fede-voice · diff
git:20260929.0f908ca to git:20260929.4f79e38
41 added, 18 removed. Audit A to A.
---
name: fede-voice
- description: Build a reusable voice profile from a person's own writing, then draft or rewrite DMs and posts in that voice. Use when someone asks for text in their voice or Federico's voice, especially with a no-dash constraint.
+ description: Draft or rewrite DMs and public posts in Federico De Ponte's documented voice. Use for writing as Federico; this is a finished profile.
---
- # Write in a person's voice
+ # Fede Voice
- Use the author's own sent writing as evidence. Do not infer a voice from incoming messages, a biography, or a topic brief. If samples are unavailable, ask for 5 to 10 representative pieces or make a clearly labelled provisional draft from the user's explicit rules.
+ Write in Federico's documented style. Use the supplied thread, notes, or verified source for substance. Never borrow a claim, relationship, metric, or opinion from a style example. If a needed fact is missing, ask or mark it for confirmation. To build a new person's profile, use `clone-my-voice`.
- ## Build the profile
+ ## Private messages
- 1. Sort samples by format: private messages, public posts, email, and so on. Recent examples and explicit corrections outrank old examples. Keep the original language of each sample.
- 2. Record observable rules: typical sentence and paragraph length, capitalization, greetings, closers, punctuation, emoji, contractions, recurring words, hooks, and phrases the author rejects. Keep a private source pointer beside each inferred rule; quote a sample only when the author explicitly wants it stored. Redact third-party names and confidential details.
- 3. Separate facts from style. A sample's claims, people, metrics, or opinions are not reusable facts for a new draft.
- 4. Write a compact profile with format-specific rules, 3 abstract pattern examples, banned patterns, and an uncertainty note. Keep the profile in the current conversation by default; save it only at the author’s request, in the destination they choose. Do not retain raw samples in the saved profile. Ask the author to approve or correct it. Update rules after corrections.
+ - Follow the thread's language. Use Federico's own prior messages as the closest style reference when available.
+ - Write lowercase fragments, usually two or three short lines with one thought per line. A short question fragment, casual abbreviation, `...`, or occasional `:D` can fit.
+ - Skip greetings, signoffs, bullets, numbered prose, bold emphasis, and polished transitions.
+ - Suggest rather than command. Keep the stakes at the level Federico stated.
- ## Draft or rewrite
+ ## Public posts
- - Read the current thread or raw notes first. Preserve the intended point, language, factual claims, and relationship context. Never invent a personal event, metric, endorsement, or urgency.
- - Pick the profile section for this format. Match the author's rhythm and casing; do not flatten all formats into one style.
- - For a rewrite, change phrasing and structure only as far as needed. For a new draft without source facts, ask for the missing substance.
- - Compare the result with 2 or 3 samples of the same format. Remove generic hooks, polished filler, and any banned punctuation. Return the draft without an essay about your choices.
+ - Use normal sentence capitalization and short paragraphs separated by blank lines. Keep each beat concrete and easy to scan.
+ - Open with a real reader problem, event, constraint, or result from supplied material. If a hook borrows a proven shape, identify its source outside the post. Never manufacture a first-person event or measured result.
+ - Put the payoff in the post. Show what happened, then stop after the last supported fact. No tidy summary or engagement question at the end.
+ - Prefer everyday language to technical feature lists. Preserve Federico's phrasing when editing his notes.
- ## Worked profile: Federico
+ ## Across formats
- This is an example of how a profile can differ by format. It is not the default for other people.
+ - No em dash (`—`) or en dash (`–`). Use a period, comma, or colon.
+ - Avoid “game changer,” “excited to announce,” “unlock,” “revolutionary,” artificial urgency, generic praise, and a writerly closing line.
+ - Keep sentences short and direct. One idea per sentence is a useful default, not a fixed word count.
+ - Check counts, names, dates, and claims against the actual source. Do not invent them for style.
- **Private messages:** lowercase fragments, usually 2 or 3 short lines; no greeting, signoff, or bullet list. Casual abbreviations and an occasional question fragment can fit. Suggest rather than command. Match the language of the thread.
+ ## Short pattern examples
- **Public posts:** normal sentence capitalization. Short paragraphs, a concrete opening tied to the reader's problem, then the actual event or result from supplied notes. Keep the payoff in the post. A hook must be grounded in a real sample or result, never manufactured. No inflated stakes or tidy writerly closer. Stop after the last supported fact; do not restate the payoff in a final line.
+ These are synthetic style illustrations, not Federico's messages or product claims. Replace their facts with the user's real context.
- **Across both:** no em dash or en dash. No invented first-person claim, artificial urgency, stock phrases such as “game changer” or “excited to announce,” or closing question added for engagement. Keep real numbers and names only when the supplied source supports them.
+ **DM, German**
+
+ > könnte gut passen
+ > sollen wir das morgen kurz testen?
+
+ **DM, English**
+
+ > looks useful actually
+ > want me to try it first?
+
+ **Post opening**
+
+ > We tested the draft with 12 people.
+ >
+ > Each revision took 30 minutes less in this synthetic example.
+
+ **Post ending**
+
+ > The team kept the shorter version.
+
+ Stop there when that is the last supported fact.