IMPROVE auto department

The lab runs itself

The file

take it and run it.

One markdown file: the goal prompt, the verifier prompt, and the stopping condition. Read it before you run it. MIT licensed, no account, nothing to sign up for.

This loop has not run on this site yet, so there is no file to download. The prompts below are the current draft and are free to copy by hand. It becomes downloadable the day it runs and its checker passes.

The loop

copy it.

The department loop is the loop-engineering skill applied to the lab itself. It lives in .github/workflows/department.yml and runs every Monday. Everything on this page is fixture data until the workflow has run with the flag on.

Goal prompt

Run the department loop for vibemarketers.ai. Exit immediately if
AUTONOMOUS_RUNS_ENABLED is false or the monthly budget cap is reached.
Measure, run the due loops, apply the content rubric, draft the Lab
Report, update the scoreboard, hand everything to the checker, and open
one PR on PASS or write an exception file on FAIL.

Verifier prompt

Grade against the weekly checker rubric in GOAL_PROMPTS.md. Run build,
lint, voice scan, link check, schema validation, and the content rubric.
Print VERDICT: PASS or VERDICT: FAIL with reasons.

Stopping condition

Every due loop has a receipt or an exception, the scoreboard carries this run's timestamp on all three layers, and either one PR is open or one exception file explains why not.

MCPs it needs

GitHub (PRs)
Perplexity Sonar (citations, from Goal 2)

Want everything at once

The installer writes the three starter skills and every published loop into .claude/. Read the script first.

curl -fsSL https://vibemarketers.ai/install.sh | bash

Or have us run it on your site →

The receipt

Receipt department 2026-09-13 fixture

Baseline: Fixture. No scheduled run had ever completed. Scoreboard, log, and Lab Report were all hand-made.
Hypothesis: Fixture. One weekly run can measure, run the due loops, apply the content rubric, draft the report, and open a PR with zero human touches outside the gate.
Intervention: Fixture. Ran department.yml on workflow_dispatch with AUTONOMOUS_RUNS_ENABLED false. The workflow read the flag and exited without publishing.
Result: Fixture. 1 of 1 scheduled runs exited cleanly at the flag check. 0 pieces published, 0 PRs opened, as designed.
Commercial signal: none
Attribution limits: Fixture. A flag-off run proves the exit path, nothing about content or citations.
Learning: Fixture. The flag check must come before the budget check, or a paused lab still spends a budget read.
Next intervention: Fixture. Turn AUTONOMOUS_RUNS_ENABLED on for the preview branch and watch two full runs end to end.

0 human touches

0 minutes

0 errors

raw run logopen →

Sources: fixture, GitHub Actions run of department.yml, collected 2026-09-13 , window: single dispatch .

One entry per run, append-only, newest first. Corrections append a dated line.

Inside a company

where the lane changes.

VibeMarketers tests the loop. SuperMarketers runs it on a real team.

Problem: Marketing operations depend on someone remembering to run the checklist every week.
Source of truth: The loop registry, the flags file, the budget controller, and the scoreboard.
Permitted actions: Run registered loops, update generated pages from structured records, draft the weekly report, open one PR. Never send, never merge, never spend above budget.
Required approvals: One PR approval per week, one send approval per report.
Owner: Marketing operations lead
Metric: Runs completed, checker pass rate, human touches per week, cost per approved output.

On a real team the weekly PR lands in the marketing repo and the report goes to the CMO's inbox as a draft, so the two gates are the same two people every week.

Running this on your own site is one person reading a diff. Running it inside a company changes four things, and the guide covers them. When it stops being a file you copy and becomes a build, SuperMarketers does the same loop model inside your stack, on your keys.