voice-calibration — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited voice-calibration (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Build a voice profile from live writing samples typed in the terminal. The profile captures sentence rhythm, rhetorical habits, vocabulary tendencies, and tonal signatures — then saves it as a reusable reference for future content generation.
AI-assisted writing blends the author's voice with the model's defaults. To capture authentic voice, every sample must be typed fresh by the user in the terminal with no AI drafting or editing. Quantity matters less than authenticity — 6-8 raw responses are enough.
Round 1 (General) → 4 prompts — capture base voice
Round 2 (Domain) → 4 prompts — capture subject-matter tone
Analysis → extract patterns
Voice profile → save to fileAsk these one at a time using AskUserQuestion. Wait for each response before asking the next. Do NOT paraphrase or suggest answers.
You have 3-4 sentences. What's something most people in your field get wrong?
Explain something you understand deeply to someone who knows nothing about it. 3-5 sentences.
Think of something you read recently that made you stop and think. What was it, and what was your reaction? Write it as if you're telling a friend.
Describe two things that most people think are similar but you think are fundamentally different. 2-4 sentences.
If --domain was passed, use that topic. Otherwise ask:
What domain will this voice profile be used for? (e.g., I-Ching commentary, product marketing, technical writing)
Then ask these one at a time, tailored to the stated domain:
Write the opening 2-3 sentences of a post about [domain topic]. Assume the reader doesn't care yet — make them care.
What's a common misconception in [domain]? Correct it in 3-4 sentences the way you'd actually say it.
Take something complex in [domain] and compress it to 1-2 sentences. Make it land.
Write the last 2-3 sentences of a post. Leave the reader with something that stays.
After collecting all 8 samples, analyze them silently. Do NOT show intermediate analysis to the user. Extract:
Generate a voice profile document with this structure:
# Voice Profile: [User Name or Handle]
Generated: [date]
Domain: [domain]
Samples: [count]
## Summary
[2-3 sentence distillation of the voice — what makes it recognizable]
## Sentence Rhythm
[Findings from sentence mechanics — concrete observations with examples from samples]
## Rhetorical Habits
[Patterns with direct quotes from samples as evidence]
## Vocabulary & Register
[Word-level observations]
## Tonal Signature
[The overall feel — how this writer comes across]
## Rules for Matching This Voice
1. [Concrete, actionable rule]
2. [Concrete, actionable rule]
3. [Concrete, actionable rule]
4. [Concrete, actionable rule]
5. [Concrete, actionable rule]
6. [Concrete, actionable rule]
## Anti-Patterns (Do NOT)
1. [Thing to avoid that would break the voice]
2. [Thing to avoid that would break the voice]
3. [Thing to avoid that would break the voice]
## Sample Sentences in This Voice
[3 original sentences written BY THE MODEL that demonstrate the captured voice — the user should read these and confirm they sound right]{skill-base-dir}/profiles/[handle-or-name].mddocs/voice-profile.md)Other skills and prompts can reference the saved profile. When generating content in the user's voice:
The profile is a living document — the user can re-run this skill anytime to update it with fresh samples.
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.