user-model-builder

# User Model Builder Build a complete userModel for **$ARGUMENTS** following the 7-phase workflow below. Each phase must complete before moving to the next. The output is a folder at `~/.claude/userModels/{name}/` containing a core persona, platform-specific dossiers, a content archive, and INDEX registration. ## Phase 1: Discovery Gather available information about the subject before creating any files. **Determine:** - Relationship to the user (collaborator, partner, observed figure) - Publishing platforms (Substack, Twitter, blog, LinkedIn) - Scrapeable content availability — dead domains require Wayback Machine CDX API - Basic identity: full name, email (`~` if unknown), phone (`~` if unknown), location, primary roles **Ask clarifying questions** if the relationship or scope is ambiguous. Proceed autonomously once direction is clear. ## Phase 2: Folder & Core Persona Create the userModel directory and core persona file. 1. Create `~/.claude/userModels/{name}/` 2. Write `{name}Model.md` following `references/persona-model-template.md` 3. Frontmatter per `references/frontmatter-schema.md` — `type: user-model`, `category: persona` 4. Required sections: Persona, Education (if known), Career Timeline, Worldview, Writing Voice (brief), Personal, Online Presence 5. Adapt to subject — omit sections that don't apply, add domain-specific sections as needed ## Phase 3: Content Scraping & Archive Scrape all discoverable published content into `archive/`. **Per platform:** | Platform | Tool | Archive path | |----------|------|-------------| | Substack | Firecrawl | `archive/substack/YYYY-MM-DD-slug.md` | | Twitter | Jina (Firecrawl blocks Twitter) | `archive/tweets/` | | Dead domains | Wayback CDX API → Firecrawl | `archive/{platform}/` | | LinkedIn | GDPR export via `linkedin-export` skill | `archive/linkedin/` | **Wayback discovery:** ``` https://web.archive.org/cdx/search/cdx?url=DOMAIN&output=json&fl=timestamp,original&collapse=urlkey ``` **Each archived file gets frontmatter:** ```yaml --- title: "Post Title" date: YYYY-MM-DD source_url: "https://..." archived_from: "{platform}" archive_date: YYYY-MM-DD --- ``` **Naming convention:** `YYYY-MM-DD-slug.md` (kebab-case slug from title). Write `archive/INDEX.md` — table of all files with dates, titles, approximate word counts. ## Phase 4: Voice Analysis & Dossiers Read the entire archived corpus before beginning analysis. Analyze each platform's content for voice, worldview, and biographical signals. For each platform, create a social dossier following `references/social-dossier-template.md`. **Frontmatter:** `type: social-dossier`, `category: {platform}-analysis`, include `corpus` field. **Required dossier sections:** - **Source** — URLs, scraping method, verification date - **Platform Profile** — Handle, legal entity, tagline, cadence, engagement metrics - **Post/Content Index** — Table: date, title, type, themes - **Voice Analysis** — Registers, tonal range, sentence mechanics, characteristic phrases. Follow `references/voice-analysis-guide.md` - **Worldview Markers** — Minimum 5 recurring themes with evidence - **Biographical Signals** — Per-post table of facts revealed. Separate distinct events. - **Calibration Quotes** — 10-15 verbatim quotes. Copy directly from archive files. Verify character-by-character: opening words, punctuation, emphasis, quote mark style. Never paraphrase. Include a Voice Evolution section when the corpus spans 3+ years or crosses a major life transition. Name dossiers descriptively: `{platform}-{publication-name}.md`. ## Phase 5: Cross-References & Registration Wire everything together. 1. Add a "Supplementary Dossiers" section to `{name}Model.md` listing all dossiers and archive 2. Update `~/.claude/userModels/INDEX.md`: - Add/update the subject's section with a Files table - Add Data Collections subsection for archive directories - Add Related links (Substack, LinkedIn, etc.) ## Phase 6: Review Launch a `code-reviewer` subagent to audit all files against `references/review-checklist.md`. **Priority checks:** 1. Calibration quotes are verbatim (opening words, punctuation, quote style) 2. Author bios are verbatim (not paraphrased) 3. Biographical facts match source (names, dates, locations — distinct events not merged) 4. Frontmatter schema compliance 5. INDEX.md consistency with files on disk 6. Cross-references resolve to real files Fix all critical and moderate findings before declaring complete. **Automated validation:** ```bash uv run ~/.claude/skills/user-model-builder/scripts/validate_usermodel.py ~/.claude/userModels/{name}/ ``` ## Phase 7: Summary Report what was built: - Files created (with paths) - Corpus stats (post count, word count, date range) - Key findings (voice registers, worldview themes, biographical discoveries) - Suggested next steps (RLAMA collection, voice model for composition, etc.) ## Reference Files | File | Purpose | |------|---------| | `references/persona-model-template.md` | Core persona section skeleton | | `references/social-dossier-template.md` | Social dossier section skeleton | | `references/frontmatter-schema.md` | YAML frontmatter spec for all file types | | `references/voice-analysis-guide.md` | Open Souls voice analysis methodology | | `references/review-checklist.md` | Post-build verification checklist | | `scripts/validate_usermodel.py` | Structural + quote accuracy validation |