A 9-point upgrade audit for the skills you already saved

How this helps you

You find the skills most likely to fail or repeat an action, and get tighter instructions for them without anything changing until you approve.

✦ Try it

Run a read-only SKILL UPGRADE AUDIT across my Grok Bot setup. Start by listing every private skill you can access, which Bots have it enabled and any skills you cannot inspect. For each skill, check nine areas: 1. Purpose: is it clear when to use it, what problem it solves and when NOT to use it? 2. Inputs: does it list the information, connected apps and permissions it needs, and what to do when something is missing? 3. Workflow: is there a repeatable sequence with clear decision rules, not a copy of one successful run? 4. Verification: does it say how to check the result against the source? 5. Output: does it define exactly what to return, including format, evidence and next actions? 6. Approvals: does it separate what the Bot can do alone from what needs my permission, especially sending, buying, publishing or deleting? 7. Failure handling: what happens when a tool fails, access expires, information is outdated or the job only partly finishes? 8. Repeatability: can it run twice without duplicate emails, records or purchases? 9. Real-world test: design a safe test with a different, messier example. Give me a SKILL HEALTH REPORT: each skill and its missing instructions, risks ranked by importance, exact improvements, skills that need testing or access, and which Bots need the skill enabled. Draft the improved instructions for my approval. Do not modify, publish, share or run live actions through any skill without my explicit approval. After approval, test the updated skill safely and report what passed, failed or remains unverified.

Fill in: No placeholders. It's Michael's prompt lightly shortened, with the same nine checks.

What they did

Michael Fenech went through xAI's guidance on what makes a good skill and noticed a gap in how most people build them. Saving a task as a skill after it worked once is only the starting point. A skill that holds up also needs to say when it should run, what information it needs, how to check its own work and what to do when something goes wrong. So he wrote a read-only audit you paste into your chief of staff. It lists every private skill it can reach, which Bots have each one enabled, and any skill it can't inspect. Then it checks each skill against nine areas: purpose, inputs, workflow, verification, output, approvals, failure handling, repeatability (can it run twice without sending a second email or buying twice) and a real-world test on a messier example. The result is a skill health report with risks ranked by importance and improved instructions drafted for his approval. Nothing is changed or run live until he says yes, and after approval the Bot tests the updated skill safely and reports what passed, failed or is still unverified. His advice: build a skill for the second run, when things don't go to plan.

Who it's for: Anyone who has saved a handful of skills and wants them to behave the same way on the tenth run as on the first.

Before you start

A chief of staff or main Bot that can see your other Bots.
At least a few saved private skills.

How to try it

Paste the prompt below into your chief of staff Bot as one message.
Read the skill health report, starting with the risks it ranks highest. Approvals and repeatability gaps on skills that send, buy, publish or delete come first.
Check the list of skills it couldn't inspect and which Bots need a skill enabled; fix access before trusting the report.
Review the improved instructions it drafts for one skill at a time and approve the ones you agree with.
Let it run the safe test it designed on the updated skill and read what passed, failed or is still unverified.

What to expect

A report listing each skill with its gaps, such as an email skill with no rule for asking before sending, or a filing skill that would create a duplicate record if it ran twice, ranked by risk, plus draft instructions for each fix.

Common mistakes

Saving a skill after one good run and never revisiting it. That's the habit this audit is built to catch.
Letting it rewrite skills straight away. The approval line keeps a working skill from being changed without you seeing the new version.
Skipping the test. Michael asks for a messier example because the first run is usually the easy case.
Ignoring the 'couldn't inspect' list. A skill the audit can't see can still be running.

Related tips

Do it by hand, save a skill, then schedule: Most automations break because they're scheduled before they've worked once.
A read-only usage audit across your whole Bot team: One prompt has your chief of staff check every Bot's routines, schedules and triggers, then report what to pause, slow, narrow or merge, and change nothing until you approve.
Set up Auto-review rules: ask first for anything you can't undo: Turn on Auto-review, then add Ask first rules for money, publishing and sending. When rules clash, Ask first wins.