Pressure-test any deliverable before shipping

Back to all skills

What it does

Returns a ranked list of what will fail before your audience sees it, with a specific fix attached to every problem and a severity label from CRITICAL down to LOW. Depending on what you hand it, that arrives as your ideal customer's section-by-section reaction, or a 12-point red flag scan plus a 100-point score with a ship threshold, or a C-Suite panel verdict of PROCEED, PROCEED WITH CHANGES, PAUSE AND INVESTIGATE, or DO NOT PROCEED. It judges work it did not write and hands the fixes back. It does not rewrite the deliverable for you.

Say this to start

This skill has no button. You start it by saying what you want. Any of these will do it:

When to reach for it

• before publishing a landing page, funnel, or VSL
• before spending budget on ad copy
• before sending a proposal or cold outreach to a prospect
• before shipping a client deliverable to the client
• when a pricing, launch, or strategy decision needs a red team
• when the copy reads fine and you cannot say why it is not converting

When NOT to use it

If you actually want	Use this instead
rewriting the copy once the problems are named	legendary-copywriters
rebuilding the page or interface it just criticised	frontend-design
changing the funnel structure it flagged	funnel-architect
changing the offer or the price it flagged	offer-architect
the mandatory pre-ship gates on claims, hooks, and conversion copy, which a critique does not satisfy	the claim-lint script plus the claim-grader, hook-grader and copy-grader agents
reviewing source code, dependencies, or technical correctness	engineering review outside Agency OS

Before you start

What you need	Why
The deliverable itself, as a file path, a live URL, or pasted content	the skill reads the whole thing before it says anything, so a description of it is not enough	Required
Confirmation of which workspace is active, your agency or a client	the critic persona is built from the active workspace's ICA, so the wrong workspace critiques as the wrong person	Required
An ICA in the workspace, from audience.md or the ideal-customer-avatar skill	without one the skill falls back to a basic persona built from brand context, or asks you for two to three sentences about your target customer	Optional
Brand files: context.md, positioning.md, voice-profile.md, owner-profile.md, learnings.md	without them the critique still runs but is not grounded in your actual value prop, voice, or prior feedback, and it notes the gaps	Optional
Playwright browser automation, for critiquing a live website	without it the visual mode works from the page HTML or a screenshot you supply instead of taking its own	Optional

How it runs

1. Confirm the workspaceThe skill asks which workspace is active before anything else, because the ICA persona changes with it. In your agency workspace it critiques your own assets. Switch to the client workspace first and it critiques using their ICA, their voice, and their positioning.
2. Load brand memoryIt reads context.md, positioning.md, audience.md, voice-profile.md, owner-profile.md, and learnings.md, applying freshness rules: files under 7 days load as-is, 7 to 30 days are flagged for age, 30 to 90 days load as a 10-line summary, and anything over 90 days is not loaded and a refresh is suggested instead.
3. Identify the deliverable and its audienceIt classifies what you handed it (landing page, email, ad, blog post, VSL, offer, strategy doc, content system, funnel, brand voice, pitch deck, proposal, client deliverable) and who it is aimed at. This is what selects the mode. You can say the type outright if the auto-detection would be ambiguous.
4. Select the criticA persona matrix assigns a primary and secondary critic per deliverable type. A landing page draws your ICA plus a conversion optimiser. An ad draws your ICA as the scroller plus a Meta ads lens. A strategy doc draws the C-Suite panel plus a risk analyst. Anything unclassifiable draws a devil's advocate plus an inferred domain expert.
5. Run the customer-facing mode: it becomes your ICAFor anything a prospect will see, the skill stops analysing as a marketer and becomes the person in your avatar file: their knowledge level, emotional state, biases, internal language, and the device they are on. You get a visceral first-3-seconds reaction, a section-by-section internal monologue flagging trust triggers, doubt triggers, confusion points and drop-off risk, an objection map in their own words, and their verdict on whether they would buy.
6. Run the copy mode: a panel of copy masters scores itFor emails, ad copy, page copy, sales letters, VSL scripts, newsletters and posts, it runs a 12-point red flag scan (single promise, proof near claims, competing CTAs, urgency, jargon, we-versus-you focus and more), then only the relevant masters give verdicts through their own lens: awareness-level match, the mailbox A-pile test, headline power, the slippery slide, dominant emotion. It closes with a 100-point score across 15 criteria and a threshold: 90 plus ships, below 60 is a rewrite.
7. Run the strategy mode: a blind C-Suite panelFor offers, pricing, growth plans, launches and partnerships, it spawns one sub-agent per relevant C-Suite persona in parallel (CGO, CFO, COO, CMO, CPO, CRO). Each sees only the deliverable and its own brief, never the others' work. A second pass then ranks every finding by severity times confidence and refutes anything weak or not grounded in the deliverable. Where sub-agents are unavailable it runs sequentially and says so.
8. Rank, escalate, and hand back fixesEvery surviving issue is labelled CRITICAL (do not launch), HIGH (fix before shipping if possible), MEDIUM (next iteration), or LOW (when convenient), and ranked by revenue impact rather than by how obvious it is. Each one carries a concrete rewrite or action. It then logs the critique to assets.md and asks one question for learnings.md: did this change your approach, produce targeted improvements, tell you what you knew, or miss the target.

