| name | prep-owner-steps |
| description | Hand the owner finished steps, not directions — deep links, paste-ready blobs, his path walked once, one batched sitting, payoff + verification stated (Q-0273 seed skill). |
prep-owner-steps
Hand venture-lab's owner finished steps, not directions.
Owner-directed (Q-0273, 2026-07-12). His words: agents give him
"directions clearly mapped out — go to github, actions, settings, rule, etc,
and paste this file from this github page in this repo — while they could
easily lead with the link and the copy/paste ready file in chat as a
separate block." The method: before handing the owner any step, ask "which
steps will he face, and are there any text blobs he might need to enter that
I can send?" — then send them. superbot carries the founding copy; this is
the kit's generalized template, inherited by every adopter.
When this runs
Every time work surfaces an owner-only step (repo settings, secrets,
portal clicks, external publish, account actions) or any instruction the
owner must execute by hand — in chat replies, owner-queue items, ⚑ blocks,
and run reports. It is the drafting half of the OWNER-ACTION contract
(control/README.md): those fields prove the ask is necessary and verified;
this method makes it executable in one sitting.
The method — prepare, don't describe
- Lead with the direct link. Deep-link the exact page (the ruleset
editor, the variables tab, the release form) — never a navigation trail
when a URL exists. A trail ("Settings → Rules → main") is the fallback
annotation under the link, for when the deep link 404s.
- Every blob he must enter ships as its own fenced block. File contents,
variable names, one-line replies, commit messages, form fields — each in a
separate copy-ready block, labeled with exactly where it goes. Never "paste
the file from this page" — fetch it and paste it INTO the chat/queue item
yourself.
- Walk his path once yourself, in your head or via probe. Enumerate
every click/field he will actually face (the screen he lands on, the
button names, the confirm dialogs) and pre-answer each. If a step's outcome
feeds a later step (a minted token → a variable), say what to carry
forward and where it lands.
- Batch to one sitting. If the task needs owner steps at multiple
points, restructure so they cluster into ONE block at the start or end —
never interleave "now go click X" through a work narrative he has to
babysit.
- State the payoff + verification. One line each: what completing the
steps unblocks, and how he (or the next agent) verifies it worked (the
exact URL/command that should now succeed).
The shape (use it in queue items and chat alike)
⚑ OWNER — <what this unblocks, one line>
1. <deep link> (fallback: Settings → … → …)
paste ↓ into <exact field>:
```<the blob>```
2. …
verify: <command/URL that should now succeed>
The bar
The owner should be able to complete every step with clicks and pastes
only — zero composing, zero fetching, zero deciding-what-you-meant. If he
has to open another page to copy something you could have copied for him,
the preparation failed.
Declared capabilities: read-only (a drafting method — the invoking session's
own capabilities do any writing).
