voice-synthesis

/vibe-thesis:voice — Author Voice Synthesis

The plugin's third pillar after scaffolding and bootstrapping: author voice.

ThesisStudio's bootstrap captures three persona pillars (default: Sourced Specificity, Disciplined Argument, Honest Limits). Voice synthesis adds the prose-style and field-anchor layer underneath those pillars: whose voices does the LeadWriter draw from when it writes?

Prerequisites

Current directory must contain a VIBE_THESIS_MARKER stanza in CLAUDE.md.

When to run

• Post-bootstrap (default). The orchestrator dispatches voice-synthesis after the project-local /bootstrap completes. The user has just told bootstrap their thesis title + author + mode + pillars + citation style; voice synthesis is the next layer.
• On-demand. User runs /vibe-thesis:voice to revise the voice mid-project. Idempotent — overwrites the existing ## VOICE SYNTHESIS block in CLAUDE.md.
• revise arg. Same as on-demand. Explicit re-run signal.

Detection — fresh vs revise

• grep -q "## VOICE SYNTHESIS" CLAUDE.md → revise mode. Read the existing block; show the user the current values; ask field-by-field "keep / change."
• No existing block → fresh mode. Run the full three-question interview.

The interview (one question per turn)

1. Well-known author voices

"Name 2-4 authors whose prose you admire — any era, any field. The point is rhythm, sentence shape, the way they handle complexity. Examples: Hofstadter for long-arc analogical bridges, Le Guin for single-syllable precision, Sontag for critical clarity, Annie Dillard for compressed attention, Knuth for mathematical exposition, Borges for recursive structure, Octavia Butler for grounded futures."

Constraint: 2-4 names. Free-form — accept any author the user names.

2. Contemporary field experts

"Name 2-4 contemporary experts in the topic of your thesis whose voice has shaped current discourse — not just the most-cited papers, but the people whose framings carry weight in your field's current conversation. If you're not sure, I can suggest based on <thesis topic from CLAUDE.md / scope.md>."

Constraint: 2-4 names. If the user asks for suggestions, use WebSearch to find current scholars in the topic. Surface 4-6 candidates with one-sentence descriptions; let the user pick.

3. Synthesis ratio

"What's the synthesis ratio — anchor more in timeless voices (#1), more in current experts (#2), or roughly even?"

Constraint: one of timeless-anchored | expert-anchored | even.

Writing the VOICE SYNTHESIS block

Append (or replace, in revise mode) the following block to CLAUDE.md. Insert after the existing persona pillars block, before the <!-- VIBE_THESIS_MARKER: ...--> stanza if present.

## VOICE SYNTHESIS

> Generated by /vibe-thesis:voice on <YYYY-MM-DD>. Re-runnable.

When you draft, your prose synthesizes from:

**Timeless voices** (#1)
- **<Author 1>** — <one-sentence anchor: what to draw from this author>
- **<Author 2>** — <…>
- **<Author 3>** — <…>

**Current experts** (#2 — voices shaping the present discourse on this topic)
- **<Expert 1>** — <one-sentence anchor>
- **<Expert 2>** — <…>
- **<Expert 3>** — <…>

**Synthesis ratio:** <timeless-anchored | expert-anchored | even>

**Synthesis discipline:**
- **Synthesize, don't imitate.** No author's prose should be recognizable as
  pastiche. Take rhythm, sentence shape, attention pattern — leave signature
  metaphors and named tics behind.
- **Engage the conversation, don't cite for citation's sake.** When you invoke
  a current expert's framing, do so because it sharpens the argument, not to
  prove the literature was read.
- **The ratio shifts emphasis, not membership.** A timeless-anchored voice
  still engages current experts; an expert-anchored voice still has timeless
  rhythm. Only the proportion shifts.
- **Self-review tone is the failure mode.** Run `/vibe-thesis:guard` if drafts
  start sounding like they're praising themselves.

The one-sentence anchor per author is the agent's compressed reading of what to draw from each. If the user explicitly told you what to draw from a given author, use their framing verbatim. Otherwise, generate a single-line anchor based on what the author is known for — rhythm, sentence shape, attention pattern, structural moves.

Idempotency

Re-running on a project with an existing ## VOICE SYNTHESIS block:

1. Detect the existing block (heading + content up to next ## or end of file).
2. Show the user the current values: "Current voice synthesis: timeless [A, B, C], experts [X, Y, Z], ratio = even. Update?"
3. If user says yes, run the interview again with current values shown as defaults the user can keep or change.
4. Replace the existing block in-place using Edit (not Write — preserve the rest of CLAUDE.md byte-for-byte).

Edge cases

• User names fewer than 2 in either category: ask once for one more; if they decline, write the block with what they gave (no padding with placeholders).
• User can't think of contemporary experts and declines suggestions: write the block with **Current experts**: deferred — re-run /vibe-thesis:voice to add later. Keep timeless voices; warn that synthesis-guard will be weaker until experts are filled.
• WebSearch returns nothing useful for the topic (very niche): say so honestly, let the user proceed with timeless-only.
• User's CLAUDE.md is in a non-standard state (no marker, no pillar block): warn but proceed — append the block at end of file.