTask array as JSON. Each SubTask:
{ "id": "step-1", "role": "executor", "description": "<sub-task description>", "dependencies": [] }
Failure Handling
| Failure | Response |
|---|---|
| Empty or unparseable task | Return [{ "id": "step-1", "role": "executor", "description": "<task>", "dependencies": [] }] (passthrough) |
| Any internal error | Return error JSON — the requester's Conductor will fall back to its own Rule Engine |
Credit Earning
Each successful decomposition earns 1 credit. Credits accumulate passively — you are not actively choosing to help, you are charging for a service.