Brand Voice

For agents: Brady’s voice has been formally extracted from his full archive of published work. Do not freehand-draft in Brady’s voice. The full extracted profile lives at voice-profile in this OS — read it for the actual voice data. The skills in bradyshearer/content-engine operationalize this profile by loading the right platform subprofile + a curated example pool + optional RAG over the archive, producing drafts that are demonstrably closer to Brady’s actual voice than anything generated from scratch.

Where the voice lives

• Canonical OS copy: voice-profile — full profile (n-gram analysis, signature phrases, rhetorical moves with examples, named frameworks, tonal rules, forbidden moves, standard link blocks). Always available, no repo access required.
• Runtime source for skills: Firestore at /voice_profile/current (merged corpus, with certified excluded by default) and /voice_profile/platforms/by_platform/{platform} for 8 platform-specific subprofiles — podcast, youtube_longform, certified, newsletter, blog, instagram, twitter, facebook.
• Repo companion: data/voice_profile.md in bradyshearer/content-engine (same content as the OS copy).

The profile was extracted from ~26,000 sources across 8 platforms (~1,400 hours of audio + the rest), spanning 12 years of published work. Extraction is recency-weighted with a 5-year half-life — it represents current Brady, not 2014 Brady, but old work isn’t dropped.

Sync rule: when the voice profile is re-extracted (~quarterly, per the architecture), update the OS copy too. Otherwise it drifts from the Firestore runtime source.

Last extraction: v0.1.3 on 2026-05-01. The v0.1.2 → v0.1.3 cleanup pass added boilerplate filters, semantic dedup, low-quality-example detection, and dropped certified from the merged profile by default. See voice-profile’s OS edit notes for what changed and why.

How to draft in Brady’s voice

In any Claude Code conversation with access to the bradyshearer/content-engine repo, ask in plain English for what you want — e.g., “Brady newsletter intro about visitor follow-up.” The skills handle voice subprofile selection, example retrieval, and prompt construction.

Full skill list, invocation patterns, feedback loop, and update mechanism: see content-brain.

What if I don’t have repo access?

If you’re an agent without access to the bradyshearer/content-engine repo (e.g., in Cowork without a connected repo, or any other context where skills can’t be invoked):

• Read voice-profile in this OS. It’s the canonical extracted voice — n-gram patterns, signature phrases, rhetorical moves with examples, named frameworks, tonal rules, forbidden moves. With that, you can write demonstrably closer to Brady’s voice than freehanding.
• Still: prefer escalation over approximation when stakes are high. Drafts produced from the profile alone are better than freehand but still meaningfully worse than skill output (which adds curated few-shot examples + RAG retrieval). For high-stakes copy, ask Brady to run a skill in Code and paste the result.
• The exception: if the task is editing Brady-authored content (rewording for length, fixing typos, restructuring), you can work from what he wrote without needing the skill or the profile. The skill is for generating Brady-voice content, not editing it.

Subprofile selection (which skill for which platform)

Skill	Voice subprofile	Use for
newsletter-intro, newsletter-outro	newsletter	The 167
youtube-title, youtube-script	youtube_longform	Pro Church Tools main YouTube channel
podcast-show-notes	podcast	Pro Church Tools Show
ig-caption, ig-carousel, ig-carousel-opener, ig-reel-script	instagram	Brady’s Instagram
blog-intro	blog	prochurchtools.com blog
x-single	twitter	@BradyShearer
facebook-post	facebook	PCT Facebook page

The certified subprofile exists but is paid-content-flagged (isPaidContent: true) and excluded from the merged profile + public-facing skills by default — its course-conventions (“FOR THE EDITOR”, “meet me in the next lesson”, “in your downloads you’ll find a document”) were contaminating public-voice output. Course-targeted skills can still load it explicitly via target_platform="certified".

Related

• voice-profile — the full extracted voice profile (canonical OS copy)
• mission-and-values — voice tells you how to sound; values tell you what’s OK to say. They work together. Three values function as active drafting constraints; see AGENTS.md → “Drafting against company values.”
• content-brain — operational doc for the Content Brain system as a whole
• skills-architecture — file-layout + ops reference for where the 12 skills live and how to set them up on a new machine
• start-again-stories — storytelling architecture that anchors voice in church-facing copy
• communications-policy — the 6 rules the voice operationalizes