gru-ai: Skill for Claude Code

Walkthrough — Cognitive Walkthrough

Role Resolution

Read .claude/agent-registry.json to map roles to agent names. Use each agent's id as the subagent_type when spawning. The CPO designs the ideal experience; the CTO traces the actual implementation.

---

Simulate user scenarios against the current system. Find what's broken, missing, or surprising.

The pattern: For each scenario, design what SHOULD happen (ideal), trace what DOES happen (actual), report the gaps.

Arguments: $ARGUMENTS

• A specific scenario name (e.g., ceo-runs-directive) → run just that one
• all → run all standing scenarios
• A free-text scenario description (e.g., "seller wants to see competitor prices") → ad-hoc walkthrough
• Empty → list available scenarios and ask which to run

Step 1: Load Scenarios

If $ARGUMENTS is a scenario name or "all":

Read standing scenarios from .context/lessons/scenarios.md.

Each scenario has:

• Name: slug identifier
• Actor: who is performing the action (CEO, seller, shopper, developer)
• Trigger: what starts the flow ("CEO types /directive improve-security")
• Goal: what the actor wants to achieve
• Critical path: the steps that MUST work for the scenario to succeed

If all, load all scenarios. If a specific name, load just that one.

If $ARGUMENTS is free text:

Treat it as an ad-hoc scenario. Spawn the CPO to formalize it:

You are the CPO. The CEO described a user scenario informally:

"{$ARGUMENTS}"

Formalize it into this structure:
{
  "name": "slug-name",
  "actor": "who is doing this",
  "trigger": "what starts the flow",
  "goal": "what the actor wants to achieve",
  "critical_path": [
    "Step 1: what should happen first",
    "Step 2: what should happen next",
    ...
  ],
  "success_criteria": "how do you know the scenario succeeded"
}

Think from the ACTOR's perspective, not the system's. What does the actor expect at each step? What would surprise or frustrate them?

CRITICAL OUTPUT FORMAT: First character must be `{`, last must be `}`. JSON only.