QA Team AI Workbench

A command-driven workflow for QA teams

QA work is often fragmented across PRD review, test design, agent testing, bug analysis, reporting, and team management. Instead of treating this as one generic generate-test-cases prompt, this skill models the workflow as explicit commands: /qa is the unified entry point for intent parsing, routing, memory management, and planning; subcommands cover /qa-prd, /qa-case, /qa-agent, /qa-bug, /qa-report, /qa-team, and /qa-explore.

Progressive loading with validation

The skill avoids injecting every prompt into context at once. It first reads the entry point and intent rules, then loads the relevant references/ prompt file. For example: review this PRD → /qa routes to PRD → load prd/prompt.md → run validation-rules.md before output. Important constraints include verb-first, independently verifiable test steps; missing data marked as {{待确认}} or -; report statistics must come from user-provided data; high-severity PRD issues, low/medium-confidence root-cause findings, memory writes, and version cleanup all require confirmation.

Boundaries

It is best for structured QA tasks with clear inputs: PRD review, test case design, AI/Agent testing, bug root-cause analysis, report drafting, and team dashboards. It does not invent compliance standards, accounts, amounts, paths, or test data, and it will not continue generation when required inputs are missing. If intent is ambiguous, it should present candidate commands for confirmation.