Skip to content

USAGE.md

Latest commit

 

History

History
344 lines (241 loc) · 7.98 KB

USAGE.md

File metadata and controls

344 lines (241 loc) · 7.98 KB
Raw

Cybos Usage Guide

Detailed workflows and best practices for using Cybos.

Research Workflows

Company Research

/cyber-init-deal "Acme Corp"
/cyber-research-company "Acme Corp"

What happens:

  • Parallel MCP calls gather data (Perplexity, Exa, Parallel AI)
  • 4 parallel agents analyze (company, market, financial, team)
  • Synthesizer consolidates with investment lens
  • Report saved to /deals/acme-corp/research/MMDD-<slug>-YY/report.md

Duration: 4-8 minutes

Technology Research

/cyber-research-tech "Trusted Execution Environments"

What happens:

  • Deep technical research via MCP tools
  • 3 parallel agents (tech, market, financial)
  • Synthesis with investment opportunity focus
  • Report saved to /research/<topic>/MMDD-<slug>-YY/report.md

Duration: 5-8 minutes

Market Research

/cyber-research-market "AI Infrastructure"

What happens:

  • Market sizing and dynamics research
  • 3 parallel agents (market, technology, investment activity)
  • Report includes TAM, key players, white spaces, timing

Duration: 8-12 minutes

Research Intensity Flags

Flag Duration Quality When to use
--quick 10-30s Basic Fast fact-checking
--standard 2-5m Good Default, most research
--deep 5-15m Excellent Critical decisions, IC prep

Content Workflows

Tweets

/cyber-tweet "AI agents using crypto wallets"

Workflow: Draft → Review → Polish → Save

Loads voice-identity.md + writing-style-en.md, creates hook-driven tweet. Output: /content/tweets/MMDD-<slug>-YY.md

Duration: 2-5 minutes

Essays

/cyber-essay "Why compute is becoming a commodity"

Workflow: Draft → Review → Polish

Structure: Hook → Stakes → Mechanism → Turn → Landing (500-2500 words)

Duration: 15-30 minutes

Telegram Posts

/cyber-post "AI automation for VCs"

Creates Russian post + English translation. Uses writing-style-ru.md.

Images

/cyber-image "Solitary figure in brutalist VC office"

Workflow: Draft prompt → Review → Generate via Nano Banana MCP

Styles: Automatically inferred from keywords:

  • info - infographics, diagrams
  • mural - sacred transformation aesthetic
  • cyberpunk - grounded futurism (default)

Duration: 3-7 minutes

Source-Driven Content

Use @path/to/file.md to reference source material:

# Expand idea into essay
/cyber-essay @content/ideas/agent-economy.md "Full essay with examples"

# Distill research into tweet
/cyber-tweet @deals/acme-corp/research/report.md "Key insight"

# Multiple sources
/cyber-essay "TEE market" @content/ideas/tee.md @research/tee-market/

Telegram Processing

/cyber-telegram                    # 1 unread dialog
/cyber-telegram --count 3          # 3 unread dialogs
/cyber-telegram --user "@username" # Specific person
/cyber-telegram --requests         # Message requests (non-contacts)
/cyber-telegram --dry-run          # Read only, no drafts

What happens:

  1. Fetches messages via GramJS MTProto
  2. Looks up entity context from database
  3. AI generates draft replies
  4. Drafts saved to Telegram (not sent)
  5. Per-person logs saved to /context/telegram/<slug>.md

GTD Runner

/cyber-gtd                  # Plan first GTD item
/cyber-gtd --count 3        # Process 3 items
/cyber-gtd --execute        # Skip plan, run immediately
/cyber-gtd --project slug   # Only tasks under # heading

Task Classification:

Pattern Workflow
"ask for call", "message" outreach
"call with", "meeting" call-prep
"podcast" podcast
company name research

Output: /content/work/MMDD-<slug>.md with draft + pending actions

DD Memo Generation

# Prerequisite: run research first
/cyber-research-company "Company"

# Generate memo
/cyber-memo "Company"

What happens:

