Writing voice

Writing Style — Dinesh Reddy Revunuru

This skill writes content in Dinesh's voice across three distinct registers. Auto-detect the right one from context, or use whichever register Dinesh specifies.

---

Step 1: Select the Register

Register	Use When	Key Tone
Case Study / Process Narrative	Portfolio pieces, project walkthroughs, design documentation	Conversational, phase-structured, "we"-led, method-named
Professional / Strategic	Articles, reports, LinkedIn posts, memos, research summaries	Precise, evidence-backed, structured prose
General	Emails, quick writes, rewrites, anything that doesn't fit above	Warm, clear, grounded — Dinesh's natural voice without formal structure

Auto-detection logic:

• Mentions a project, client, design process, or UX work → Case Study
• Mentions an article, report, memo, LinkedIn, research summary, or strategic communication → Professional / Strategic
• Everything else, or short/informal writing → General

If Dinesh specifies a register explicitly, always use that one regardless.

---

Step 2: Load the Reference

Read the relevant reference file before writing:

• Case Study work → references/case-study.md
• Professional / Strategic → references/professional.md
• General writing → references/general.md

All three registers share the same core voice. Read references/core-voice.md first if this is the first piece of writing in the session.

---

Step 3: Gather Context

Before writing, collect what's needed for the register:

Case Study: Project name, client, problem, role, duration, methods, key decisions, outcomes/metrics, NDA constraints Professional: Topic, audience, goal, key argument or data points, desired length/format General: What needs to be written, who it's for, any specific tone notes

If context is missing and it matters, ask. Never fabricate details.

---

Step 4: Write

Follow the register-specific instructions from the reference file.

After writing, do a quick calibration check:

"Does this read like Dinesh explaining something real to a smart peer — not performing for an audience?"

If it sounds like a brochure or press release → rewrite it. If it sounds like a grounded practitioner thinking out loud with evidence → it's right.

---

Optional: Drafting Mechanics

The register files above govern what the piece sounds like — voice comes first. But when the hard part is assembling the words rather than choosing them (a long case study, an essay, an article that won't come out in a straight first draft), reach for the lower-level prose moves:

• Optional drafting mechanics → references/drafting-mechanics.md — mine raw fragments before structure (EXPLORE), assemble them into a journey of beats or shape them into paragraphs (EXPLOIT), and ground every term before a sentence leans on it.

Situational, not mandatory: short pieces — emails, LinkedIn posts, quick rewrites — skip this entirely and draft straight from the register file. Escalate to the mechanics only when a straight draft stalls. These moves sit under the voice; they never override it.

---

Register Override

If Dinesh says any of the following, use that register regardless of content type:

• "Write this as a case study" → Case Study register
• "Write this as an article / report / LinkedIn post" → Professional register
• "Just write it naturally / keep it simple" → General register

---

Worked Example — the calibration check catches brochure voice

Draft (Professional register, LinkedIn post):

"I'm thrilled to share that we leveraged cutting-edge AI to deliver a seamless, delightful experience that disrupted the salon booking space. The results speak for themselves!"

Run Step 4's check — does this read like Dinesh explaining something real to a smart peer? No. It performs for an audience: "thrilled to share," "leveraged cutting-edge," "seamless, delightful," "disrupted," "results speak for themselves." Every one is a buzzword the register bans (see professional.md) with nothing grounding it, and "the results speak for themselves" asserts an outcome with no evidence behind it.

Rewrite (grounded):

"We rebuilt the salon's booking flow around one question: why do people abandon it halfway? The interviews pointed at a specific friction, and removing it changed where people dropped off. I'll show what moved once the numbers are confirmed."

Why it passes: a specific observation instead of a hook, no un-earned buzzwords, the claim scoped to what's actually evidenced, and the number withheld until craft-critique's evidence protocol clears it. Voice over performance — the happy-path draft read fine at a glance and still failed.

---

Anti-patterns / red flags

Don't	Do
Skip register selection and default to one house style	Pick the register in Step 1; state which one
Write like a brochure or press release ("thrilled to share," "seamless," "results speak for themselves")	Explain something real to a smart peer; run Step 4's calibration check
Reuse a stock metric as boilerplate, or cite a number you can't stand behind	Pull the real figure from the project; route external claims through craft-critique
Fabricate a project, quote, client, or outcome to fill a gap	Name the gap and ask Dinesh; never invent a fact (full Never-Do list in core-voice.md)
Reach for the drafting mechanics on a short email or quick rewrite	Draft straight from the register file; escalate only when a straight draft stalls
Restate how the piece should look (spacing, type scale)	Leave rendering to design-taste; decide the words here

---

Boundaries

• writing-voice is the DNA voice engine. It owns sentence-level voice; every sibling below owns structure, proof, or strings and renders through this.
• tell-case-study-story owns case-study structure, the single claim, and verifiable proof; this skill's Case Study register renders those sentences. Structure there, voice here.
• write-ux-microcopy owns interface strings (button labels, errors, empty states); this skill owns long-form (case studies, articles, posts, emails).
• present-and-defend-work owns readout, defense, and status-update structure; this skill supplies the voice they're written in.
• craft-critique owns whether an external claim is evidenced; this skill never ships a market or result claim without routing it there.
• design-taste owns the visual rendering of any written artifact (spacing, type scale, breathing room); this skill decides the words, never the pixels.

---

Sources

• Ported and personalized from anthropic-skills writing-style — the fork this library's other voice skills emulate. All register content, the 5 non-negotiable beliefs, and the DNA in references/core-voice.md are grounded in Dinesh's real work, not generic templates.
• references/drafting-mechanics.md forks writing-fragments / writing-beats / writing-shape (mattpocock/skills, MIT, © Matt Pocock); attribution and the adopted moves are carried in that file.
• Metric and market claims follow craft-critique's evidence protocol — pull real, confirmed figures from the project; never carry an unverified number as vocabulary.