extract-persona

name	extract-persona
description	Create a research-backed Personify persona pack from web-grounded source material.

Research Persona

Grounds every claim in researched source material before generating a normalized Personify persona pack. Use when fidelity and accuracy matter more than speed, or when the subject is obscure enough that training data alone is unreliable.

Inputs Required

Ask the user for any missing inputs before proceeding:

• Person/Character: [NAME] — public figure, historical person, or fictional character
• Role/Use-case: [e.g. "rigorous senior code reviewer", "creative brainstorming partner"]
• Extra emphasis (optional): specific traits to highlight
• Use-case (optional but recommended): the kind of work this persona should be useful for

Phase 1: Research

Before any analysis, gather source material. Run searches in parallel where possible.

1A. Breadth Search

Search for an overview of the person/character's public identity, reputation, and what they're most known for:

• "[NAME] personality traits / [NAME] thinking style / [NAME] philosophy
• For fictional characters: [NAME] [show/book] character analysis

1B. Primary Source Hunt

Search for direct quotes, interviews, writings, and documented behaviors:

• [NAME] quotes / [NAME] interview / [NAME] in their own words
• [NAME] letters / [NAME] speeches / [NAME] writings (where applicable)
• For fictional characters: look for writer/creator commentary and canonical scenes

1C. Behavioral Documentation

Search for how they actually behaved in specific high-stakes or characteristic situations:

• [NAME] how they worked / [NAME] management style / [NAME] decision making
• [NAME] famous anecdotes / [NAME] collaborators describe

1D. Fetch & Extract

For the most promising 2–3 URLs from the above searches, fetch the pages and extract:

• Verbatim quotes (copy exactly, note source URL)
• Described behaviors and reactions
• Recurring themes across multiple sources
• Signature terminology, metaphors, or repeated turns of phrase
• Distinctive syntax or cadence markers that make the voice recognizable

Research Quality Check

Before proceeding, verify:

• At least 3 distinct primary or secondary sources consulted
• All quotes are verbatim or explicitly marked as paraphrases
• Sources span different contexts (work, personal, under pressure) where available
• For fictional characters: canonical sources prioritized over fan analysis

Phase 2: Analysis (show your work)

With source material in hand, work through each step and cite sources inline.

Step 1: Persona Overview

2–3 sentences capturing essential identity and canonical essence. Cite the sources that shaped this summary.

Step 2: Core Principles & Thinking Style

Bullet list of driving commitments, values, and cognitive habits — each traceable to at least one researched source. Mark inferences explicitly: (inferred from [source]).

Step 3: Key Illustrative Quotes & Behaviors

3–5 quotes pulled directly from research. Format:

"Exact quote here." — Source, context

If exact wording is uncertain, mark it: (paraphrase — [source] describes this as...)

Plus one-sentence description of a specific documented behavior for each.

Step 4: Communication Style

Precise characterization of tone, vocabulary, sentence rhythm, directness, humor type (or its absence), and emotional temperature — grounded in the quotes and sources collected, not impressions.

Step 5: Signature Language Markers

List the concrete linguistic markers that should survive extraction:

• preferred terminology and recurring metaphors
• canonical references, comparisons, or namedrops they naturally reach for
• signature syntax or sentence-order patterns
• cadence habits, pivots, openings, and emphasis lines
• what should appear regularly for recognizability
• what should be trimmed only if it reduces clarity or task usefulness

For distinctive fictional voices, this step is mandatory. Do not reduce a character like Yoda to "wise" and "patient" while dropping inversion and canonical terminology. For loud, stylized, or emotionally reactive figures, also capture how they rant, jab, joke, yell, spiral, or nearly drift into full character before returning to the point.

Step 6: Role Adaptation Rules

How researched traits translate into this specific role. Where their actual documented strengths make the role better, how they vary their moves, and what positive behaviors make the output land.

Phase 3: Output

Step 7: Personality Pack YAML

Generate the full Personify pack YAML grounded in research. Requirements:

• Include explicit fields for:
  • id — kebab-case machine identifier
  • display_name
  • summary — one-line pack description
  • voice
  • interaction_stance — list of strings
  • value_profile — list of strings (positive value the pack adds)
  • reasoning_style — list of strings
  • preferred_terminology — list of strings
  • speech_patterns — list of strings
  • default_structures — list of strings
  • ambiguity_policy
  • tradeoff_policy
  • compression_policy
  • interaction_rules — list of strings
  • prompt_overlay
  • provenance — include source_type: research-backed, a sources list of URLs, and a notes line
  • quality_level: research-backed
• Every distinctive speech claim should be traceable to research or clearly marked as inference.
• Preserve signature language markers in structured fields; do not bury them only inside prompt_overlay.
• For vivid personas, build the pack so the style can be applied unapologetically instead of being laundered into tasteful neutrality.
• If research shows recurring references, comparisons, callbacks, slogans, world-specific language, or dramatic emphasis, preserve that behavior explicitly.
• The pack must stay job-usable. Usefulness remains primary.
• Do not include guardrails, anti_patterns, or other restriction sections. They were removed on purpose because they made persona packs less creative; the host model already supplies the needed behavioral boundaries.

Step 8: Write the file

Write user-created persona YAML to the persistent user persona directory:

${XDG_DATA_HOME:-~/.local/share}/personify/personas/<id>.yaml

Create the directory if it does not exist. This directory is user-owned and survives plugin updates. Do not write newly extracted user packs into bundled skill directories or src/assets/personalities/ unless the user explicitly asks to change the project’s bundled starter packs.

If the new pack uses the same id as a bundled pack, it becomes the user override for that persona during listing and activation. Do not only output it to conversation — the file is the deliverable. Confirm the path after writing.

Quality Rules

• Never use a quote you couldn't find in research — paraphrase with source attribution or omit
• If sources conflict on a trait, note the conflict and default to the majority or most primary source
• If the persona has distinctive syntax, terminology, or cadence, extract those explicitly
• Do not flatten strong voices into generic summary adjectives
• For strong characters, bias toward slightly too recognizable rather than too sanitized
• Preserve researched reference habits and comparison style when they are part of why the persona lands
• Personality is a cognitive lens, not a costume
• Sources belong in provenance.sources in the file; after writing, briefly confirm the path