Guided Skill Chaining Pattern
Each skill's output determines which skill runs next. Deterministic in nature, but conditional — the sequence is decided at runtime, not fixed in advance.
Why This Pattern
The core trade-off is cost and latency versus correctness. Chaining through the genie means every hop costs an LLM call — worth paying only when a step's output genuinely needs interpretation or user feedback before the next call happens. The cookbook is blunt about the anti-pattern: if it's just a sequence of deterministic API calls, don't make the LLM "reason" through them.
The reasoning lesson is the discipline, not the mechanism: chain skills through a genie because the LLM's judgment is actually needed between hops — not because chaining is available.
When to Use It
- A skill call needs a prerequisite skill's output as input (e.g., retrieving a template before creating a request).
- Complex, multi-step chains where each step's output shapes the next call — vendor matching, then department standardization, then GL code inference, then purchase order creation.
- The output of one skill needs to be passed to the LLM for processing or user feedback before the next call.
When NOT to Use It
- Do not use this pattern if the output of a skill doesn't need to reach the LLM for processing or user feedback. A sequence of pre-determined API calls belongs in a single recipe or skill instead — chaining it through the genie adds latency and cost for no benefit.
How It Works in Workato
- Provide instructions in the genie prompt to call one skill after another.
- Provide instructions in the skill description to call a prerequisite skill before the current skill can run.
- Add a field to the skill's input schema that explicitly expects the output of a prior skill call, structurally enforcing the correct sequence.
Source: Agentic Design Cookbook V2.0, pp. 12–13