CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

build-the-job

Andrew Rendon's rules for building a job the business picked, so it runs as drafts from a scheduled task. Use when Andrew or the owner builds, sets up, tests or proves a job, alone or alongside the factory skills (amp, doll, pathfinder, triangle, verify, blind-review, wrap).

Quality

86%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Low

Low-risk findings worth noting

SKILL.md
Quality
Evals
Security

What you cannot know without being told:

  • The business's own tools first. For each tool in the brain's tools-and-accounts, ask whether that tool can do the job itself, with its own feature or through its Claude connector. Say which tools you checked and what each can and cannot do. Only then a Claude scheduled task. Custom code last.
  • How a Cowork scheduled task works (support.claude.com/en/articles/13854387, checked 2026-11-14). A task has a name, a prompt, an approval mode, a frequency (hourly, daily, weekly, on weekdays, or manually), and optionally a model and a folder. It runs remotely, even when the computer is asleep or the app is closed, unless it needs local files or apps; then it runs only locally. It uses the connectors and plugins set up in Cowork. So the task reaches the brain through the Google Drive connector, not a folder on the computer.
  • The owner's judgment is asked, not inferred. Where the job decides something the owner decides today (which day to offer, how far back to look, what to leave out, what to do when something is missing), ask the owner how they decide. Past examples show what they did, not their rule.
  • One folder per job. Everything for the job goes in the brain under jobs/<job name>/: its steps, its examples, the task prompt and your working papers. Working papers name a customer by first name and street only, never a surname, house number, email or phone number.
  • Held-back answers stay out of the brain. Keep what the owner actually sent for the held-back examples in a folder beside the brain, never inside it, and never point the task at it.
  • Proved before it is done. The job is done only when its task prompt has run in a fresh sub-agent on the held-back examples and every draft has been compared with what the owner actually sent: send as written, one small edit, or rewrite. Give the sub-agent the task prompt, the brain, and for each held-back customer only what the task would see before the owner answered (their email, the visit note); never the answers or the owner's replies to them. Report the count.
Repository

4W-dev/andrewrendon-plugin

Path

plugins/andrewrendon/skills/build-the-job/SKILL.md

Last updated
First committed

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.