Menu
Extract a brand's unique voice from existing content, build a do/don't guide with specific language rules, and maintain tone standards to detect drift over time. Works with X, LinkedIn, Instagram, and TikTok.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Analyze A/B tests across hook variations, posting times, content formats, and CTAs — declare statistical winners, document learnings, and build an always-on optimization engine.
Track post-level metrics across all platforms — impressions, engagement rate, profile clicks, follows, link clicks — with weekly aggregation and trend detection for performance analysis.
Generate weekly and monthly performance summary reports — top/bottom posts, content type analysis, pillar performance, follower growth, conversion tracking, and actionable recommendations.
Write Instagram and LinkedIn captions with scroll-stopping hooks, value-dense bodies, platform-appropriate CTAs, and strategic hashtag sets — optimized for each platform's reading behavior and visual context.
Select and write the optimal call-to-action for any social post — follow, reply, DM, save, share, link, or signup — matched to platform, content type, content value, and audience psychology.
Generate 5-10 hook variations for any post, thread, or caption across X, LinkedIn, Instagram, and TikTok — using different styles (question, contrarian, statistic, story, observation, promise) so the best one can be tested or selected.
| name | brand-voice-system |
| description | Extract a brand's unique voice from existing content, build a do/don't guide with specific language rules, and maintain tone standards to detect drift over time. Works with X, LinkedIn, Instagram, and TikTok. |
| version | 2.0.0 |
| author | Crewm8 |
| maintainer | Gokul (github.com/gokulb20) |
| license | MIT |
| homepage | https://crewm8.ai |
| tags | ["social","strategy","brand-voice","tone","consistency","identity"] |
| related_skills | ["positioning-statement","content-pillars"] |
| inputs_required | ["social-post-samples","founder-interviews-or-voice-references","existing-brand-guidelines-if-any"] |
| deliverables | ["brand-voice-guide-document"] |
| compatible_agents | ["hermes","claude-code","droid","cursor","windsurf","openai","openclaw","generic"] |
| execution_tier | 2 |
| required_tools | ["FIRECRAWL_API_KEY"] |
| required_env_vars | ["FIRECRAWL_API_KEY"] |
Extract, codify, and maintain the brand's unique voice across all social content. This skill defines how the brand sounds so every post, reply, and DM feels like the same person wrote it — no ghostwriting whiplash, no platform dissonance.
A brand without a documented voice sounds like four different people on four different platforms. This skill gives you a single source of truth for tone, vocabulary, and platform-specific flex — so every piece of content reinforces the same identity.
| Component | What It Covers |
|---|---|
| Voice traits | 3-5 adjectives describing the brand (e.g., "warm, direct, playful, zero-BS") |
| Do guide | Specific language patterns, sentence structures, preferred words |
| Don't guide | Banned phrases, corporate jargon, cliches, competitor vocabulary |
| Platform flex | How the voice adapts per platform |
| Edge policy | Swearing, controversy, humor boundaries |
| Voice tests | 5 sample sentences to check new content against before publishing |
Gather source material. Collect the last 3 months of social posts, blog content, newsletters, podcast transcripts, founder interviews, and customer support replies. The more source material, the more accurate the voice extraction.
Analyze for patterns. Read everything and note:
Distill into 3-5 voice traits. Each trait gets:
Build the Do guide:
Build the Don't guide:
Define platform voice flex:
Write the voice test. 5 pre-written sentences that the agent can test any new content against before publishing. Each sentence tests a different voice dimension (tone, vocabulary, structure, energy, platform-fit).
Document edge policy explicitly:
# [Brand Name] Voice Guide v[date]
## Voice Traits
1. **[Trait]** — [Definition]. Example: "[real quote]" (not: "[counterexample]")
2. ...
## Do This
- [specific language pattern]
- [preferred phrase or construction]
- [opening pattern that works]
## Don't Do This
- [banned phrase or pattern]
- [banned format or word]
- [tone violation to catch]
## Platform Voice Flex
- **X:** [how the voice changes]
- **LinkedIn:** [how the voice changes]
- **IG:** [how the voice changes]
- **TikTok:** [how the voice changes]
## Edge Policy
- Swearing: [Yes / Light / No]
- Controversy threshold: [Describe — what requires founder approval?]
- Humor boundary: [Self-deprecating / Absurdist / None / Depends on topic]
## Voice Test
1. [Passing sentence] → [Why it passes]
2. [Failing sentence] → [Which voice dimension it violates]
3-5. ...
| Platform | Voice Characteristic |
|---|---|
| X | Fastest pace. Shortest sentences. Most direct. First 3 words carry the hook. |
| Slightly longer sentences. More structure. Still human — "corporate you" is banned. | |
| Visual-first; voice supports the image. Captions are warmer, more personal, emoji-tolerant. | |
| TikTok | Most casual. Caption is secondary to video script. Voice must feel like a friend talking. |
The skill is complete when:
Run 3 high-performing posts through the guide. They should match every rule. Run 1 post that felt wrong — it should flag specific violations. Show the guide to someone who knows the brand but didn't write the guide. Ask: "Does this sound like them?" If they hesitate, the extraction missed something.
Input: 12 recent X posts, 5 founder podcast transcripts, and a "voice is feeling off" complaint from a team member.
Process: The agent reads all source material, identifies that the founder writes long, thoughtful sentences on X that get truncated and lose momentum. The guide captures: "X voice = 1 thought, 1-2 sentences max. Save the nuance for LinkedIn." It also catches that the brand uses "honestly" as a crutch word (3+ times per post) — banned.
Output: A voice guide with 4 traits (Direct, Warm, Unpolished, Opinionated), a Do guide with 7 rules, a Don't guide with 7 entries, platform flex for all 4 platforms, and a voice test with 5 sample checks.