Copied to clipboard!
Module 5 Level 2: Applied AI Design Systems & Skills intermediate 30 min

Auditing Design Skills

Three gates: description, methodology, agent-readiness.

What you'll be able to do

  • Audit a design skill against the three-gate rubric
  • Score each gate PASS, PARTIAL, or FAIL with quoted evidence
  • Produce a go/no-go verdict and a prioritized fix list

Audit a Design Skill Against the Three Gates

You found (or wrote) an AI design "skill" and you want to know if it is actually good before you leverage it on real work. You audit it against three gates: description quality, methodology completeness, and agent-readiness. AI helps you score it fast and surface the gaps. You stay the art director who decides whether it passes and what to fix first.

Recommended tool

claude

Good at reading a skill against a rubric and pointing to the exact line that fails a gate. Keep the skill text in the chat so every finding cites where the problem lives.

Bring to the exercise

  • The design skill you want to audit (its description and instructions)
  • What you actually want the skill to do for you
  • One example of a good output you would expect from it
Step 1

Load the Skill and the Rubric

Set the three gates before any scoring so the audit is structured, not vibes.

I am auditing an AI design skill against three gates. Confirm the rubric before I paste the skill:

GATE 1, DESCRIPTION QUALITY: Is it clear what the skill does, when to use it, and when NOT to? Would the right person know to reach for it?
GATE 2, METHODOLOGY COMPLETENESS: Does it lay out real steps, inputs, and outputs? Could someone follow it and get a consistent result, or is it hand-wavy?
GATE 3, AGENT-READINESS: Could an AI assistant actually run this end to end? Are inputs, outputs, and guardrails explicit, or does it assume a human fills gaps?

Score each gate PASS / PARTIAL / FAIL with a one-line reason. Do not invent strengths it does not have.

Confirm the rubric, then I will paste the skill.
What good output looks like
  • Restates the three gates clearly.
  • Agrees to PASS / PARTIAL / FAIL with reasons.
  • Commits to not inventing strengths.

Verify before using AI's output

Step 2

Score Gate 1: Description Quality

Check whether the skill clearly states what it does and when to use it.

What to substitute before pasting

  • [PASTE THE DESIGN SKILL: DESCRIPTION + INSTRUCTIONS] The full text of the design skill you are auditing.
  • [WHAT YOU WANT IT TO DO] In your own words, the job you want this skill to do.
Here is the skill:
[PASTE THE DESIGN SKILL: DESCRIPTION + INSTRUCTIONS]

Here is what I actually want it to do for me: [WHAT YOU WANT IT TO DO]

Score GATE 1 (Description Quality). Tell me:
- Is it clear what it does, when to use it, and when not to?
- Quote the exact lines that earn or lose the gate.
- One concrete fix to raise the score.
What good output looks like
  • A PASS / PARTIAL / FAIL with quoted lines.
  • Names whether "when NOT to use it" is present.
  • One concrete fix, not a vague suggestion.

Verify before using AI's output

Step 3

Score Gates 2 and 3: Methodology and Agent-Readiness

Check whether the steps are complete and whether an AI could actually run it.

Now score GATE 2 (Methodology Completeness) and GATE 3 (Agent-Readiness) the same way.

For GATE 2: are the steps, inputs, and outputs real and followable? Where would two people get different results?
For GATE 3: could an AI run this end to end? Are inputs, outputs, and guardrails explicit, or does it quietly assume a human fills the gaps?

Quote lines. Give each a PASS / PARTIAL / FAIL and one concrete fix.
What good output looks like
  • Separate scores for Gate 2 and Gate 3, each with quoted lines.
  • Points to where two people would diverge (Gate 2).
  • Names the implicit human assumptions (Gate 3).

Verify before using AI's output

Step 4

Decide and Prioritize

Turn three scores into a go/no-go and a short, ordered fix list you own.

Summarize the three gate scores in one table. Then give me:
- A verdict: USE AS-IS, USE WITH FIXES, or DO NOT USE YET.
- The top three fixes in priority order, highest impact first.

I make the final call. Lay out the evidence so I can decide, do not soften the verdict to be nice.
What good output looks like
  • A clean three-gate summary table.
  • A clear verdict, not a hedge.
  • Three prioritized fixes, highest impact first.

Verify before using AI's output