dotHuman is a method that gives your AI agents a workspace that’s durable, reliable, and transferable — so their work is spec-driven, reviewed by you, and remembered between sessions. This page follows real work up the ladder: unproven → audited → durable.
The protocol loop
Every goal runs the same loop — spec-driven, reviewed, banked. Put a real one through it below, and watch a project play out goal by goal — every step backed by a real .human/ file.
Unproven — every request becomes a spec-driven goal, not a one-off command.
This one’s a deliberately small “hello world” — picked so the whole trail fits on one screen. The same loop runs refactors, migrations, and net-new products the same way.
The machinery, not a testimonial.
This is founder-proof today. The receipts grow as more teams run the protocol on real work — and land here, cleared, as they do.
You want a small page that shows a different quote each day, starting from an empty folder. Here’s the whole run — what it decides, the bug it hits, and the notes it leaves so the next task starts ahead. Every line has the real file behind it.
Before
my-project/
empty folder
→the protocol
After
index.html
“The obstacle is the way.”
— Marcus Aurelius
A “quote of the day” page — no build, no dependencies, just a file you open. The box on the right is rendered live.
Prerequisite — once per project
First, it runs dotHumanize.
Before any goal, a one-time step turns the empty folder into a dotHumanized workspace — it sets up the .human/ notebook everything else writes to. Every project starts here, once.
The request — “show a quote of the day” — became a small written plan: what done means, the steps, a dated checklist. A target, not a vibe.
peek at the goal folder it created
Goals/2026-06-30_quote-of-the-day/
2026-06-30_quote-of-the-day/
├─spec.md# what "done" means├─plan.md# the approach├─tasks.md# the checklist, dated└─RESUME.md# where to pick up
Wrote down what it didn’t know.
Up front, it wrote down the tricky part: make today’s quote the same all day but different tomorrow — in the viewer’s timezone, not the server’s. Naming the unknown is what forces it to be handled.
peek at the spec it wrote
spec.md
# Goal — quote of the dayIntent: show one quote, chosen by the date, on load.
Unknown: keep the pick stable for the day — and "a day"
must mean the viewer's day, not the server's.
Done when: same date → same quote; no build, no deps.
Caught a real bug — and saved the fix for good.
The obvious approach picked the wrong day for anyone west of London. It caught that, fixed it, and wrote the rule into a reusable runbook — durable notes the next task inherits for free.
peek at the log entry & the runbook
captainslog.md · Lesson
Lesson — the simple "days since 1970" is a UTC day; out
west it flips at the wrong midnight. Verified: 6 vs 5
at 10pm. Fix: count from local midnight.
↓ saved into a runbook
# Evergreen/daily-pick.md
Index by the viewer's local day, from a fixed epoch.
Left itself a briefing for next time.
An AI’s memory resets between sessions — always. So it wrote a note for the next session: what this is, how it’s built, what to watch for. Next time, it’s up to speed in seconds instead of starting from scratch.
peek at the briefing it left
Comprehension/_contextreset.md
What it is: a "quote of the day" web page. No build, no deps.
Architecture: index.html — pick by the local day, then render.
Gotcha: "today" must be the local day (see the runbook).
Start here: read this, then the active Goal.
Goal 2extend it — the next goal, same project
Started the next goal already ahead.
A second job — add a daily color to match the quote — began by reading those notes. The timezone rule it learned last time? Reused as-is, no rediscovery. Each task begins further ahead than the last.
peek at the reused code
index.html — one rule, two values
// the same pick rule, reused for the color:
const pick = QUOTES[dayIndex(today, QUOTES.length)];
const accent = THEMES[dayIndex(today, THEMES.length)]; // reused
Turned a repeat into a reusable skill.
Because the same trick worked twice, it saved it as a small skill it can reach for again — “a value that changes once a day.” The project is now smarter than when it started.
Every step above is one turn of the loop you saw — plan a goal → review it → bank what you learned — with you signing off. After two goals the workspace carries more than after one, and the next goal starts from all of it.