Vibe Portrait

Vibe Portrait

Developer personality portrait generator with subcommands.

Command routing

Determine which subcommand the user wants based on their message:

If ambiguous, ask the user which action they want.

Generate (full analysis)

Step 0: Ask analysis mode

Before reading any data, ask the user which analysis mode they prefer:

How thorough should the analysis be?

1. ⚡ Quick (cost-efficient) — Sample ~200 messages. Fast, low token cost, good enough for most portraits.
2. 🔍 Full (comprehensive) — Read ALL messages. Higher token cost, but captures every nuance and evolution of your personality. Best for users with long history who want maximum accuracy.

If the user doesn't respond or says "just do it" / "默认" / "whatever", default to Quick mode.

In Quick mode, follow the sampling strategy described in Step 1c. In Full mode, read all lines from history.jsonl (and any imported files). Skip the sampling strategy entirely. If the file exceeds 3000 lines, still read in batches (e.g., 500 lines at a time) to avoid tool errors, but process every line.

Step 1: Locate and read conversation data

1a: Local history

Always read from ALL available sources regardless of which terminal the user is running. A developer's personality is shaped by all their AI interactions, not just one tool.

Check and read from every source that exists:

Merge all messages from all sources into one pool before sampling. Use ts timestamps (Codex history) or timestamp (Codex sessions) for chronological sorting when available.

1b: Multi-machine sync via Git repo

Vibe Portrait uses a private GitHub repo to sync persona data across machines. This is the recommended approach for users with multiple development machines.

First-time setup (if no portrait repo exists yet):

After analysis is complete (Step 7), the skill will offer to create a private repo. See Step 7 for details.

If the user already has a portrait repo:

Ask the user for their portrait repo URL, or check if ~/.vibe-portrait-repo exists (a file containing the repo URL, written during first-time setup).

# Check for existing repo config
cat ~/.vibe-portrait-repo 2>/dev/null

If found, clone or pull the repo into a temp directory:

# Clone if not already local
REPO_URL=$(cat ~/.vibe-portrait-repo)
REPO_DIR=~/.cache/vibe-portrait-sync
git clone "$REPO_URL" "$REPO_DIR" 2>/dev/null || (cd "$REPO_DIR" && git pull)

Then load existing analysis files from $REPO_DIR/analysis/*.json. When merging with the current machine's analysis:

• Each machine's analysis is stored as a separate JSON file: analysis/{hostname}-{date}.json
• To merge: average the numeric scores across all machine analyses, weighted by message count
• Union all quotes (deduplicate by text)
• Union all detected domains and tools
• The persona skill (me/SKILL.md) is always regenerated from the merged result

If the user provides a .jsonl file directly (fallback):

Read it as a supplementary source. Merge messages from all sources before sampling:

• Deduplicate messages with identical display content
• If timestamps are available, sort merged messages chronologically
• Treat all sources equally

1c: Read strategy

Use a bash command to count total lines first:

wc -l ~/.claude/history.jsonl

Quick mode (default): Sample strategically:

• Read the first 50 lines (early personality)
• Read the last 100 lines (current personality)
• Read 50 random lines from the middle (evolution)
• Total: ≤200 messages per source file

Full mode: Read all lines. For large files (>500 lines), read in batches of 500 lines using offset/limit to avoid tool errors. Process every message.

Parse each line as JSON and extract the display field. Skip lines where display is empty, null, or a slash command (starts with /).

Minimum threshold: If fewer than 20 non-empty messages are found across all sources, inform the user that there isn't enough data for a meaningful portrait. Suggest they try again after more conversations.

Step 2: Analyze across six dimensions

Read the analysis framework for detailed signal definitions: → references/analysis-framework.md

Analyze across six dimensions: Communication Style, Technical Breadth, Technical Depth, Decision Patterns, Collaboration Style, Work Rhythm. See the reference file for detailed signal definitions.

Language detection (for i18n): if >50% of sampled messages are in Chinese, set meta.lang = "zh". Otherwise "en". This determines the portrait page language.

For each dimension, produce: a score (0-100), a one-line summary, and 2-3 evidence quotes (keep quotes short, anonymize sensitive content).

Step 3: Compute personality type, rating, and famous match

3.1 MBTI Mapping

Read: → references/mbti-mapping.md

Map observed signals to four axes:

• E/I: delegation breadth vs self-reliance
• S/N: implementation focus vs architecture focus
• T/F: logic-driven vs people-aware
• J/P: structured planning vs exploratory iteration

Each axis: 0-100 score. Letter determined by which side of 50.

3.2 Developer Rating

Read: → references/rating-rubric.md

Rate on a six-tier scale (provide BOTH label for Chinese and labelEn for English):

Always include strength highlights AND growth areas, even for top-tier ratings.

3.3 Famous Person Match

Read: → references/famous-matching.md

Match the user across 3 independent dimensions (Technical Spirit / Strategic Mind / Communication Soul), each picking a different historical or contemporary figure. No predefined list — the AI uses its own knowledge. See reference file for rules and anti-bias constraints.

