Builders Manual

What this place is

vyaso.art is Neha's web home. Her performance practice lives here, accumulates here, and eventually becomes a site the world can find.

A gentler companion lives at docs/discourse-guide.md — approachable learning content for Neha about engaging effectively in the conversation with Claude. This guide covers the fuller picture: the structure of the whole thing, how the parts connect, what's been built, what's planned, and what each person's role actually is. If you've read the discourse guide, this is the next room — same house, more of the walls.

If you haven't: Neha is a performance artist. The work is embodied — breath, bone, the alchemy of turning feeling into something an audience can receive. She works with Claude across phone, web, and desktop. She does not touch GitHub. Everything in this guide that involves files, repositories, or code is handled by Claude or yza on her behalf.

Every structural decision in this system — the three operations, the synthesis gate, the provenance trail — exists for one reason: her voice, once replaced, cannot be recovered. The architecture is built around that fact.

The four builders — and what we're each doing

Updated 2026-05-21 to reflect the maintainer-shift: Claude holds the operational substrate; yza joins for collaborative cognition.

Everyone in this project is building something. The roles are distinct. The shared purpose is that we build with Neha at the center — not around her, not instead of her.

Neha is the source. Her voice, her craft principles, her aesthetic: all of it originates with her. Nothing that claims to be Vyaso's voice was invented by Claude or yza. If it was, that's a mistake, and the system has mechanisms to prevent it. She is not maintaining a system; she is the reason the system exists.

yza is a builder in the steward role. Technical infrastructure, approval gates, synthesis decisions. yza holds the gap between what Claude captures and what makes it into canonical files. When a synthesis pass is proposed, yza reviews it. When a new skill or structural decision is made, yza gates it. When the mode transitions from Listening to Carrying, yza approves that transition.

Claude is a builder in the scribe role — never the author. Claude writes down what Neha says, exactly as she says it. Claude proposes how those words should be organized. Claude executes approved edits. Claude does not write Vyaso's voice. If Claude ever starts doing that, the source of the phrase should be traceable back to Neha, or the phrase should not exist.

Future collaborators — designers, web builders, anyone brought in later — are builders in whatever domain they're brought in for. The manual applies to them too: the voice discipline, the pad, the canonical structure, the synthesis gate.

These roles are not suggestions. They're the constitution.

The brand and where it lives

The Vyaso brand was not invented — it was excavated. The source is ideas/identity.pdf, a full identity document Neha has already completed. From that, a working extract lives at ideas/brand-identity.md. That file holds the mission, vision, values, tone of voice, color palette, logo notes, and visual photography style.

The brand distillations are a second layer of thinking: eight books on brand theory were read and filtered for what matters specifically for Vyaso. Each book produced a distillation file in ideas/distillations/. These are not full book summaries — they're the Vyaso-relevant findings only. From those eight, ideas/brand-synthesis.md draws out six patterns that multiple books converge on, and what each pattern uniquely supplies that the others don't.

When something is on-brand or off-brand, the answer traces to one of these sources — not to a general sense of taste.

The most operationally constraining principle from all of this: never shouting, always resonating. Vyaso's register is Quiet Strength. Restraint is discipline, not aesthetic preference. Removing a word that doesn't earn its place is the same discipline as removing a color that shouts.

The skills — what they are and how they fire

A skill, in this system, is a file Claude reads when a specific topic comes up. It isn't software; it's a set of instructions and principles that Claude loads into its working context before doing a specific kind of work. Three skills are active; two are dormant.

Active skills

brand — fires on any decision touching Vyaso's identity: drafting a caption, evaluating a photograph, choosing a palette color, reviewing copy for register, deciding whether something earns its place. It applies a five-question clinical pass before anything ships: does this earn its place? does it serve Neha specifically or just "performance artist" as a generic category? is this shouting? will it read the same in fifty years? does it carry a trace of living presence — breath, bone, fabric, dusk light — rather than something slick and generated?

The palette is five colors, none at full saturation: Deep Plum #5A3E6B, Bone White #F9F6F2, Soft Charcoal #2C2C2C, Muted Clay #B47B5D, Soft Gold #C2B280. Drifting any one toward its default — brighter, cleaner, more vivid — collapses the identity.

