- Home
- Skills
- Writing & Content
- Voice Checker: AI-Tell Review for Lyrics and Prose
More screenshots
Works with the AI Agents you already use
Voice Checker: AI-Tell Review for Lyrics and Prose
An authenticity audit for lyrics and prose: three structured reads that flag AI-writing tells with quoted evidence and questions — never rewrites, never a verdict.
$5
Voice Checker: AI-Tell Review for Lyrics and Prose
Example session with this skill installed
Audit this before it goes public. Hook from my new track: "Through the darkness and the pain I rise above it all / demons in my rearview, I'm answering the call." And the release blurb: "A groundbreaking new chapter, this record weaves together a tapestry of late-night stories. It was recorded in a single night and can be heard as the sound of a city learning to breathe again."
- Read your context and instructions
- Compiled the voice checker
Three reads, run in order.
Ear read: the blurb moves at one even speed with no rough edges — every sentence complete, every transition smooth. In the hook, the second line explains the first line's feeling instead of extending it, and all four beats land with the same weight.
Fingerprint read. [warning] The abstraction stack: "darkness," "pain," and "demons" across the hook's two lines, with "demons" the only noun a camera could film. [warning] The borrowed carpet, twice in three blurb sentences: "weaves together" and "a tapestry of" — phrases that fit any record and therefore describe none. [warning] The press-release superlative: "groundbreaking" asserts a quality the blurb never once describes. [info] Inspiration poster: "rise above it all" appears once. In horrorcore that phrasing can be an honest genre nod, so it is flagged a single time, as a question.
Intent read. [info] Agency audit on the blurb: three passives in two sentences ("was recorded," "is woven," "can be heard"), which quietly removes the artist from a paragraph whose whole job is the artist. The hook's arc also resolves completely in two lines — wound, demons, triumph — with no thread left hanging.
Summary: 4 warnings, 2 infos across the submission; no fully clean section to name this time. Every finding in the full report is phrased as a question, the sharpest being: if the darkness in the hook had a shape, what would the camera actually film? Nothing was rewritten. The deliverable is the report; what changes is the writer's call.
Connects securely to your tools. The creator never sees your data.
About this skill
This skill answers one question about a piece of writing: would a reader believe a human wrote it? It does not grade craft, rhyme, or grammar. It hunts the habits that make machine-written text sound like machine-written text — and reports them as questions with evidence, so the writer decides what to keep.
The audit runs three separate reads, because each catches what the others miss. The ear read goes through the piece aloud, hunting too-smooth grammar in rough scenes, lines that perform and explain themselves at once, and metronome cadence that feels filled-in rather than decided. The fingerprint read hunts the specific moves language models overproduce: abstraction stacks (three big nouns doing the work of one image), decoded metaphors that translate themselves, inspiration-poster phrases, press-release superlatives, and borrowed-carpet phrases in prose that fit any artist and describe none. The intent read steps back to the whole piece: arcs that resolve too neatly, narrators so generic the song could be about anyone, throat-clearing openings, and passive constructions that remove the artist from their own story.
Every finding follows a fixed template — the quoted evidence, the pattern name, why it reads as machine-written, and a question, never a suggested replacement line. Severity is two levels only, warning and info, assigned by a decision table that weighs repetition, position, and genre: a pattern that is a convention of the stated genre is an info at most, and a choice the writer declares intentional is closed, not re-argued. Lyrics and prose are audited under different rule sets, always.
The report names the clean files too — knowing which tracks passed untouched is proof the writer's own voice is already on the page, and it is half the value of the audit.
The audit is advisory by design: it produces a report of findings and questions, never edits, scores, or pass/fail grades, and it will not rewrite a flagged line for you. It judges text against the conventions of the genre you state — give it the wrong genre and the findings will be calibrated wrong. Very short pieces (a two-line bio) yield thin evidence and the report says so. It also cannot tell you who or what actually wrote a text; it identifies patterns that read as machine-written to human audiences, which is a question about effect, not authorship.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 24 days ago
Permissions
No special permissions declared or detected