Step 4: Generate the portrait HTML

Read the HTML template: → templates/portrait.html

The template contains a PORTRAIT_DATA JavaScript object near the top of the <body>. Replace the ENTIRE PORTRAIT_DATA object with the computed data. The structure is:

const PORTRAIT_DATA = {
  meta: {
    username: "...",         // infer from system username or ask
    generatedAt: "...",      // today's date
    messageCount: 0,         // actual count of sampled messages
    dateRange: "...",        // earliest to latest message date
    confidence: "...",       // "Early Sketch" / "Clear Portrait" / "Deep Portrait"
    lang: "..."              // "zh" if user's messages are predominantly Chinese, otherwise "en"
  },
  personality: {
    mbtiType: "XXXX",
    axes: {
      EI: { score: 0, letter: "X", label: "..." },
      SN: { score: 0, letter: "X", label: "..." },
      TF: { score: 0, letter: "X", label: "..." },
      JP: { score: 0, letter: "X", label: "..." }
    },
    summary: "..."
  },
  radar: {
    technicalDepth: 0,
    technicalBreadth: 0,
    communication: 0,
    decisionSpeed: 0,
    collaboration: 0,
    creativity: 0
  },
  rating: {
    tier: "...",
    label: "...",             // Chinese label: 夯爆了/夯/人上人/NPC/拉/拉完了
    labelEn: "...",           // English label: Legendary/Elite/Above Average/NPC/Below Average/Inactive
    emoji: "...",
    color: "...",
    reason: "...",
    strengthHighlights: [],
    growthAreas: []
  },
  famousMatch: {
    technical: {
      name: "...", emoji: "...",
      dimension: "Technical Spirit", dimensionZh: "技术灵魂",
      reason: "...", sharedTraits: []
    },
    strategic: {
      name: "...", emoji: "...",
      dimension: "Strategic Mind", dimensionZh: "思维内核",
      reason: "...", sharedTraits: []
    },
    communication: {
      name: "...", emoji: "...",
      dimension: "Communication Soul", dimensionZh: "表达人格",
      reason: "...", sharedTraits: []
    }
  },
  communication: {
    languages: {},
    directnessScore: 0,
    avgMessageLength: 0,
    topKeywords: [],
    questionRatio: 0
  },
  technical: {
    domains: {},
    topTools: []
  },
  workRhythm: {
    hourlyActivity: [/* 24 integers, one per hour */],
    sessionPattern: "...",
    avgSessionLength: "...",
    projectCount: 0
  },
  quotes: [
    { text: "...", context: "..." }
    // 3-5 representative quotes
  ]
};

IMPORTANT:

• Replace ONLY the PORTRAIT_DATA object. Do not modify any other part of the template.
• All values must be valid JavaScript (strings quoted, arrays bracketed, no trailing commas).
• The rest of the template's JavaScript will automatically render everything from this object.
• Keep quote texts under 100 characters. Anonymize any sensitive data (API keys, tokens, passwords, personal URLs).

Step 5: Generate persona skill

Read the persona skill template: → references/persona-skill-template.md

Generate a multi-file persona skill. Read the template for the full structure: → references/persona-skill-template.md

Create the following files under ~/.claude/skills/vibe-portrait-personas/me/:

me/
├── SKILL.md                        # Entry point (~80 lines)
├── portrait-meta.json              # Timestamps for incremental updates
└── references/
    ├── thinking-patterns.md        # From: technical depth + decision patterns
    ├── decision-framework.md       # From: decision patterns + collaboration style
    ├── communication-style.md      # From: communication style analysis
    ├── engineering-philosophy.md   # From: technical depth + breadth + anti-patterns
    └── mindset-markers.md          # Abstracted attitudes (NO raw quotes)

portrait-meta.json is critical — it stores lastMessageIndex (line count of history.jsonl at analysis time) and updatedAt timestamp. These enable the update subcommand to do incremental analysis later.

Create directories if they don't exist. me/ is reserved for the user's own persona.

Step 6: Write output files

Write the following files to the current working directory:

1. HTML Portrait:

vibe-portrait-YYYY-MM-DD.html

2. Portrait Image (auto-export):

After writing the HTML, attempt to generate a PNG screenshot of the portrait. Try these methods in order:

• If Playwright MCP is available (check for playwright_navigate and playwright_screenshot tools):

  1. Navigate to the local HTML file: file:///absolute/path/to/vibe-portrait-YYYY-MM-DD.html
  2. Wait for charts to render (2-3 seconds)
  3. Take a full-page screenshot, save as vibe-portrait-YYYY-MM-DD.png

• If Playwright is not available, try bash:

  # Try with npx playwright if installed
  npx --yes playwright screenshot --full-page "file://$(pwd)/vibe-portrait-YYYY-MM-DD.html" "vibe-portrait-YYYY-MM-DD.png" 2>/dev/null

• If neither works, skip silently. Tell the user they can use the "Export as Image" button on the HTML page instead.

3. Analysis JSON (for sync/merge):

vibe-portrait-analysis-YYYY-MM-DD.json

Contents:

{
  "version": "1.0",
  "machine": "<hostname from `hostname` command>",
  "exportedAt": "YYYY-MM-DD",
  "sourceFiles": ["~/.claude/history.jsonl"],
  "messageCount": 200,
  "portraitData": { /* the full PORTRAIT_DATA object */ }
}

3. Persona skill (also written in Step 5):

~/.claude/skills/vibe-portrait-personas/me/SKILL.md

Step 7: Sync to portrait repo

After writing all output files, ask the user:

Sync to your portrait repo?

1. Yes — create a new repo (first time)
2. Yes — push to existing repo
3. No — keep everything local