cortex-qa

name	cortex-qa
description	Use when generating QA baselines, running scenario-driven gameplay QA, or starting an interactive QA session

Cortex QA

Intent Routing

• "generate baseline", "initialize QA", "set up QA profile" -> Baseline Mode
• "run this scenario", "execute gameplay test", "scenario file" -> Scenario Mode
• "interactive QA", "exploratory session", "step through PIE testing" -> Interactive Mode

Only ask a follow-up when the request is genuinely ambiguous between scenario execution and interactive exploration.

Baseline Mode

Prepare QA context for a project and generate an initial game profile.

1. Launch the cortex-toolkit:qa-engineer agent.
2. Ask it to:
   • verify MCP connectivity and PIE readiness
   • inspect key gameplay actors and systems
   • produce a first-pass QA profile using resources/game-profile-template.md
3. Save the generated profile in the project QA workspace requested by the user.

Keep this flow non-destructive. Prefer discovery and documentation over test mutation.

Scenario Mode

Execute a predefined gameplay QA scenario and return findings with report artifacts.

1. Use the Task tool with subagent_type: "cortex-toolkit:qa-engineer" and max_turns: 35.
2. Provide the scenario path or scenario content and request:
   • scenario execution via QA composite tools
   • structural issue detection after each step
   • screenshot capture on assertion failures
3. Require a final summary with:
   • pass/fail status
   • major and critical findings
   • paths to generated report files

If the agent returns completed, present the scenario results. If it returns blocked or partial, report what completed, what remains, and what blocked execution. If there is no status line, treat the run as partial and summarize what findings were collected.

Interactive Mode

Drive live exploratory testing in PIE with tight observe-act-assert loops.

1. Use the Task tool with subagent_type: "cortex-toolkit:qa-engineer" and max_turns: 75.
2. Instruct the agent to operate in guided mode:
   • execute one user-requested step at a time
   • report findings after every step
   • suggest the next highest-value probe
3. Continue until the user ends the session or a critical issue is found.

If the session reaches the turn limit, report what was covered and tell the user to continue by invoking cortex-qa again. Do not treat the turn limit as a failure.