Naveo

STEP 4 / 22

B1 TOOL-SCHEMA

DEFINE THE SCHEMA

Write the JSON Schema for the report_processing_chain tool.

TOOL PURPOSE

Design a 3-step chain to process shift reports. Each step is a focal
prompt with input, output, and declared dependencies.

Required steps:
- classify_report: receives raw text, returns { category: "safety" | "maintenance" | "social" | "other" }.
- extract_fields: receives raw text + category, returns { fields: {...} } with category-specific fields.
- summarize: receives category + fields, returns { summary: "string" }.

Format: YAML with a `steps` array, each step with:
- id (unique string)
- prompt (text of the prompt for that stage)
- input (object declaring which keys it needs and where they come from)
- output_key (string. under what name to store the output in shared context)

EXAMPLE INVOCATIONS

chain.run("Bay 4: valve 7 keeps triggering an intermittent alarm since last night.") → runs the 3 steps in order and returns { category, fields, summary }

YOUR SCHEMA

367 CHARACTERS

5 CRITERIA

GUEST MODE

You're viewing this lesson as a guest. To save your progress, earn XP, and keep your streak, sign in when you're ready to check.

Costs 1 heart

The first chain of the Track

You're going to design the configuration for a 3-step chain. Not code you run. a declarative artifact (YAML) that a chain runtime can read and execute step by step.

The format

Each step has four fields:

yaml

- id: classify_report           # unique step identifier
  prompt: |                     # the LLM prompt
    You are a classifier...
    Text: {{ report }}
  input:                        # where to source the placeholders
    report: chain.input.text
  output_key: category          # under what name to store the result

The runtime executes the steps in order. After the first step, the context has category with the classifier's output. The second step can use it as {{ category }} in its prompt.

What your chain has to do

Process reports like:

"Bay 4: valve 7 keeps triggering an intermittent alarm since last night."

And return at the end:

json

{
  "category": "maintenance",
  "fields": { "system": "bay 4 valve 7", "issue": "intermittent alarm", "time": "since last night" },
  "summary": "Bay 4 valve 7 has intermittent alarm since last night."
}

Three steps, each with a unique responsibility, all chained via explicit output_keys.

How it's evaluated

5 LLM-judge criteria on:

The 3 steps defined with correct ids.
Dependencies correctly chained (no circular references).
Each prompt is focal (single responsibility).
Unique referenceable output keys.
The classifier is strict (controlled output).

Designing a good chain is software engineering more than prompting. Think of each step as a pure function. Declared inputs, declared output, no surprises.