feat: Add speed-prototyping skill

[thisyearnofear]

Summary

Adds speed-prototyping skill for rapid iterative development workflows.

Why this is specifically good for Cerebras

The skill operationalizes a key insight: when inference is fast, trying multiple approaches costs less than deliberating about which one to pick.

Traditional LLM workflow	Cerebras workflow
Plan carefully, execute once	Execute immediately, iterate fast
Debug failing approach	Pivot to new approach after 2 attempts
Batch changes to minimize latency cost	One change per iteration

This only works when each iteration is cheap. With Cerebras's millisecond inference, the "try both" strategy becomes faster than the "think harder" strategy.

Changes

.opencode/skill/speed-prototyping/
├── SKILL.md (68 lines)
└── examples/
    └── fetch-retry.md

Usage

Activates on natural language like:

• "prototype a rate limiter for our API"
• "spike out auth with JWT"
• "try a few approaches for caching"

Testing

bun dev
# Then invoke: /skill speed-prototyping