Imported from chendren/computer (skills/computer-operations/SKILL.md). Install upstream with npx skills add chendren/computer --skill computer-operations. Copyright stays with the author.

Computer Operations

The Computer plugin provides a Star Trek LCARS-themed interface for voice transcription, AI analysis, chart generation, web search, monitoring, logging, comparison, summarization, translation, explanation, knowledge management, workflow pipelines, and export/reports.

Server Management

• Start: /computer:computer or bash "${CLAUDE_PLUGIN_ROOT}/scripts/start.sh"
• Stop: /computer:computer stop
• Status: /computer:status
• Port: 3141
• URL: http://localhost:3141

API Endpoints (push data to the UI)

All POST endpoints broadcast to connected WebSocket clients for real-time display.

Commands

Agents

UI Panels

Desktop Notifications

macOS desktop notifications fire automatically for:

• Analysis complete: When a new analysis is posted
• Monitor alerts: When a monitor status is "alert" or "triggered"
• Monitor updates: General notification for monitor status changes

Smart Voice Routing

When the user speaks via microphone, the input is analyzed for command intent:

• "Computer, analyze [text]" → auto-routes to /computer:analyze
• "Computer, search [query]" → auto-routes to /computer:search
• "Computer, compare [items]" → auto-routes to /computer:compare
• "Computer, summarize [text]" → auto-routes to /computer:summarize
• "Computer, monitor [target]" → auto-routes to /computer:monitor
• "Computer, translate [text]" → auto-routes to /computer:translate
• "Computer, explain [topic]" → auto-routes to /computer:explain
• "Captain's log [entry]" → auto-routes to /computer:log
• "Computer, remember [fact]" → auto-routes to /computer:know
• "Computer, status" → auto-routes to /computer:status
• "Computer, briefing" → auto-routes to /computer:brief

A green "Routed" badge appears in the UI when voice routing is active.

LCARS Chart Colors

Primary: #FF9900, #CC99CC, #9999FF, #FF9966, #CC6699, #99CCFF, #FFCC00 Background: #000000, Text: #FF9900, Grid: #333333

Data Storage

JSON files in ${CLAUDE_PLUGIN_ROOT}/data/:

• transcripts/ — Voice and file transcripts
• analyses/ — Analysis results, summaries, explanations, translations
• sessions/ — Conversation logs
• logs/ — Captain's log entries
• monitors/ — Monitor configurations and status
• comparisons/ — Comparison results
• vectordb/ — LanceDB vector database (knowledge embeddings, chunks, metadata)

Cross-Referencing Prior Results

When the user references earlier work ("what did we find about X?", "in the last analysis...", "compare this to what we found before"), you should:

1. Fetch stored data: Query the relevant API endpoint to retrieve prior results

   curl -s http://localhost:3141/api/analyses | jq '.[0:5]'
   curl -s http://localhost:3141/api/logs
   curl -s http://localhost:3141/api/transcripts
   curl -s http://localhost:3141/api/knowledge

2. Search through results: Look for matching content, topics, or entities
3. Reference naturally: "As noted in the analysis from [timestamp], the sentiment was..."
4. Link findings: Connect current work to prior results when relevant

Proactive Suggestions

Based on context, suggest relevant follow-up actions:

• After an analysis: "Would you like me to compare this with a previous analysis?" or "Shall I search for more information on [top topic]?"
• After a search: "Would you like me to analyze these findings in more detail?"
• After a transcription: "Would you like me to summarize the key points?" or "Shall I extract action items?"
• After a comparison: "Would you like a deeper explanation of the key differences?"
• After a monitor triggers: "The monitor detected a change. Would you like me to analyze the difference?"
• After any result: "Shall I store this in the knowledge base for future reference?"

Pushing Data to UI

From any command or agent, push data to the running UI by writing JSON to a temp file and using curl:

# Write JSON (avoids shell escaping issues)
# Then POST with file reference:
curl -X POST http://localhost:3141/api/analysis -H 'Content-Type: application/json' -d @/tmp/computer-result.json
curl -X POST http://localhost:3141/api/charts -H 'Content-Type: application/json' -d @/tmp/computer-chart.json
curl -X POST http://localhost:3141/api/search-results -H 'Content-Type: application/json' -d @/tmp/computer-search.json
curl -X POST http://localhost:3141/api/logs -H 'Content-Type: application/json' -d @/tmp/computer-log.json
curl -X POST http://localhost:3141/api/monitors -H 'Content-Type: application/json' -d @/tmp/computer-monitor.json
curl -X POST http://localhost:3141/api/comparisons -H 'Content-Type: application/json' -d @/tmp/computer-comparison.json
curl -X POST http://localhost:3141/api/knowledge -H 'Content-Type: application/json' -d @/tmp/computer-knowledge.json

Vector Knowledge Base

The knowledge base uses LanceDB for vector storage with Ollama nomic-embed-text (768-dim) for embeddings.

Chunking Strategies

• fixed — N-character chunks with overlap
• sentence — Split on sentence boundaries (default for short facts)
• paragraph — Split on double newlines (default for medium text)
• sliding — Fixed window with configurable step
• semantic — Split when embedding similarity drops below threshold
• recursive — Split by headers, then paragraphs, then sentences (default for long docs)

Search Methods

• vector — Cosine similarity nearest neighbors
• keyword — BM25-style text search
• hybrid — Combined vector + keyword (default, best general-purpose)
• mmr — Maximal Marginal Relevance (diversity-promoting)
• multi_query — Multiple query variations merged via Reciprocal Rank Fusion

Knowledge Ingestion Format

{
  "text": "The content to store",
  "title": "Optional title",
  "source": "user|analysis|search|monitor|import",
  "confidence": "high|medium|low",
  "tags": ["tag1", "tag2"],
  "chunk_strategy": "paragraph",
  "chunk_options": {}
}

Knowledge Search Format

{
  "query": "search text",
  "method": "hybrid",
  "limit": 10,
  "metadata_filter": { "source": "user", "confidence": "high", "tags": ["tag"] }
}

TTS Acknowledgements

For short status messages, include "speak": true in status broadcasts to have the Computer speak:

curl -X POST http://localhost:3141/api/tts/speak -H 'Content-Type: application/json' -d '{"text":"Analysis complete."}'

Max 300 characters. Only use for brief acknowledgements, not full results.