JoshuaShepherd Movement Leader Voice
Read a movement leader's own corpus and produce the canonical guide at docs/voice/{SLUG_UPPER}_VOICE.md. Downstream systems (Writing Studio, AI Lab, voice-fidelity evals, editors) load this file.
Discipline: describe the voice from the corpus, on the leader's own terms — never compare to another author. Every marker, phrase, and sample anchors in something the leader actually wrote or said.
Invocation
/movement-leader-voice {leader-slug}
/movement-leader-voice Brad Brisco
/movement-leader-voice --audit docs/drafts/article.md
/movement-leader-voice --refresh
Run from the leader repo root (e.g. movement-leader-websites/michael-cooper/).
| Flag | Effect |
|---|---|
{slug} or display name |
Defaults to cwd repo if obvious |
--audit {path} |
Score draft against Section 16 rubric; do not rewrite guide |
--refresh |
Regenerate from corpus; append version-history row |
| (empty) | Infer leader from repo; ask if ambiguous |
Inputs (priority order)
| Source | Path | Yields |
|---|---|---|
| Book corpus | docs/books/{book-slug}/chapters/*.md |
Highest — rhythm, lexicon, samples |
| Voice analysis | docs/movement_leader_research/{slug}/profile/voice-analysis.md |
Marker hypotheses |
| Collated substrate | **/{SLUG}_RESEARCH_COLLATED.md |
Fingerprint, quotes |
| Biography | **/profile/biography.md, **/_misc/committed-voice.md |
Approved first-person |
| Articles / talks | docs/**/articles/, transcripts |
Speaking vs writing |
If docs/books/ is empty, derive from research dossier and mark the guide research-grounded (lower confidence; tag unverified sample passages).
Execution protocol
Read this skill, then execute phases 1–3 to completion unless --audit (audit-only mode).
Phase 1 — Inventory
ls docs/books/*/chapters/ 2>/dev/null
ls docs/movement_leader_research/*/profile/voice-analysis.md 2>/dev/null
ls docs/voice/ 2>/dev/null
For large corpora, delegate per-book reading via Task (subagent_type: explore, one agent per book) → each returns markers, lexicon, structural habits, 3–5 exact sample passages with attribution. Synthesize centrally.
Phase 2 — Write
Write docs/voice/{SLUG_UPPER}_VOICE.md with all 18 sections per reference.md. Marker weights sum to 100%. Every sample passage is an exact, attributed quote.
Phase 3 — Verify
- All 18 sections present; titles match schema
- Weights sum to 100%; measurable targets per marker
- Self-contained — no cross-author comparisons
- Sample passages exact and sourced; "Do not invent" lists specific refusals
- Pre-output checklist and audit rubric reference the same markers
- Sources list real files read
Print a session report: slug, output path, corpus basis, marker count, sample-passage count.
Key rules
- Self-contained, never comparative.
- Corpus-grounded — read books before writing the guide.
- No fabricated quotes or biography.
- Markers are measurable — numeric targets, not mood words.
- One canonical file —
docs/voice/{SLUG_UPPER}_VOICE.md; leave research files untouched. - Reference implementation: brad-brisco repo,
docs/voice/BRAD_BRISCO_VOICE.md.
Additional resources
- Full 18-section schema: reference.md