Voice Brand Profile Builder

Sub-skill: Build / update the Voice & Brand Profile

by sergebulaev · MIT license · ★ 3,059 on GitHub · updated Sep 18

Builds or refreshes ../../../references/voice-profile.md so every writing skill in

# Sub-skill: Build / update the Voice & Brand Profile

Builds or refreshes `../../../references/voice-profile.md` so every writing skill in
this bundle drafts in the user's real voice instead of a generic "human" voice.
Runs on any agent (Claude Code, Codex, OpenClaw): the core path needs only the
user's own writing pasted in. Apify is an optional accelerator, never required.

## When this runs

- User says "build my voice profile", "learn my voice", "set up my profile", or
  invokes `linkedin-humanizer --mode profile`.
- Also offer it the first time a writing skill runs and finds `filled: no`.

## Inputs (any one is enough)

1. **Pasted samples (portable default).** Ask for 3-6 of the user's own real
   LinkedIn posts or comments. This alone is enough; no token, no history needed.
2. **Apify-assisted (optional).** If `APIFY_TOKEN` is set and the user gives their

Overview

Builds or refreshes ../../../references/voice-profile.md so every writing skill in this bundle drafts in the user's real voice instead of a generic "human" voice. Runs on any agent (Claude Code, Codex, OpenClaw): the core path needs only the user's own writing pasted in. Apify is an optional accelerator, never required.

Manual install (npx, ZIP)

Claude Codenpx skills add sergebulaev/linkedin-skills --skill sub-skill-build-update-the-voice-brand-profile -g -a claude-code

Claude.ai / desktopZip the folder and upload it under Settings → Capabilities → Skills.

Codex, Cursor, other agentsPoint them at SKILL.md. The other files are plain text.

Connects with Agent Alley MCP · agentalley.io/mcp

Files included

voice-profile.md3.0 KB

Read voice-profile.md here

sergebulaev/main1 file

voice-profile.md3.0 KB

voice-profile.md

Show the full text57 lines

sub-skill-build-update-the-voice-brand-profile/voice-profile.md57 lines · 3.0 KB

PreviewCode

Outline

RawView on GitHub

Sub-skill: Build / update the Voice & Brand Profile

Builds or refreshes ../../../references/voice-profile.md so every writing skill in this bundle drafts in the user's real voice instead of a generic "human" voice. Runs on any agent (Claude Code, Codex, OpenClaw): the core path needs only the user's own writing pasted in. Apify is an optional accelerator, never required.

When this runs

User says "build my voice profile", "learn my voice", "set up my profile", or invokes linkedin-humanizer --mode profile.
Also offer it the first time a writing skill runs and finds filled: no.

Inputs (any one is enough)

Pasted samples (portable default). Ask for 3-6 of the user's own real LinkedIn posts or comments. This alone is enough; no token, no history needed.
Apify-assisted (optional). If APIFY_TOKEN is set and the user gives their profile URL, pull recent activity with lib.fetch_user_recent_comments(username=...) (and any post URLs they share via lib.fetch_post) to gather more samples. Treat as an accelerator on top of, not a replacement for, pasted samples.
Manual. The user can also just tell you their niche, rules, and links.

Steps

Gather 3+ real samples of the user's writing (pasted or pulled).
Extract the voice fingerprint from the samples, not from assumptions:
- sentence-length rhythm (short/medium/mixed, and how often a long line appears)
- recurring openers and transitions they actually use
- punctuation habits (soft .. pause? never em dashes? line breaks per idea?)
- vocabulary they lean on, and any words/cliches they clearly avoid
- emoji and hashtag behavior
Infer niche, ICP, and pillars from the sample topics; confirm with the user rather than guessing.
Capture hard rules and CTA/link style the samples reveal or the user states.
Write ../../../references/voice-profile.md: fill sections 1-5, copy the 2-4 strongest lines verbatim into "Signature examples", and set the Status block to filled: yes, source: <pasted|apify|manual>, updated: <today's date>.
Show the user the filled profile for approval before saving, and tell them any writing skill will now match it automatically. They can edit the file anytime.

