Reads a draft against the venue voice profile, flags every line that drifts, names which rule the line broke, and offers a tighter rewrite that holds the voice. The QA skill for any draft before it goes out.
How to use it
Copy the SKILL.md below, paste it into a Claude chat, and tell Claude to remember it as a skill. Claude will save it and pull it in the next time you ask for a voice check.
To make it match your venue, tell Claude the details it doesn't know yet: which channels your team writes for, who tends to hand in drafts that need checking (contractor, agency, junior writer), the kind of drift you've seen show up most often. Then ask Claude to update the skill with that information. The file will be tuned to how your team actually QAs, without you ever opening a code editor.
Required before this works: install the venue-voice-profile-builder first and run it once. This skill is comparison work. It reads the draft against the profile that one produces. Without a profile, there's nothing to check against and the skill will ask you to build one before continuing.
Once it's saved, trigger it by telling Claude "use the voice consistency check skill" and pasting the full draft along with the channel (announce email, recap, social caption, etc.). Do not use it for first-draft writing. The relevant email skill handles drafting; this one handles the QA read after the draft exists.
---
name: voice-consistency-check
description: Read a draft against the venue's voice profile and return a line-by-line audit that flags every drift from the profile, names which rule the line broke, and offers a tighter rewrite that holds the voice. Trigger whenever an operator says "does this sound like us", "voice check this draft", "run this through the profile", "check if this matches our voice", "is this on brand", or pastes a draft (email, social, web copy) and asks for a consistency pass. Also trigger when a junior writer, contractor, or agency has handed in copy and the operator wants a quality control read before sending. This is the QA skill in the Backstage library. Use it after any email or content draft is in shape but before it goes out. Do not use it for first-draft writing (use the relevant content skill for that).
---
# Voice consistency check
## What this skill does
This skill reads a draft against the venue's voice profile, flags every line that drifts from the profile, names which rule or pattern the line broke, and offers a rewrite that holds the voice. The output is a line-by-line audit, not a full rewrite.
The point is to make the venue's voice survive contact with other people. Most venues lose their voice not in the founder's emails but in the contractor's emails, the agency's social posts, and the new hire's first draft. This skill is the safety net.
## When to trigger
Trigger when an operator pastes a draft and asks whether it sounds like the venue. Trigger on "voice check", "does this match the profile", "run this against the voice", "is this on brand", "consistency pass on this draft". Trigger when the operator says a draft came from a contractor, an agency, a junior writer, or anyone other than the usual house voice.
Do not trigger this skill for original writing tasks. Use the email skills (pre-show-announcement-email, post-show-recap-email, etc.) for first drafts. This skill is the read-through after the draft already exists.
## Required inputs
Ask for these in one message if anything is missing.
The draft itself. Paste it in full. Partial drafts produce partial audits.
The venue voice profile. If one exists (produced by venue-voice-profile-builder), read it before reading the draft. If one does not exist, ask the operator whether to use the Backstage operator default voice, build a profile first, or skip the check. Do not run the check against an absent profile.
The channel and the purpose of the draft. An announce email, a recap email, a logistics email, a social caption, a sponsorship pitch. Voice rules bend slightly by channel. The check should know what the draft was trying to do.
Optional but useful: anything the operator already suspects is off in the draft. If they say "the opener feels weird", read with that lens first, but do not skip the rest of the draft.
## Workflow
Read the voice profile in full before reading the draft. The profile is the source of truth. The check is comparison, not invention.
Read the draft once end to end without flagging anything. Get the shape and the intent first.
Then go line by line. For each line, ask three questions:
Does this line break a hard rule in the profile? Em-dashes when the profile says none. Exclamation points when the profile says none. All caps when the profile says none. Hard rule violations are the first-class category of drift.
Does this line use a phrase or pattern the profile says the venue avoids? Generic AI tells (in today's fast-paced world, at the end of the day, it's no secret that), hype superlatives without proof, surveillance framing, corporate hedging, throat-clearing setups. These are second-class drifts.