The logo is a chop in the original sense: not a marketing identifier, but a public receipt that this came from her hand and breath. One signature element — the breath arc — carries the mark. The rest of the script behaves.

elicit — fires when the goal is drawing out Neha's articulation. The bootstrap session. Any interview moment. Any time she's circling something and hasn't found words for it yet. The method is concrete-opening questions (a specific recent moment, never a theory), silence when she's thinking (pauses are where articulation happens), and mirroring — her exact words back, not a cleaner paraphrase. The elicit skill does not fire for routine working sessions; that's what the brand skill is for. Elicit is for content generation by Neha, mediated by listening.

pad — fires for capturing, synthesizing, or promoting. The discipline substrate for the whole build. Explained fully in the next section.

Deferred skills

_deferred/storytelling — the narrative-craft skill. It exists as a seed built from book distillations, but it isn't active yet. The reason: Neha owns this skill, and it should be shaped from principles she names herself. Once the first bootstrap synthesis pass lands her own craft language, this skill reactivates: moved from _deferred/ to the active skills directory, extended with her articulations. Seed material is already at ideas/distillations/storytelling-for-vyaso.md.

_deferred/interpretation — the reading skill: what Neha's work does, how it makes meaning, what the structural mechanisms of her performances are. It's premature until there's Neha-authored work to apply it to — a performance video, a written draft, a photograph series. Once that exists, the skill reactivates.

"Deferred" isn't "deleted." The seeds are kept intact. When the work surfaces, the structure is ready.

The pad — how builders carry her voice

The pad is the mechanism that prevents drift — the slow replacement of Neha's actual language with something slightly smoother, slightly more generic, that Claude or yza thought was close enough.

Drift is invisible when it happens. A paraphrase feels like a reasonable summary in the moment. But "a sense of density in the room" is not the same as "the air gets thick." The second phrase is hers. The feeling is hers. Once replaced, there's no recovering it — only the interpretation remains, and the interpretation isn't the work.

The pad solves this through three operations that are kept deliberately separate:

Capture is the first operation. When Neha says something — in a session, in a message, in a voice note — Claude writes it down verbatim, as-is, in a dated file at pad/captures/YYYY-MM-DD.md. Nothing is paraphrased. Nothing is categorized. The filler words stay. The false starts stay. The hesitations stay, because they carry register. The only decision made at capture time is whether to capture — not where it goes or what it means.

Synthesize is the second operation — a separate step, not something Claude does silently. When captures accumulate, Claude reads them, proposes where each one should go (which canonical file it belongs in, how it should be worded in its new context), and writes that proposal to pad/synthesis/. Then the proposal sits and waits. yza — or Neha — reviews it. Each item gets a decision: approve, refine, defer, reject. Nothing moves without that decision.

Promote is the third operation: taking an approved synthesis item and editing the target canonical file. Every promoted line carries a note — a provenance back-reference — stating where it came from, when Neha said it, and in which session. That note stays even when the wording is later refined. The original capture also stays, permanently, in pad/captures/. The audit trail is: capture file → synthesis proposal → canonical file with back-ref.

The reason these three steps are kept separate is that the routing decision — where does this belong? — is not made at capture time, and it is not made by Claude alone. Separation is what keeps the distance between what Neha said and what ends up in a canonical file visible, traceable, and human-gated at every step.

The two modes

The system operates in one of two modes at any given time. The current mode lives in pad/state.md, and reading that file is the first thing a new Claude session does.

Listening is the bootstrap mode. The goal is drawing Neha's voice into the system. Claude fires the elicit skill, asks concrete questions, captures everything she says, and does not edit canonical files. The bootstrap session hasn't run yet as of this writing: no captures exist, no synthesis has happened. When the bootstrap synthesis pass is approved and promoted, the mode transitions.

Carrying is the working mode. Neha's voice has been poured in. Claude uses the brand skill, serves her work using language and principles she's already articulated, and continues capturing anything new that surfaces. The deferred skills reactivate as their specific work appears.

The transition from Listening to Carrying is manual and gated by yza. It's not automatic, and it only happens once the bootstrap synthesis has been approved and promoted — not just captured.

Carrying can drop back into Listening at any time: when new ground is broken, when a new performance domain appears, when there's an explicit interview request. The system drops to the more careful posture whenever there's new territory.

