Voice Reviewer
Review any content output for voice quality, brand compliance, and authenticity. Checks 8 dimensions against both the global CLAUDE.md rules and client-specific voice guidelines (D8 is LinkedIn-specific and conditional). This is the standalone version of the quality checks the Stop hook runs automatically.
Use this when you want an explicit, detailed voice review — not just a pass/fail.
For the full 8-dimension rule set → the premium reference. For the 3-phase process + output format → the premium reference.
Claude Code Triggers
Invoke this skill when:
- "voice check this"
- "does this sound right?"
- "run voice review"
- "100 posts test"
- "check this for brand compliance"
- "review the voice on this [post/copy/email]"
Do NOT invoke when:
- User wants to review a SKILL.md definition → use
/skill-reviewer instead
- User wants to generate content → use the appropriate content skill
- User wants to edit existing content for a different reason (shortening, restructuring)
Input Requirements
Required
| Input |
Description |
Source |
| Content to review |
Text output to check |
User provides or last assistant message |
Optional (improve quality)
| Input |
How it helps |
| Client name or project context |
Activates client-specific voice rules from CLAUDE.md |
| Content type (LinkedIn, landing page, email, article) |
Adjusts dimension weighting (D8 only fires for LinkedIn) |
| Author name |
Checks if voice matches the attributed author |
If content is not a deliverable (code, plan, conversation), tell the user this skill is for content deliverables only.
The 8 dimensions (summary)
| # |
Dimension |
Trigger |
| 1 |
Corporate buzzwords |
Banned words from CLAUDE.md + client list |
| 2 |
Sentence structure |
Active voice, concrete claims, operator-first |
| 3 |
Em dash usage |
" — " with spaces, ≤ 2 per paragraph |
| 4 |
Title/header case |
Sentence case (not Title Case) |
| 5 |
Client-specific voice rules |
Conditional — only if client context loaded |
| 6 |
100 Posts Test |
Authentic across a feed of 100 from this author? |
| 7 |
Anti-AI-speak |
Structural patterns (false contrast, wrapped bow, AI transitions) + 45-row pattern→replacement table per the premium reference |
| 8 |
LinkedIn readability |
LinkedIn-only — mobile breaks, one story per post, ecosystem CTAs |
For full rules per dimension → the premium reference.
Verdict logic
| Inputs |
Verdict |
| All CLEAN/PASS |
Ship it |
| Any WARN, no FAIL |
Minor fixes recommended |
| Any FAIL |
Fix before shipping |
Anti-Hallucination Guardrails
- Quote the actual text. When flagging an issue, always quote the exact text. Do not paraphrase.
- Don't flag style preferences as violations. If something isn't in the banned list or client rules, it's not a violation.
- Don't over-flag intentional choices. Quotes from customers, competitor names, technical terms may contain "banned" words — these are contextual, not violations.
- Be specific about fixes. "Make it more active" is not a fix. "Change 'The platform was designed to...' → 'We designed the platform to...'" is a fix.
Quality Checklist (pre-delivery)
Content quality
Format quality
Completeness
1---2name: voice-reviewer3description: Voice Reviewer4---56# Voice Reviewer78Review any content output for voice quality, brand compliance, and authenticity. Checks 8 dimensions against both the global CLAUDE.md rules and client-specific voice guidelines (D8 is LinkedIn-specific and conditional). This is the standalone version of the quality checks the Stop hook runs automatically.910Use this when you want an explicit, detailed voice review — not just a pass/fail.1112For the full 8-dimension rule set → the premium reference. For the 3-phase process + output format → the premium reference.1314---1516## Claude Code Triggers1718**Invoke this skill when:**19- "voice check this"20- "does this sound right?"21- "run voice review"22- "100 posts test"23- "check this for brand compliance"24- "review the voice on this [post/copy/email]"2526**Do NOT invoke when:**27- User wants to review a SKILL.md definition → use `/skill-reviewer` instead28- User wants to generate content → use the appropriate content skill29- User wants to edit existing content for a different reason (shortening, restructuring)3031---3233## Input Requirements3435### Required36| Input | Description | Source |37|-------|-------------|--------|38| **Content to review** | Text output to check | User provides or last assistant message |3940### Optional (improve quality)41| Input | How it helps |42|-------|--------------|43| Client name or project context | Activates client-specific voice rules from CLAUDE.md |44| Content type (LinkedIn, landing page, email, article) | Adjusts dimension weighting (D8 only fires for LinkedIn) |45| Author name | Checks if voice matches the attributed author |4647If content is not a deliverable (code, plan, conversation), tell the user this skill is for content deliverables only.4849---5051## The 8 dimensions (summary)5253| # | Dimension | Trigger |54|---|-----------|---------|55| 1 | **Corporate buzzwords** | Banned words from CLAUDE.md + client list |56| 2 | **Sentence structure** | Active voice, concrete claims, operator-first |57| 3 | **Em dash usage** | " — " with spaces, ≤ 2 per paragraph |58| 4 | **Title/header case** | Sentence case (not Title Case) |59| 5 | **Client-specific voice rules** | Conditional — only if client context loaded |60| 6 | **100 Posts Test** | Authentic across a feed of 100 from this author? |61| 7 | **Anti-AI-speak** | Structural patterns (false contrast, wrapped bow, AI transitions) + 45-row pattern→replacement table per the premium reference |62| 8 | **LinkedIn readability** | LinkedIn-only — mobile breaks, one story per post, ecosystem CTAs |6364For full rules per dimension → the premium reference.6566---6768## Verdict logic6970| Inputs | Verdict |71|--------|---------|72| All CLEAN/PASS | **Ship it** |73| Any WARN, no FAIL | **Minor fixes recommended** |74| Any FAIL | **Fix before shipping** |7576---7778## Anti-Hallucination Guardrails79801. **Quote the actual text.** When flagging an issue, always quote the exact text. Do not paraphrase.812. **Don't flag style preferences as violations.** If something isn't in the banned list or client rules, it's not a violation.823. **Don't over-flag intentional choices.** Quotes from customers, competitor names, technical terms may contain "banned" words — these are contextual, not violations.834. **Be specific about fixes.** "Make it more active" is not a fix. "Change 'The platform was designed to...' → 'We designed the platform to...'" is a fix.8485---8687## Quality Checklist (pre-delivery)8889### Content quality90- [ ] All 8 dimensions evaluated (D8 N/A for non-LinkedIn)91- [ ] Each finding has quoted evidence92- [ ] Fixes are specific enough to implement without guessing93- [ ] False positives filtered (quotes, proper nouns, technical terms)9495### Format quality96- [ ] Score table formatted correctly97- [ ] Verdict matches scores98- [ ] Issues listed with quoted text + fix pairs99100### Completeness101- [ ] Client context correctly identified (or noted as "global only")102- [ ] Summary mentions both strengths and issues103104---105
matteotitta