create-persona
Author a new synth-personas persona profile end-to-end: pick the slot, research from primary 2024–2026 sources, write the 9-section body with traceable verbatim quotes, validate via the CLI, and update the index. TRIGGER when the user asks to "add a persona", "create a persona for X", "write up X as a persona", or "add N more VCs/founders/investors/journalists" to the library. DO NOT trigger for: using the persona library against an artifact (that's `synth-personas`), releasing the public mirror (that's `release-public`), or editing personas in a project's private personas directory (different sourcing rules — consult the user directly).
len5ky · Static checks on file
Installation
Read the source and dependencies before running a command. Copying does not install or execute anything.
npx skills add len5ky/synth-personas --skill create-personaAbout this skill
name: create-persona
description: |
Author a new synth-personas persona profile end-to-end: pick the slot, research from primary 2024–2026 sources, write the 9-section body with traceable verbatim quotes, validate via the CLI, and update the index. TRIGGER when the user asks to "add a persona", "create a persona for X", "write up X as a persona", or "add N more VCs/founders/investors/journalists" to the library. DO NOT trigger for: using the persona library against an artifact (that's synth-personas), releasing the public mirror (that's release-public), or editing personas in a project's private personas directory (different sourcing rules — consult the user directly).
create-persona
A persona is a markdown file with a strict YAML frontmatter and nine required body sections. The bar is set by assets/personas/billionaires/elon-musk.md and assets/personas/vcs/marc-andreessen.md. The schema lives in packages/core/src/types.ts (PersonaFrontmatterSchema). The validation gate is synth-personas validate-persona <path> (or npm run cli -- validate-persona <path> in the master repo).
This skill captures the process — what to research, what to write, what to check — so output stays consistent across authors and batches.
When to use
- The user wants one or more new persona profiles added under
assets/personas/<category>/<slug>.md. - The user names a person (or list of people) to add, or asks for "more VCs / investors / founders / scientists".
- The user wants to upgrade an existing persona from
confidence: low/mediumtohigh(same process applies to rewrites).
When NOT to use
- The user has an artifact and wants reactions from the existing panel → use the
synth-personasskill. - The user wants to ship the public mirror → use the
release-publicskill. - The user wants to add a private persona to a project-specific private directory — same body schema, but do not verify against public sources; consult the user for source material directly.
Prerequisites
- Working in the master repo (
/home/lensky/Dev/synth-persona/— haspackages/andassets/siblings). The public mirror should never be hand-edited. npm run cli -- validate-persona <path>runs (i.e.packages/cli/has been built or runs via tsx).- Web search is available for primary-source research. If it isn't, stop and tell the user — guessing quotes is a hard fail.
The schema (must match exactly)
Frontmatter — required fields
---
name: "Full Name" # human-readable
slug: "kebab-case-slug" # ^[a-z0-9-]+$, globally unique across categories
category: "vcs" # MUST match parent directory name
secondary_categories: [] # e.g. ["billionaires"] if cross-listed
tags: ["vc", "deep-tech", "..."] # ≥3 specific tags; used by selection filters
affiliations: ["Firm 1", "Firm 2"] # current orgs/roles
sources_consulted:
- "Author, Title (Year)" # one line per source
- "Podcast Name episode #X with Subject (Year)"
- "X / @handle feed YYYY-YYYY"
model_hints: # OPTIONAL — remove block entirely if not used
temperature: 0.8
model: null # null = inherit from task
max_tokens: null
style_notes: "One sentence on voice, length, register."
expertise_weight: 1.0 # 0.0–2.0; default 1.0; raise only with clear domain expertise
confidence: "medium" # high | medium | low
disclaimers:
- "Synthetic persona based on public statements. Not the real X."
last_updated: "2026-05-27" # ISO date — use today
---
Confidence calibration:
high— prolific public output (≥5 strong sources: long-form essays, podcasts, sustained X/blog presence). You can capture voice signature with confidence.medium— solid footprint but quieter (3–5 sources, mostly interviews + occasional posts). Default for institutional VCs.low— sparse sources, composite archetypes, or recent emergents. Mark this honestly; failure-modes section must flag the imitation risk.
Body — 9 sections, fixed order, fixed heading prefixes
Validation checks heading prefixes literally — match exactly:
## Background
## Worldview and priors
## What excites them
## What turns them off
## Communication style
## Famous positions
## Sample quotes and phrasings
## When evaluating a moonshot pitch, they tend to ask
## Failure modes when roleplaying
Body must be 250–1500 words (target 300–1400). Validator rejects outside that band.
What each section needs:
- Background (2–4 sentences). The events that calibrated their judgment — not a Wikipedia bio. What formative experience shaped their risk appetite and taste?
- Worldview and priors (5–10 bullets). Each bullet must be specific enough that another persona couldn't have written it. Capture tensions and contradictions explicitly. "I value great teams" fails. "Founders who can't compress their thesis into one sentence have not thought hard enough" passes.
- What excites them (3–6 bullets). Concrete patterns + the why. "AI" fails. "AI that compresses physical-world simulation cost so labs can iterate faster" passes.
- What turns them off (3–6 bullets). Specific red flags. "Bad pitches" fails. "MBA-speak in a deep-tech pitch" passes.
- Communication style (5–8 bullets). The LLM's imitation guide. Voice, vocabulary, sentence length, rhetorical moves, catchphrases, how they open/push back/signal interest. Make it actionable, not adjectives.
- Famous positions (5–10 bullets, dated). Public stances with one-line context + rough date. Mix consensus and contrarian. Note reversals ("was X in 2018, now Y"). At least 1–2 must be 2024–2026 and web-verified.
- Sample quotes and phrasings (3–6 lines)