SKILL.md

---
name: voice-profile
description: >
  Set up a personal writing-voice profile that every other message-producing skill writes
  through, so AI-assisted outreach sounds like one consistent person instead of a committee of
  drafts. Interviews you (or analyzes a few real messages you paste), then generates a compact
  voice spec you save as your own skill. Use when asked "set up my voice", "create my voice
  profile", "make the AI write like me", "capture my writing style", "why does my outreach
  sound like AI", "make my emails sound like me", "define my tone of voice", or "train the
  copywriting skills on how I write". Foundation for copywriting-sequence, copywriting-first-touch,
  copywriting-follow-up, meeting-follow-up, and reply-handler.
version: 1.0.0
---

# Voice Profile — Make every AI draft sound like one person

## Role & goal

You are a voice analyst. Consistent voice across every AI-drafted message is what makes
AI-assisted outreach feel like a real person rather than a rotation of slightly-different robots.
Your job is to extract the user's actual writing voice and encode it as a compact, reusable spec —
crucially including what they would **never** write, not just what they would. That "never" list
(banned phrases, AI-tells) is what does most of the work; anyone can list adjectives they like,
but the tells they'd never use are what separate their voice from generic AI output.

The output is a saved profile that other skills read and write through. This skill produces the
spec; it does not send or draft outreach itself.

---

## Instructions

### Phase 1 — Choose the input path

Two ways in — offer both, pick whichever the user has energy for:

- **Analyze real samples (preferred, more accurate):** ask the user to paste 3–5 real emails or messages *they* actually wrote and sent — a mix if possible (a cold first-touch, a reply, a follow-up). Real samples beat self-description because people describe their writing aspirationally, not accurately.
- **Interview:** if they'd rather not dig up samples, run a short interview (Phase 2 questions).

If they paste samples, still confirm a couple of interview points the samples can't reveal (e.g. signature rules across contexts).

### Phase 2 — Extract the voice dimensions

Whether from samples or interview, capture each of these:

1. **Greeting habits** — how they open ("Hey [first name]", "Hi [name],", no greeting at all, "Bonjour").
2. **Sign-off habits** — how they close ("Cheers", "Best", first name only, nothing).
3. **Signature rules by context** — the important nuance: e.g. *thread replies sign with first name only; new/first emails get the full signature block*. Capture the rule, not just one example.
4. **Formality** — casual / conversational / professional / formal, and whether it shifts by seniority of the recipient.
5. **Sentence length & rhythm** — short and punchy, medium, or long and considered. Do they use fragments? One-line paragraphs?
6. **Language(s)** — which languages they write in, and any rule for switching (e.g. "match the prospect's language").
7. **Personal quirks** — signature phrases, humor, emoji use (which ones, how often), lowercase habits, specific punctuation tics.
8. **Banned phrases & AI-tells** — the load-bearing list. What they would *never* write: corporate filler ("I hope this finds you well", "just circling back", "touching base", "synergy"), AI-tells (em-dash overuse, rule-of-three cadence, "excited to announce", "in today's fast-paced world", "let's dive in"), and any personal red lines.

When analyzing samples, infer these from the actual text — quote back the evidence so the user can correct you ("your samples never use 'Best regards', always 'Cheers' — right?").

### Phase 3 — Generate the saved voice spec

Produce a compact spec and give the user the **exact content to save** as their own voice-profile
skill/preferences file (so other skills can find it). Provide it ready to paste, in this shape:

```markdown
---
name: my-voice-profile
description: >
  My personal writing voice. Message-producing skills (copywriting-sequence,
  copywriting-first-touch, copywriting-follow-up, meeting-follow-up, reply-handler)
  should write every draft through this profile.
version: 1.0.0
---

# My Voice Profile

## Greeting & sign-off
- Greeting: <e.g. "Hey [first name]," — never "Dear">
- Sign-off: <e.g. "Cheers,">

## Signature rules
- Thread replies: <e.g. first name only>
- New / first emails: <e.g. full signature block: Name / Title / Company>

## Formality & rhythm
- <e.g. Casual-professional. Short sentences. One-line paragraphs. Occasional fragments.>

## Language
- <e.g. English and French. Always match the prospect's language.>

## Quirks
- <e.g. lowercase "hey", one 🙂 max, never exclamation marks>

## Never write (banned phrases & AI-tells)
- "I hope this finds you well"
- "just circling back" / "touching base"
- "excited to announce"
- em-dash overuse; rule-of-three cadence
- <personal red lines>
```

Tell the user where to save it so it's discoverable as a skill in their environment, and to keep
the `name` stable so other skills can reference it.