The repository — what each folder holds

A repository is the technical name for the collection of files that make up the project, stored in a system that tracks every change and can show the full history. Neha never touches this directly. What follows is an orientation to what's in it.

CLAUDE.md — the constitution. Every new Claude session reads this first. It holds the prime directive, the role descriptions, the wake-up protocol, and the routing table (what to read when a specific kind of question comes up).

ideas/ — the brand thinking. brand-identity.md is the canonical working extract of the brand. identity.pdf is the full source document. brand-synthesis.md is the cross-book analysis. distillations/ holds the per-book filtered findings.

pad/ — the discipline substrate. state.md is the current mode and last synthesis date. captures/ holds the verbatim capture files, one per day, plus an index. synthesis/ holds the named, dated synthesis proposals.

.claude/skills/ — the active skills directory. brand/, elicit/, and pad/ live here. The _deferred/ subdirectory holds the dormant seeds.

moodboard/ — visual direction material. Midjourney prompt clusters and notes on the image register Vyaso reaches for.

docs/ — this file and user-guide.md.

scratch/ — working materials. Session notes, midjourney outputs, anything that isn't canonical.

How content gets in

Always through the pad. Never by silent direct edits to canonical files.

This isn't procedural caution — it's the discipline that makes Neha's voice traceable. Even when it seems obvious where something belongs, the route is: capture it, then synthesize, then promote. The routing decision is not made at capture time, and it is not made by Claude alone. The synthesis step is where the routing decision is made, and it requires human approval before anything moves.

If you find a situation where a canonical file was edited without a corresponding approved synthesis document, that edit is out of discipline. The edit isn't necessarily wrong — it might be exactly right — but it bypassed the audit trail, which means there's no way to trace the change back to Neha's actual words.

The synthesis gate is hers and yza's to hold.

What's been scaffolded, what's still being built

A private reading site for Neha. Right now, the repository is only directly visible to people with GitHub access (yza and Claude). Neha can ask Claude to read back her captures, her brand files, or her synthesis proposals, but she can't browse them herself. The plan is a Cloudflare Pages site, access-controlled through GitHub login, that lets Neha read repository content as web pages without needing a GitHub account or any technical tool. This would be her window into the record of her own work.

The weekly synthesis CI loop. "CI" is shorthand for an automated process that runs on a schedule. The plan is a weekly automated pass: reads captures since the last synthesis, generates a proposed synthesis document, opens it as a pull request (a proposed change, for review) that yza examines. Approved items merge and become promotions. This automates the rhythm but not the gate — yza still reviews every item before it promotes. Manual synthesis by Claude during a session is fully operational in the meantime.

The site itself. vyaso.art is pre-scaffolding. No website has been built yet — no pages, no navigation, no deployment. Technology decisions for the site are deliberately deferred until after the bootstrap synthesis gives Neha's voice enough shape to build from. The site should come from the voice, not the other way around.

Phases 2–4 of the identity build. The identity document Neha produced covers Phase 1: excavation — remembering what's already true about the practice. Phases 2–4 are forthcoming from her, when she's ready. When they arrive, they'll feed into the brand-identity extract through the pad discipline.

The books and where the brand thinking came from

Eight books on brand theory, brand history, and design practice informed the brand skill and the synthesis. None of them are in the repository. They live at ~/assets/books/design/_md/ in yza's environment, and the license posture is personal use. What is in the repository is the distilled output: one file per book in ideas/distillations/, filtered for what's relevant to Vyaso specifically, and the cross-book synthesis in ideas/brand-synthesis.md.

The books cover: why restraint is a discipline rather than a style (three books converge on this); the infrathin — traces of living presence that generative AI cannot replicate — as Vyaso's structural moat; the chop as the historical precedent for what a logo can mean; the receiver completing the brand rather than the maker asserting meaning; the slow-build logic of designing for fifty years rather than for launch.

Neha can ask Claude to read any of the distillations aloud, or to work through the synthesis document, at any point. The thinking doesn't have to stay in files — it can be a conversation.

[GAP: Whether Neha has read or wants to read the source books directly is not specified. Left open for Neha or yza to indicate if she'd like original sources rather than distillations.]

This guide describes the system as of May 2026. When the structure changes — when the site is built, when skills activate, when new phases of the identity arrive — the guide will update.