Cringe check

Skill cabbagecachekid/red-pen/skills/cringe-check

Keep AI-assisted writing honest and yours: an AI-written check, a cringe check, and a full-review orchestrator (anti-inflation + fact passes). Composable Claude Agent Skills.

Install

npx -y skills add cabbagecachekid/red-pen --skill cringe-check

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Audits professional and application copy for tone and positioning that reads as cocky, parroting, or presumptuous — solo-hero framing where the author is the only agent, positioning-against unnamed others to look smart by contrast, parroting a job description's distinctive phrases verbatim, prescribing a client's reality or priorities in a proposal, undermining a polished deliverable with unsolicited "happy to convert it" offers, and overclaiming authority or flattening hobby and enterprise work into false parallel. Use when reviewing a resume, cover letter, portfolio page, proposal, or pitch, or when the user asks for a "cringe check" or worries their copy sounds arrogant, try-hard, or like it is sucking up.

SKILL.md

4.7 KB, as published. Nobody here has run it

Cringe Check

Audit copy for the positioning moves that make a smart, qualified author look cocky or try-hard. This is a tone and stance check — it looks at how the author positions themselves relative to other people, the reader, and the truth. It does not touch sentence mechanics (that is the ai-written-check skill).

The frame to hold: the strongest copy reads as a generous collaborator who is genuinely good, not a solo hero or a sage on a mountain. Most cringe comes from the author accidentally instrumentalizing other people or claiming more than they own.

How to run it

Walk the six dimensions below. For each hit, output: the dimension name, the offending line quoted, and a collaborator-voice rewrite. Be specific — a real rewrite, not "soften this."

The six dimensions

1. Solo-hero check. Does every story use "I" as the only agent? Are partners — PMs, engineers, designers, leadership, participants — instrumentalized or invisible? Fix: name collaborators with their own agency.

2. Positioning-against check. Does the copy set up unnamed others (engineers, leadership, "most designers," "stock tools") as deficient so the author looks smart by contrast? Fix: let contrasts come from honest difference, not implicit putdown.

3. Curiosity check. Does the copy show any open question, uncertainty, or genuine interest in the reader's specific work — or is it all answers and claims? All-answers reads as closed and cocky.

4. Generosity check. Are the people in the stories collaborators, or background characters receiving the author's brilliance? Credit shared wins as shared.

5. Live-question check. Does the author name something they're still figuring out? (Especially load-bearing for research / AI roles, where admitting a live uncertainty is the credential.)

6. JD-mirror check. Does the copy lift the job description's most distinctive verbs or slogans verbatim ("kill bad ideas," "ship fast," "strategy-level not feature-level")? Verbatim lifts read as parroting. Soft echoes — same concept, the author's own words — are fine and usually desired.

Two more, for client / proposal copy

7. Don't-prescribe. In a proposal, don't tell the client what their reality is — "the highest-risk feature in your product," "the biggest open question is whether users will pay," "if they don't trust it they'll leave." That reads as a stranger diagnosing their business. Bring craft (methodology), relevant market understanding, and relevant experience instead. Describe what the work can surface; don't preview its conclusions.

8. Don't-undermine-bespoke. When sending a polished, hand-built artifact (a custom deck, a tuned layout), don't tack on "happy to convert this to Figma/Notion/slides if easier." It undercuts the thing you just made and volunteers free work before the engagement is scoped. The format you sent is the format. (Format flexibility is fine when the artifact is genuinely rough — a draft doc, a quick recording.)

Overclaim and scale (runs through all of the above)

Authority you don't have: "leads," "owns," "decides," "built from scratch," "established the function" — only if literally true. Contributor ≠ owner. Early adopter ≠ the person who sets strategy.
False parallelism: do not use matched grammar (parallel verbs, parallel clauses) to imply that a weekend project and a multi-year enterprise effort are the same scale. Parallel grammar implies parallel scope to the reader. Keep different scales in different scopes.
If asked to make a small thing sound bigger, decline and surface the scale issue rather than inflating it.

For worked examples and the reasoning behind each dimension, see references/POSITIONING.md.

What this skill does NOT do

It does not check sentence mechanics or AI tells — use ai-written-check.
It does not verify facts — that belongs to full-review.
It never inflates. When the honest version is smaller, it says so.