subagents-discipline_skill

Overview

This skill enforces a verification-first development discipline for multi-agent implementation tasks. It ensures developers inspect real data, run component and feature tests, and include a required DEMO block before claiming work is done. The goal is fewer integration surprises and reliable end-to-end features.

How this skill works

Invoke this skill at the start of any implementation task to apply three rules: look at actual data before coding, run both component and feature tests, and use available tools to verify behavior. It guides the developer through explicit commands to fetch real interfaces, run isolated checks, then exercise the integrated flow and capture evidence.

When to use it

• At the start of any task that touches external data, APIs, databases, files, or config
• Before writing migrations, API handlers, or UI integrations
• When working on epic sub-tasks that must match a design contract
• When you need to avoid integration regressions and ambiguous assumptions
• Whenever claiming a feature as complete or ready for review

Best practices

• Fetch and inspect the actual data/schema/output before coding (not docs or assumptions)
• Always produce both a component test and a feature (end-to-end) test
• Include a DEMO block with exact commands, steps, and observed results
• Check available MCP/tools and be resourceful (browser automation, curl, DB queries)
• If end-to-end testing is impossible, document what was verified, why, and what remains

Example use cases

• Implementing an API endpoint: inspect request/response shapes, run unit tests, then curl the endpoint and verify the UI path
• Adding a database migration: list table columns, write migration against observed schema, run app flows that read/write the table
• Building a frontend component: log actual props from the backend, render isolated tests, then run an integrated user flow
• Working on an epic child: pull and follow the design doc exactly, verify field names and types match the parent contract