Smart-Kickoff is a launcher for AI coding agent sessions created by 📝One Inc that treats startup as a load contract, deciding which 📝context an agent must read now, which loads only when a 📝deterministic trigger fires, and which is never loaded by default.
Most agent setups treat context as a pile — project instructions, operating manuals, safety rules, incident history, and task notes all pushed into the launch prompt because each one once mattered. Smart-Kickoff separates those sources by timing rather than by topic. A resolver reads the work memo that initiated the session, writes a manifest of the sources it selected and the load class assigned to each, and refuses oversized undeclared context that cannot produce a compact launch slice. The manifest doubles as the audit trail, recording what loaded and why without pasting private source bodies into the prompt.
Launch runs two independent checkpoints. The first asks whether the briefing is readable: it resolves the Context memo the roadmap points at and verifies that the memo is small enough and structured with a Load Contract the launcher can parse. A Context memo written as one oversized block with no Load Contract fails here, and the launcher refuses to start rather than loading it anyway. The second asks which repository the session targets, inferring it from the memo's 📝prose and halting to ask a human when that inference is not confident. Both checkpoints fail closed — an ambiguous answer stops the launch instead of guessing.
The same timing model governs state-changing work after launch. Operations touching code, data, credentials, deployments, or shared memory trigger mutation gates that require their exact safety sources before proceeding, rather than trusting that semantic retrieval surfaced the right document earlier. The distinction the design rests on is that discovery is not authorization: a fuzzy match is not a safety gate.
We built Smart-Kickoff at One Inc because "read everything first" felt safe and behaved like a tax. Every session paid for stale manuals it never opened, while the one rule that mattered sat buried beside old incident history. Our fix was not more context — it was making each source declare when it earns its place, then failing closed the moment that claim is malformed.
Build It
Hand this to your coding agent. It retrieves the paired 📝Smart-Kickoff — Agent Buildbook, then runs that buildbook's discovery against your own stack before building anything:
Read the Smart-Kickoff Agent Buildbook — with MythOS MCP, memo brianswichkow-ed4e43; without it, fetch https://mythos.one/me/brianswichkow/ed4e43. Then: ask me how technical I want this walkthrough, inventory my setup, interview me where it's unclear, and propose a phased plan. Don't build a phase until I approve it.