Hard rules

Build the fingerprint from the user's ACTUAL samples. Never invent a voice.
Preserve their quirks (a favorite phrase, an unusual rhythm). Those are the point. Only the generic AI-tell scrub still applies to drafts later, not to the profile itself.
Keep it honest about coverage: with 3 samples say the profile is a first pass and will sharpen as they add more; suggest re-running after 10+ posts.
Never put secrets, private data, or anything the user did not provide into the file.

The filled profile is read by linkedin-post-writer, linkedin-comment-drafter, linkedin-reply-handler, and linkedin-repurposer before they draft.
Re-run this any time the user's voice or focus shifts to refresh the profile.

1	# Sub-skill: Build / update the Voice & Brand Profile
2
3	Builds or refreshes `../../../references/voice-profile.md` so every writing skill in
4	this bundle drafts in the user's real voice instead of a generic "human" voice.
5	Runs on any agent (Claude Code, Codex, OpenClaw): the core path needs only the
6	user's own writing pasted in. Apify is an optional accelerator, never required.
7
8	## When this runs
9
10	- User says "build my voice profile", "learn my voice", "set up my profile", or
11	invokes `linkedin-humanizer --mode profile`.
12	- Also offer it the first time a writing skill runs and finds `filled: no`.
13
14	## Inputs (any one is enough)
15
16	1. Pasted samples (portable default). Ask for 3-6 of the user's own real
17	LinkedIn posts or comments. This alone is enough; no token, no history needed.
18	2. Apify-assisted (optional). If `APIFY_TOKEN` is set and the user gives their
19	profile URL, pull recent activity with `lib.fetch_user_recent_comments(username=...)`
20	(and any post URLs they share via `lib.fetch_post`) to gather more samples.
21	Treat as an accelerator on top of, not a replacement for, pasted samples.
22	3. Manual. The user can also just tell you their niche, rules, and links.
23
24	## Steps
25
26	1. Gather 3+ real samples of the user's writing (pasted or pulled).
27	2. Extract the voice fingerprint from the samples, not from assumptions:
28	- sentence-length rhythm (short/medium/mixed, and how often a long line appears)
29	- recurring openers and transitions they actually use
30	- punctuation habits (soft `..` pause? never em dashes? line breaks per idea?)
31	- vocabulary they lean on, and any words/cliches they clearly avoid
32	- emoji and hashtag behavior
33	3. Infer niche, ICP, and pillars from the sample topics; confirm with the user
34	rather than guessing.
35	4. Capture hard rules and CTA/link style the samples reveal or the user states.
36	5. Write `../../../references/voice-profile.md`: fill sections 1-5, copy the 2-4
37	strongest lines verbatim into "Signature examples", and set the Status block to
38	`filled: yes`, `source: <pasted\|apify\|manual>`, `updated: <today's date>`.
39	6. Show the user the filled profile for approval before saving, and tell them
40	any writing skill will now match it automatically. They can edit the file anytime.
41
42	## Hard rules
43
44	- Build the fingerprint from the user's ACTUAL samples. Never invent a voice.
45	- Preserve their quirks (a favorite phrase, an unusual rhythm). Those are the
46	point. Only the generic AI-tell scrub still applies to drafts later, not to the
47	profile itself.
48	- Keep it honest about coverage: with 3 samples say the profile is a first pass and
49	will sharpen as they add more; suggest re-running after 10+ posts.
50	- Never put secrets, private data, or anything the user did not provide into the file.
51
52	## Related
53
54	- The filled profile is read by `linkedin-post-writer`, `linkedin-comment-drafter`,
55	`linkedin-reply-handler`, and `linkedin-repurposer` before they draft.
56	- Re-run this any time the user's voice or focus shifts to refresh the profile.
57