  1. Load all research from /deals/<company>/research/
  2. memo-analyst (Opus) applies investment rubric
  3. memo-writer (Sonnet) fills template
  4. Output: /deals/<company>/memo/memo.md

Duration: 8-15 minutes

Morning Brief

/cyber-brief              # Generate brief
/cyber-brief --email-days 7  # 7 days of emails

Data sources: Telegram, Gmail, Calendar, GTD.md

Web view: http://localhost:3847

Social Scheduling

/cyber-schedule @content/tweets/file.md              # Schedule content
/cyber-schedule @content/file.md --image @img.png    # With image

Schedules to Twitter and/or LinkedIn via Typefully.

Projects

/cyber-init-project "Cyber Accelerator Q1"
/cyber-project cyber-accelerator-q1
/cyber-projects
/cyber-gtd --project cyber-accelerator-q1

Projects are # headings in GTD.md. Use /projects/ folder for multi-month initiatives with artifacts.

Agent System

Agent Model Purpose
company-researcher Haiku Business model, product, traction
market-researcher Haiku TAM, dynamics, competitive landscape
financial-researcher Haiku Funding history, metrics, valuation
team-researcher Haiku Founder backgrounds, assessment
tech-researcher Haiku Technology deep-dives, moat analysis
quality-reviewer Sonnet Gap analysis (deep mode only)
content-writer Sonnet Tweets, essays
memo-analyst Opus Strategic investment analysis
memo-writer Sonnet Memo from template
synthesizer Sonnet Consolidate research

Agents run in parallel via Task tool.

MCP Integration

Tiered Strategy

1. FAST SEARCH (seconds)
   └─ perplexity search OR exa search

2. DEEP RESEARCH (1-5 minutes)
   └─ parallel-task createDeepResearch

3. TARGETED EXTRACTION
   └─ exa getContents (PRIMARY)
   └─ Fallback: parallel-search web_fetch

4. HARD SCRAPING (last resort)
   └─ playwright for JS-heavy sites

Primary MCP Servers

Server Purpose
exa Web search, content extraction (PRIMARY)
perplexity Fast search + deep research
parallel-task Deep research tasks
nano-banana Image generation (Gemini 3.0)
typefully Social scheduling
gmail Email + Calendar (unified)

Best Practices

1. Research Before Memo

Always run /cyber-research-company before /cyber-memo.

2. Iterative Content

Content workflows expect feedback:

  • "Good" or "ship it" = approved
  • "More technical" = adjust tone
  • "Different hook" = revise opening

3. Use Ideas Folder

Capture rough thoughts in /content/ideas/ and expand later:

/cyber-essay @content/ideas/agent-econ.md "Full essay"

4. Monitor Logs

Check /cyber-log regularly to track activity.

5. Update Deal Context

When new information emerges, update /deals/<company>/index.md.

Troubleshooting

Hook Not Loading

chmod +x .claude/hooks/load-context.ts
bun .claude/hooks/load-context.ts < /dev/null

MCP Server Not Found

  • Check API key in .env
  • Verify .mcp.json configuration
  • Test: npx -y @perplexity-ai/mcp-server --help

Research Returns Limited Data

  • Run research again (MCP services may have been slow)
  • Try different MCP tools
  • Check API quotas

Image Generation Fails

  • Check GEMINI_API_KEY
  • Simplify prompt if safety checker triggered
  • Use continue_editing for refinements

Development

Adding a Workflow

  1. Create: .claude/skills/<Skill>/workflows/<workflow>.md
  2. Create command: .claude/commands/cyber-<command>.md
  3. Reference workflow with @.claude/skills/...

Adding an Agent

  1. Create: .claude/agents/<agent-name>.md
  2. Define purpose, output format, guidelines
  3. Reference in workflow as Task: <agent-name>

Customizing Brand Voice

Edit:

  • context/style/voice-identity.md - shared persona
  • context/style/writing-style-en.md - English style
  • context/style/writing-style-ru.md - Russian style

See ARCHITECTURE.md for full technical reference.