Add a spec-first / scoping workflow guide

## Summary

Add a **spec-first / scoping workflow** guide (and optionally a `spec-writer` skill). We use AI for well-defined scope; the missing piece is documented discipline for *getting to* that scope. Spec-first work means fewer correction loops, which means fewer tokens per task — the metric that actually drives cost.

## The guide should cover

- Why scoping/spec-first reduces cost (tokens-per-task, not per-token price).
- A practical loop: spec → plan → execute → verify.
- How to write a good spec / acceptance criteria for an agent.
- When a task is too vague to hand to an agent, and how to tighten it.
- Optionally: a `spec-writer` skill that helps draft a spec from a rough idea.

## Notes

- Back the findings with web research across multiple independent sources.
- Keep it tool-general (GitHub Copilot CLI, OpenCode, Pi).
- Tie in with existing material on agent mode and token budget.