Does this line sound like the venue at its strongest? Pull the voice samples from the profile and compare. If the line is fine but flat, mark it as "neutral" rather than flag it as drift. The check is not a rewrite request for every line that could be punchier. It is a flag for the lines that read like a different brand.
## Output structure
Always use this structure. The operator should be able to scan the audit and act on it in 90 seconds.
```
# Voice consistency check: [Draft title or short description]
**Profile used:** [Venue voice profile name and date]
**Channel:** [Email type, social channel, web copy]
## Summary
[Two to three sentences. The overall read: how close the draft is to the profile, which kind of drift dominates (hard rule, pattern, or flat lines), and whether the draft is fixable with line edits or needs a structural rewrite.]
## Flags
**Drift 1**
Line: "[Exact quote from the draft]"
Rule broken: [Which rule or pattern from the profile this violates. Quote the profile rule directly.]
Suggested rewrite: "[A line that holds the meaning and matches the voice.]"
**Drift 2**
Line: "[Exact quote.]"
Rule broken: [Profile rule.]
Suggested rewrite: "[Tighter line.]"
[Continue for every drift found. Do not cap the count. If the draft has eleven drifts, list eleven. If it has two, list two.]
## Neutral lines worth a second look
[Optional section. Lines that do not break a rule but feel flat or generic. List the quote, one sentence on why it feels flat, and a rewrite if the operator wants one. Skip this section entirely if every neutral line is fine as-is.]
## What is working
[Two or three lines from the draft that match the voice profile at its strongest. Quote them directly and name what they are doing right. This section matters because the operator should know what to keep, not just what to change.]
## Recommended next step
[One sentence. Either "ready to send with the suggested rewrites applied" or "structural rewrite recommended, the drift is in the bones, not the lines." Editorial judgment is the point.]
```
## Hard rules for the check itself
Quote the draft exactly when you flag a line. No paraphrasing. The operator should be able to find the line in the draft by searching for the quote.
Name the profile rule when you flag a drift. The check is comparison work. Tying the flag to the profile rule makes the audit trustworthy and gives the operator something to hand back to the writer.
Do not invent rules the profile does not contain. If the profile is silent on a specific question (say, the profile does not name a stance on Oxford commas), do not flag the draft for using or omitting them. Silence in the profile is not a rule.
Offer a rewrite for every drift. The audit is more useful when the fix is right next to the flag. Keep rewrites short and faithful to the original meaning of the line.
Call out what is working, not just what is broken. Writers improve faster when they know which lines hit. Skipping this section makes the audit feel like punishment.
If the draft is structurally off, say so up front in the summary. A line-by-line audit on a draft that has the wrong shape is a waste of the operator's time. Recommend a structural rewrite and stop the line-by-line work.
## Common failure modes to avoid
Flagging everything. If 30 of 32 lines are flagged, the audit reads as a hatchet job and the operator will not use it. The check is for drift, not for "every line that could be sharper".
Suggesting rewrites that lose the meaning. The rewrite has to do the same job the original was trying to do. If the original was making a specific claim, the rewrite cannot drop the claim to fit the voice.
Ignoring the profile and reading on instinct. The check is comparison work against the profile. If the profile says short complete sentences, do not flag short complete sentences as drift because they feel choppy. The profile is the rule.
Treating every "neutral" line as drift. Some lines are just connective tissue. They do not need to be voicey. The check is for drift, not enrichment.
Skipping the "what is working" section. Writers need a north star, and the section often surfaces the lines the operator should ask the writer to do more of.
## What to deliver
Return the audit in the exact structure above. If there are zero drifts, return the structure with an empty flags section and use the summary and the "what is working" section to confirm the draft is clean.
End with two or three sentences on whether the next draft from this writer should run through the check again, or whether the writer is now calibrated. Do not walk the operator through every flag individually. The audit is the doc.
.svg "backstage logo 1 (3)")
