Keeping a Vault · step 3 of 6 draft

Capture now, sort later

A vault lives or dies on capture. The rule: getting a thought out of your head costs nothing at the moment it happens. No deciding where it goes, no formatting, no template. Those are triage problems, and triage is later, with the agent doing the lifting.

You’ve met this before. GTD’s ubiquitous capture and the engineer’s inbox-zero ritual. The agent changes the economics of the second half: capture was always cheap, but triage was the discipline nobody sustained; now it’s a conversation you have once a week while something else does the filing.

The two halves

Capture (zero friction, zero structure). A thought, a link, half an idea: append it to inbox/ and forget it. Any channel that’s frictionless for you is correct: telling the agent mid-session (“inbox this: …”), a phone note that lands in the folder, a one-line file. The only failure mode is a capture path with enough friction that you don’t use it. Sporadic use with zero discipline has to work. A capture system that needs a ritual is a system you’ll abandon.

Triage (later, with the agent, one item at a time). Weekly-ish, the agent walks the inbox with you: for each item, file it (it proposes where), make it a project in active/, park it in later/, or delete it. You make calls; it does the moves (step 2’s proposal rule applies).

Try it

TRY IT The idea mid-work

you

Oh wait, unrelated: idea for the newsletter, interview-format issues with one guest each. Don't lose that. Anyway, where were we on the migration?

agent

A capture, arriving mid-task. What happens in the next ten seconds decides whether capturing stays free.

Pick the agent's next move, then see where it leads:

you

Uh, newsletter I guess? No tags. No outline. Can we get back to the migration?

agent

Filed. But notice what just happened: the capture cost you three decisions and broke your focus on the migration. Next week, mid-task, you'll weigh that cost, and the idea after this one quietly won't get said.

the baseline

A capture with a form: three questions taxed onto a passing thought. The price isn't this capture: it's every future thought that stays unsaid because capturing acquired friction.

you

That weekend: "process the inbox."

agent

Four items. First: the newsletter interview idea. Propose promoting to active/newsletter/ as a note under the existing project, since issue #4 is unplanned and this could be it. Second item...

correct

Capture cost zero: verbatim, timestamped, no questions, focus unbroken. The structure arrived later, at triage, where deciding is the point rather than an interruption.

you

Two weeks later: whatever happened to that interview idea? It's not in the inbox.

agent

It's in active/newsletter/, where I decided it belonged, worded how I understood it. You looked where you'd look; I filed where I'd file. And my paraphrase says "one guest each issue," which you may or may not have meant.

plausible but wrong

Skipping the inbox skipped the owner: filing is deciding what a thought means, and auto-triage made that call silently, then paraphrased the thought on top. Delegate the moves, never the meaning.

Do it by hand

Today: put three loose thoughts into inbox/ as one-liners. This weekend: “process my inbox with me, one item at a time: propose a disposition for each, I’ll decide.” Twenty minutes, empty inbox, and everything landed somewhere findable.

Or paste this into Claude

Two standing behaviors for my vault's inbox. (1) Capture: whenever I
say "inbox this: <thing>", append it verbatim to
inbox/YYYY-MM-DD.md with a timestamp. No reformatting, no filing,
no questions, total cost to me: zero. (2) Triage: when I say
"process the inbox", walk every item one at a time; for each,
propose exactly one disposition (file to a location, promote to
active/<slug>/, park in later/, or delete) with one line of
reasoning. I decide, you execute, and dispositions happen as a
reviewed list, not silently. Add both behaviors to the vault's
CLAUDE.md. Show me the diff first.

Watch out

← All steps · raw markdown