Hey everyone,
Regarding the short course Spec-Driven Development with Coding Agents:
We’ve heard that the course can be confusing to follow along with, so I’ve put together a summary of the key concepts to help you stay oriented as you work through the material.
This isn’t a replacement for the lessons — it’s more like a map you can return to when things feel unclear. Use it to check where a topic fits in the bigger picture, or to review the core ideas after a session.
Hope it helps. Feel free to reach out if you have questions. 
# Spec-Driven Development with Coding Agents — Course Summary
## Core Idea
Instead of vibe coding (prompt → hope → fix → repeat), you write a **specification** first. The spec is the “brain”; the agent is the “muscle.”
-–
## Key Concepts by Lesson
### Why SDD (L1)
Vibe coding doesn’t scale. Specs create a persistent artifact that survives context window resets, reduces cognitive overhead, and improves how accurately the agent matches your intent.
### Workflow Overview (L2)
Three layers:
- **Constitution** — project-level
- **Feature specs** — feature-level
- **Replanning** — between features
Analogy: you’re the architect with blueprints; the agent is the builder.
### The Constitution (L4)
Three files written *with* the agent in conversation:
- `mission.md` — why + who
- `tech-stack.md` — how it’s built
- `roadmap.md` — what and in what order
### Feature Loop (L5–L7)
For each feature: **Plan → Implement → Validate**
- Each step happens in a fresh agent context
- Work in its own Git branch
- Spec files are committed alongside code
### Replanning (L8)
Between features, update the constitution and roadmap. Catch drift before it compounds.
- Small scope changes → do it now
- Large scope → schedule on roadmap
### MVP (L10)
Once your constitution and early feature specs are solid, you can tell the agent to implement the rest of the roadmap at once. Treat it as a stress test of your specs.
### Legacy / Brownfield Projects (L11)
Same process — the agent reverse-engineers a constitution from existing code, READMEs, and TODOs. Then continue with the normal feature loop.
### Custom Workflows / Skills (L12)
Automate repetitive SDD prompts using **Agent Skills** (a standardized format). Skills can be project-scoped or global. MCP servers are being partly replaced by Skills + CLI tools.
### Agent Replaceability (L13)
By following open standards (AGENTS.md, Agent Skills, MCP, ACP), your workflow isn’t locked to one agent or IDE.
-–
## SDD Workflow at a Glance
```
Constitution (once)
└── mission.md
└── tech-stack.md
└── roadmap.md
Feature loop (repeating)
└── Plan → feature spec (plan + requirements + validation criteria)
└── Implement → agent builds in a branch
└── Validate → human reviews, tests, commits, merges
Replanning (between features)
└── Update constitution if needed
└── Revise roadmap
└── Improve your workflow (skills, automation)
```