contract-validation

Contract Validation Skill

Validates agent input and output contracts using the Contract-First design pattern. Every OMT agent has a contract in contracts/<agent-name>.json that defines required inputs, outputs, and validation rules.

Core Concepts

Agent Contract Structure

Every agent must define:

1. Input Contract: What the agent needs to start (sources, required fields, validation rules)
2. Output Contract: What the agent must produce (required fields, destinations)
3. Validation Rules: How to verify correctness (see references/validation-rules.md)

Contract Validation Flow

1. Read agent contract: contracts/<agent-name>.json
2. Gather input data from contract's source locations
3. Validate input contract
4. Execute agent logic
5. Validate output contract
6. Update workflow-state.json with results

How to Validate

Step 1: Read the Contract

Use the Read tool to load the agent's contract definition:

Read contracts/dev.json

The contract specifies input_contract.source — a list of file paths to gather input from.

Step 2: Validate Input Before Execution

Before starting agent work:

1. Gather input data — For each source in input_contract.source, use the Read tool to check the file exists and has content:

   • Read outputs/pm.md (requirements from @pm)
   • Read outputs/arch.md (architecture from @arch)
   • Use Glob to find existing files matching patterns (e.g., tests/**/*.test.ts)

2. Check required fields — For each field in input_contract.required:

   • Verify the data exists and is not empty
   • Apply validation rules (e.g., fileExists, minLength:N)
   • If any required field fails: stop and report errors

3. Record validation — Read lib/contract-validator.ts for validation logic, then perform equivalent checks using Read/Glob tools. Log the result.

Step 3: Validate Output After Execution

After completing agent work:

1. Collect output data — Gather all outputs specified in output_contract.required:
   • Use Glob to list created files (e.g., tests/**/*.test.ts, src/**/*.ts)
   • Capture execution results (e.g., test status: "15/15 passed")