| name | design-doc |
| description | Write a short design doc (RFC) before building — problem, goals & non-goals, proposed approach, alternatives considered, risks, and a validation plan — and get sign-off before the agent writes code. Part of the "keep the agent on track" suite. Use to stop an agent from charging ahead on the wrong solution. |
Design doc (decide the approach first)
Why
The most expensive agent mistakes are approach mistakes — built quickly, wrong direction. A one-page design doc forces the choice into the open and gets agreement before any code.
Structure (DESIGN.md)
# Problem / context
What's the situation, why act now.
# Goals & non-goals
Goals: what success includes.
Non-goals: what we are explicitly NOT doing (this is the valuable part).
# Proposed approach
The plan, in enough detail to critique.
# Alternatives considered
Option B, Option C — and why not.
# Risks & mitigations
What could go wrong; how we'd catch or contain it.
# Validation plan
How we'll know it worked.
Protocol
- The agent drafts, a human approves, then implementation starts.
- If the approach changes mid-build, reopen the doc — don't silently pivot.
Anti-patterns
- Writing the design doc after the code (theater).
- No non-goals — scope creeps without them.
- No alternatives — a decision with one option isn't a decision.