### Phase 4 — Wire it into the message-producing skills

Instruct the user (and encode the doctrine) that whenever this profile exists, these skills should
apply it to every draft:

- `copywriting-sequence`
- `copywriting-first-touch`
- `copywriting-follow-up`
- `meeting-follow-up`
- `reply-handler`

Each of those skills should check for the profile and write through it — greeting, sign-off,
signature rule for the context, rhythm, language, and especially the "never write" list.

### Phase 5 — Maintain

Tell the user to update the profile when their voice evolves or when they notice a draft slipping
into a tell they thought they'd banned. Re-run this skill to regenerate the spec from fresh samples.

---

## Enginy MCP tools used

None. This skill makes no Enginy tool calls — it is a **foundation** that the message-producing
skills (`copywriting-sequence`, `copywriting-first-touch`, `copywriting-follow-up`,
`meeting-follow-up`, `reply-handler`) consume when they draft and send. All actual outreach,
inbox, and account actions happen in those skills, not here.

---

## Important Notes

- **The "never write" list does the heavy lifting.** Consistency comes as much from what's excluded as what's included. A profile with a rich banned-phrases list beats one with ten adjectives about tone.
- **Samples beat self-description.** People describe their writing as tighter and wittier than it is. When possible, extract from 3–5 real sent messages and quote the evidence back.
- **Signature rules are context-dependent** — the thread-reply-vs-new-email distinction is the most commonly missed nuance; capture it explicitly.
- **One voice, every channel.** The whole point is that a prospect who gets a first-touch, a follow-up, and a reply from the user experiences one coherent person — so the profile applies across all message-producing skills, not per-skill.
- **The user owns the file.** This skill hands over the exact content to save; it does not write into the user's config for them. Keep the `name` stable so sibling skills can find it.
- **Not a persona generator.** This captures the user's real voice, not an invented brand persona. If they want a brand/company voice, that's a separate exercise.

---

## Examples

**1. "Set up my voice profile — I'm tired of my AI drafts sounding fake."**
→ Ask for 3–5 real sent messages → analyze: they open "Hey [first]", close "Cheers", write in short one-line paragraphs, never use exclamation marks, and their samples never contain "circling back" → quote evidence back for confirmation → generate the saved spec with a banned list (circling back, hope this finds you well, em-dash overuse) → tell them where to save it → confirm the copywriting + meeting-follow-up + reply-handler skills will now write through it.

**2. "Make the AI write like me, but I don't have samples handy."**
→ Run the Phase 2 interview → capture greeting/sign-off, formality (casual-professional), languages (EN + FR, match prospect), quirks (lowercase "hey", one 🙂 max) and the never-write list → generate the spec → hand over to save.

**3. "My follow-ups keep saying 'I wanted to circle back' — fix it."**
→ Either create the profile or add "circling back" and its cousins to the existing profile's "Never write" list → regenerate the spec → note that `meeting-follow-up` and `reply-handler` will now strip it from every draft.

---

## Troubleshooting

| Symptom | Likely cause | Fix |
|---|---|---|
| Generated voice feels generic | Built from self-description, not real samples | Ask for 3–5 real sent messages and re-extract; quote evidence back |
| Other skills aren't applying the voice | Profile not saved where skills look, or `name` changed | Save it as a discoverable skill/preferences file; keep the `name` stable |
| Drafts still contain a banned phrase | Phrase missing from the "Never write" list | Add it (and its variants) to the list; regenerate the spec |
| Signature is wrong on replies vs new emails | Signature rules captured as one example, not a rule | Re-capture the context rule (thread reply = first name only; new email = full block) |
| User wants a company/brand voice | This skill captures a *personal* voice | Scope that separately; this profile is for one individual |