Build Notes

SmartWriter Build Notes

Porting a coding agent's capabilities into a writing product — everything that broke, and every tradeoff along the way

00 "Just a Wrapper" Is Harder Than It Sounds The opening post of a vibe-coding series: porting a coding agent's capabilities into a writing product, and everything that broke along the way

Engine & Rhythm

01 The Agent's Heartbeat: What Really Happens Inside "One Conversation" Pulling apart the Agent Loop to understand what's actually happening inside "one conversation" 02 Context Amnesia: Compaction Is a Budget War The context window auto-compacts when it fills up — how a writing agent keeps its core thesis intact after that happens

Memory & Session

03 Close the Window, Pick Up Where You Left Off: One Writing Task = One Session How a writing task survives a shutdown and resumes — and how multiple tasks stay cleanly separated 04 Getting It to Actually Know You Better Over Time: Memory Isn't One Big Pot, It's Three Drawers Taking apart 'gets better the more you write' — how long-term, cross-article memory is actually built

Steering the Model

05 The System Prompt Isn't a Persona — It's the Foundation of Behavior Control The system prompt is the first place agent behavior gets shaped — and it's not just about writing a persona 06 Skills Are the Manual: Where Concrete Best Practices Live Where does genre-specific writing methodology actually live? The answer is Skills, loaded on demand 07 What's Behind a Slash: Commands, Output Contracts, and Flow Design The trigger trap hiding in slash commands, and how to design the output contract behind each one

World Interaction

08 Don't Let the Model Talk Its Way Through It: Custom Tools, MCP, and the Boundary of Determinism Equipping a writing agent with tools: build a custom tool, or plug into external MCP 09 An Agent Can't Touch the World Directly: The Six-Step Permission Chain and Deleting Bash Tools are a double-edged sword — a six-step evaluation chain, and deleting Bash outright, frame every call 10 Hooks Are the Real Hooks: Injection, Backstops, and Audit Trails Beyond permissions, hooks are the all-purpose hook for injecting your own logic into how an agent runs

Orchestration & Delegation

11 Plan Before You Act: Writing's "Option C" Why structurally complex writing tasks need a custom plan-before-acting solution 12 Send Out a Clone: Context Isolation, Final Review, and Why Understanding Must Converge Farming out heavy, easy-to-derail work to clones running in isolated context

Vertical Business

13 Not "Just Another Generic Agent": Product Flow Design Built for Writing Product design that bridges standard agent capabilities, one at a time, to real writing tasks 14 Turning the Agent's Inner Monologue into an Interface Users Actually Understand: Streaming, Diffs, and Cards Translating precise, opaque backend agent state into an interface non-technical users can actually understand

15 An Agent Grown Into the Business: What Machine Do 12 Parts Add Up To? The finale: putting 12 parts back together to finally answer "isn't this just a wrapper?"