revibe（revibe 分析）

Purpose

Analyze codebases to understand architecture, file roles, design patterns, execution flows, and system design decisions. Returns structured insights inline and saves agent_context.json for persistent codebase understanding.

Revibe turns any GitHub repository into a complete architecture map — modules, dependencies, call chains, design decisions — in about 2 minutes.

When to Use

• User asks to "analyze", "understand", or "explain" a codebase or repository
• User clones or forks a new repo and wants to understand it quickly
• User says "what does this project do" or "how is this structured"
• User asks about architecture, patterns, or design decisions of a repo
• Another skill needs codebase context to work effectively (use agent mode)

Requirements

• REVIBE_API_KEY (required): API key for authentication. Free signup at https://app.revibe.codes → Settings → API Keys. Format: rk_live_xxxxx. Sent as X-Revibe-Key header.

Privacy Note

This skill sends your repository's GitHub URL to revibe.codes for analysis. Source code is stored securely in Google Cloud Storage to enable features like code exploration and re-analysis. Only you can access your uploaded projects. If you're working with private or sensitive repositories, review revibe.codes/privacy before proceeding.

Tip: To reduce permission prompts, you can optionally add Bash(curl *revibe.codes*) to your allowed tools via /allowed-tools.

Behavior

Step 1: Extract the Target

Determine the GitHub URL from one of these sources (in priority order):

1. Explicit URL in the message — e.g., https://github.com/owner/repo or owner/repo
2. Current working directory — If the user says "analyze this", "understand this project", or doesn't specify a repo, detect it automatically:

   # Get GitHub URL from current git repo
   REMOTE_URL=$(git remote get-url origin 2>/dev/null)

Convert SSH URLs to HTTPS:

   # git@github.com:owner/repo.git → https://github.com/owner/repo
   if [[ "$REMOTE_URL" == git@github.com:* ]]; then
     REMOTE_URL="https://github.com/${REMOTE_URL#git@github.com:}"
   fi
   # Strip .git suffix
   REMOTE_URL="${REMOTE_URL%.git}"

1. Ask the user — Only if not in a git repo and no URL provided: "Which repository would you like me to analyze?"

When using the current directory, confirm briefly: "Analyzing {repo_name} from your current directory..."

Step 2: Submit for Analysis

# Set up auth
API_KEY="${REVIBE_API_KEY:-}"
AUTH_HEADER=""
if [ -n "$API_KEY" ]; then
  AUTH_HEADER="-H \"X-Revibe-Key: $API_KEY\""
fi

# Submit
RESPONSE=$(curl -s -X POST "https://app-backend.revibe.codes/api/v1/analyze" \
  -H "Content-Type: application/json" \
  ${AUTH_HEADER} \
  -d "{\"github_url\": \"${REMOTE_URL}\"}")

JOB_ID=$(echo "$RESPONSE" | jq -r '.job_id')
STATUS=$(echo "$RESPONSE" | jq -r '.status')

Step 3: Wait for Completion

If status is "completed" or "already_analyzed", skip to Step 4.

Otherwise, poll every 10 seconds (max 3 minutes):

while [ "$STATUS" != "completed" ] && [ "$STATUS" != "error" ]; do
  sleep 10
  POLL=$(curl -s "https://app-backend.revibe.codes/api/v1/analysis/${JOB_ID}/status")
  STATUS=$(echo "$POLL" | jq -r '.status')
  PROGRESS=$(echo "$POLL" | jq -r '.progress // "working"')
  STEPS_DONE=$(echo "$POLL" | jq -r '.steps_completed')
  STEPS_TOTAL=$(echo "$POLL" | jq -r '.steps_total')
  # Show progress: "Analyzing... file_roles (5/10)"
done

If status is "error", tell the user: "Analysis failed. You can try directly at https://revibe.codes"

Step 4: Show Summary

SUMMARY=$(curl -s "https://app-backend.revibe.codes/api/v1/analysis/${JOB_ID}/summary")

Format the response as:

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  Revibe Analysis: {name}
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Architecture: {architecture_pattern}
Language:     {language}
Size:         {total_files} files
Entry point:  {entry_point}

Key Modules:
  {module.name}  — {module.description} ({module.loc} LOC)
  {module.name}  — {module.description} ({module.loc} LOC)
  ...

Patterns: {patterns joined by ", "}
Database: {database.type}, {database.tables} tables    ← only if database exists

agent_context.json saved to current directory.