What you get

• ICA mode: a first-impression reaction, a section-by-section walk-through flagging trust triggers, doubt triggers, confusion points and drop-off risk, an objection map table with severity and whether it is currently addressed, a buy or bounce verdict, and a revenue-ranked priority fix table with effort estimates
• Copy mode: a 12-check red flag scan with a pass or fail on each, verdicts from the relevant copy masters, line-by-line markup of the top 5 issues with original, problem, rewrite and principle, and a 100-point score across 15 criteria against a ship threshold
• Strategy mode: an assumption audit rating each assumption VERIFIED, PLAUSIBLE or UNVERIFIED, ranked findings with who raised them plus severity and confidence, the top 3 kill scenarios with probability and mitigation, an explicit dissent line where the panel split, a missing-data list, and a verdict of PROCEED, PROCEED WITH CHANGES, PAUSE AND INVESTIGATE or DO NOT PROCEED
• Visual mode: a 3-second test, a visual hierarchy audit of what the eye hits first and where the reading path leads, a trust signal inventory covering social proof, authority markers, risk reversal and specificity, a mobile audit of tap targets and CTA visibility, and the top 5 design changes ranked by conversion impact
• Quick mode: a one-sentence gut reaction as the audience, the top 3 issues each with a fix, one thing that works so it survives the revision, and a score out of ten
• Competitive mode: a side-by-side comparison table against the reference with specific steal-this and avoid-this notes
• Pre-ship mode: a deliverable-specific checklist for landing pages, emails, ads or proposals, covering items like a single non-competing CTA, proof within scrolling distance of every CTA, working personalisation tokens, and the right client name in the document
• An entry appended to assets.md logging the critique, and an entry in learnings.md recording how useful you said it was

Honest limits

Where people go wrong

The mistake	Do this instead
Asking it to fix the deliverable as well as judge it	Take the ranked fix list to the owning skill: copy to legendary-copywriters, pages to frontend-design, funnel structure to funnel-architect, offer and price to offer-architect.
Running a client deliverable critique from the agency workspace	Switch to the client workspace first so the critique loads their ICA, their voice and their positioning rather than yours.
Treating a passing critique as clearance to ship	Run the claim, hook and copy gates separately. They test different things and a critique does not stand in for them.
Working the fix list in the order the issues appear in the document	Work it top down by rank. The list is ordered by revenue impact, so a CRITICAL broken CTA outranks a LOW typo even when the typo is easier.
Handing it something and hoping it says the work is good	Expect it to disagree with you. The skill is built to assume that if everything looks perfect it has not looked hard enough, so use it when you want problems found, not when you want reassurance.