/archive/library
The source documents behind the archive: working lenses, operating guides, and the engineering decisions that made the projects necessary.
10 on the shelf · 5 field notes · 22 decision records
35 more in the deep stacks — ⌘K finds them
The operating guides behind the work, mirrored from the live harness
N°000 · guided overview
The day-to-day operating system around the projects: skills, parallel lanes, verification, overnight gates, and the publishing loop.
start here · guide
How I build real products with agents. Me first, shareable second. Every claim in here either carries a receipt from my own systems or is marked > Backlog: with the trigger that will make it real. …
start here · guide
Rewritten 2026-07-13. The v3 manual grew into a 1,100-line platform spec that taxed every read and got skipped under pressure — which showed up as burnout from juggling sessions, manual UI testing …
guide · How I work
The timeless agent-design material extracted from the v3 operating manual (2026-07-13) so the daily runbook could get short. This is product-design knowledge, not daily process: read it when design…
guide · How I work
This is written specifically for your situation: a 16-agent portfolio built on one architectural theorem (a deterministic, unit-testable gate owns every irreversible action), a shared agent-core ha…
Read in the project that made them necessary
13 decision records
latest 3 of 13
9 decision records
latest 3 of 9
First-party notes from the workbench, then the idea lenses — mirrored with source and revision history
note · from the workbench
Before an agent ships: twelve design questions, a deterministic owner for every irreversible action, risk tiers set by blast radius, and a precise understanding of what a green check does — and does not — prove.
note · from the workbench
Tests catch broken code; evals catch a system that confidently does the wrong thing. For agent products, the second category is where life and death happen — so the suite grows from reality, and the ruler itself gets audited.
note · from the workbench
The scaffolding around AI agents has one job: the model proposes, deterministic code disposes. Hooks, gates, and anti-cheat lines make a green check mean something — and the harness itself is a product, with its own gate.
note · from the workbench
Coding agents work best inside a loop they can verify themselves. Designing that loop — what the agent starts with, what it can check, when it stops — beats prompting harder, every time.
note · from the workbench
The task loop that ships real features with two human touchpoints: correct a three-line contract at the start, judge an evidence packet at the end. Everything between runs without you.
lens · Hackathon lenses
2026-07-08. Third lens. v1 asked "can Claude + MCPs do this today?" v2 asked "when the model can do everything, what's left that's ours?" This doc asks: if we committed to building it, what does th…
HISTORY: Hackathon Idea Backlog · IDEAS 2.0 — Build the Stakes, Not the Smarts
lens · Hackathon lenses
Event prompt: pick one clinical or operational workflow; build an agentic system that makes it faster, smarter, or safer; ship something a clinician or patient-facing team could use Monday. Abridge…
lens · Hackathon lenses
Copy, don't import: take the file, strip what's unused, own it in the new repo. Declare reused code to organizers when rules require. /hack reads this file during scaffolding.
lens · Hackathon lenses
Researched 2026-07-05. Goals: connect with builders, learn how others build, meet hiring managers/recruiters, and ship demos that become talking points. Winning is a side effect.