Explore deeper:
  [1] Architecture diagram
  [2] File roles & structure
  [3] System design Q&A
  [4] Execution flows
  [5] Database schema
  [6] Full interactive analysis → {project_url}

Step 5: Save Agent Context

Always save the agent context file after showing the summary:

curl -s "https://app-backend.revibe.codes/api/v1/analysis/${JOB_ID}/agent-context" \
  > agent_context.json

This file gives the agent (and other skills) structured codebase understanding for future tasks.

Step 6: Handle Drill-Down

When the user picks a number (1-5), fetch that section:

# Map numbers to section names
SECTIONS=("technical_architecture" "file_roles" "system_design_qa" "story_flow" "database_schema")
SECTION_NAME="${SECTIONS[$CHOICE - 1]}"

SECTION_DATA=$(curl -s "https://app-backend.revibe.codes/api/v1/analysis/${JOB_ID}/section/${SECTION_NAME}")

Rendering sections:

• Architecture diagram [1]: Extract data.diagram, render as HTML file with mermaid.js (see Diagram Rendering section), open in browser. Show data.summary and data.layers as a markdown table in terminal.
• File roles [2]: Show as a table — file path, role, description. Group by module if available.
• System design Q&A [3]: Show Q&A pairs. These are the "why" behind architecture decisions.
• Execution flows [4]: Render each flow's diagram as HTML file, open in browser. Show narrative in terminal.
• Database schema [5]: Render ER diagram as HTML file, open in browser. Show table details in terminal.

For option [6], tell the user to open the URL in their browser.

After showing any section, offer to explore another: "Pick another section [1-6] or ask me anything about this codebase."

Agent Mode (Non-Interactive)

When another skill or the agent itself needs codebase context (not a direct user request):

1. Submit analysis (same as above)
2. Wait for completion
3. Fetch and return ONLY the agent_context.json — no formatting, no menu
4. Save to current directory silently

Detection: If the request comes from another skill or is prefixed with context like "I need to understand this codebase to..." — use agent mode.

Diagram Rendering

When a section contains Mermaid diagram data (diagram_mermaid or diagram field), render it as an HTML file and open in the browser:

1. Write the mermaid code to /tmp/revibe-diagram.html:

<!DOCTYPE html>
<html><head><meta charset="utf-8">
<title>TITLE</title>
<script src="https://cdn.jsdelivr.net/npm/mermaid@11/dist/mermaid.min.js"></script>
<style>
  body { background: #1a1a2e; display: flex; justify-content: center; padding: 40px; font-family: system-ui; }
  .container { max-width: 1200px; width: 100%; }
  h1 { color: #e0e0e0; font-size: 1.4rem; }
  .mermaid { background: #16213e; border-radius: 12px; padding: 24px; }
  .mermaid svg { width: 100%; }
</style>
</head><body>
<div class="container">
  <h1>TITLE</h1>
  <div class="mermaid">
MERMAID_CODE_HERE
  </div>
</div>
<script>mermaid.initialize({ startOnLoad: true, theme: 'dark' });</script>
</body></html>

1. Open: open /tmp/revibe-diagram.html (macOS) or xdg-open (Linux)
2. In the terminal, show a brief note: "Diagram opened in browser → /tmp/revibe-diagram.html"

Error Handling

Examples

Basic analysis:

User: analyze https://github.com/expressjs/express
→ Submit, wait, show summary card + save agent_context.json

Quick shorthand:

User: /revibe fastify/fastify
→ Resolve to https://github.com/fastify/fastify, analyze

Context for another task:

User: I need to add a feature to this repo, first help me understand it
→ Analyze, show summary, then use agent_context.json for subsequent work

Drill down:

User: show me the architecture diagram
→ Fetch architecture section, render Mermaid as ASCII

Configuration

Get an API key:

1. Sign up free at https://app.revibe.codes
2. Go to Settings → API Keys
3. Click "Generate API Key"
4. Add to OpenClaw config: openclaw config set REVIBE_API_KEY rk_live_xxxxx

Security & Privacy

• API key required: Anonymous usage is not supported. A free account and API key are required for all analysis requests
• No changes to your repo: Revibe only reads your code — the API key grants analysis access, never write operations on your repos
• Public & private repos: Public repos work with just an API key. Private repos require connecting your GitHub account at app.revibe.codes
• Code storage: Repo contents are stored securely in Google Cloud Storage for analysis and re-analysis
• Network calls: This skill calls app-backend.revibe.codes/api/v1 endpoints only
• Local output: Saves agent_context.json to your working directory and opens /tmp/revibe-diagram.html for diagram viewing