Before you save any prompt as a skill — 7 checks.

Not for every decision. For the prompt → skill moment specifically. The gate that catches bad skills before they live in your stack forever. Run this once per skill before you click save in Skill-Creator.

A bad prompt fired once costs you nothing. A bad SKILL lives in your stack forever — until you delete it.

The 7 Checks

Check 01 · Bias

The Bias Check

"Did I ask the question that wants confirmation, or the one that wants truth?"

The Question Flip from Section 4. "Should I launch this?" gets 5 reasons to launch. "Is this a bad idea?" gets 5 reasons it might fail. Both prompts work — but the second one tells you something useful.

If you fail: Flip your core ask. Make Claude argue the side you're avoiding.

Check 02 · Inversion

The Inversion Check

"Have I considered what the failure mode of this prompt would look like?"

If this skill fires on a bad day, with stale context, what's the worst output it could produce? If the worst case is "embarrassing client deliverable" — your prompt isn't tight enough.

If you fail: Add an anti-goal section. Name the 3 worst outputs you can imagine and tell Claude not to produce them.

Check 03 · Specificity

The Specificity Check

"Are Role, Context, Task, and Output ALL clearly named?"

Open your prompt. Find each one. If any are missing or implied, the skill will produce inconsistent output. Especially Output — what format, length, voice exactly?

If you fail: Re-run Prompt-Maker. Add the missing piece. Don't try to patch it in by hand.

Check 04 · Anti-Goal

The Anti-Goal Check

"Did I name what I DON'T want, not just what I do?"

80% of taste lives in what you'd never accept. "Write a sales page" gets generic AI output. "Write a sales page that doesn't sound like every other coach's sales page — no urgency tactics, no 'transform your life,' no fake scarcity" gets something usable.

If you fail: Add 3-5 specific things you'd reject. Look at the LAST thing Claude wrote that you rewrote — that's the anti-goal.

Check 05 · Stack Test

The Stack Test

"Could I add ONE cheat code to make this 30% sharper?"

Strategy skills get sharper with Red Team or Pre-Mortem baked in. Decision skills get sharper with Inversion. Research skills get sharper with First Principles. One drop-in code in the prompt body changes the output ceiling.

If you fail: Pick the most relevant code from the toolkit and add it as a step in the instructions.

Check 06 · Voice Lock

The Voice Lock Check

"Does it properly inherit my Instructions for Claude?"

Your Instructions for Claude has your banned words, tone rules, defaults. A good skill doesn't restate them — it inherits them. If your skill says "no preamble" again, you're duplicating. Trust the global layer.

If you fail: Delete every line that's already in your Instructions for Claude. If it's in the global layer, it doesn't need to be in the skill.

Check 07 · Trigger

The Trigger Test

"Will the description actually fire when I need it to?"

Open your description. Are there 3+ trigger phrases in YOUR actual language? Not "for writing" — "caption this," "turn this riff into a post," "draft me an IG caption." If you have to slash-call every time, the description is too vague.

If you fail: Rewrite the description. Use the exact phrases you'd type when asking for this skill. Include what it's NOT for too.

When to run this checklistRight before you click save in Skill-Creator. Not during the prompt-build phase — there you're iterating, not gate-keeping. The pressure test is the last beat before a prompt becomes a permanent fixture in your stack.

If You Fail 2+ Checks

Don't Save Yet

If you're failing 2+ of these checks, the prompt isn't ready to be a skill. Run it 2-3 more times as a one-off. Iterate. The Auto-Feedback Loop will help.

If You're Failing 0-1

Fix that one. Save the skill. The other 6 will catch what slipped. Iterate via Auto-Feedback Loop in the first 5-10 uses — most skills don't ship perfect, they ship and learn.

A bad prompt costs you a chat. A bad skill costs you a habit.