Anthropic (Claude)

Anthropic’s Claude models provide advanced reasoning capabilities for security analysis. This guide covers configuration and best practices.

Available Models

Authentication

API Key Setup

1. Get your API key from console.anthropic.com
2. Run the authentication command:

cyberstrike auth login
# Select: Anthropic
# Enter your API key

Environment Variable

export ANTHROPIC_API_KEY="sk-ant-..."

Add to your shell profile for persistence:

echo 'export ANTHROPIC_API_KEY="sk-ant-..."' >> ~/.zshrc

Configuration File

{
  "provider": {
    "anthropic": {
      "options": {
        "apiKey": "{env:ANTHROPIC_API_KEY}"
      }
    }
  }
}

Model Configuration

Set Default Model

{
  "model": "anthropic/claude-sonnet-4-20250514"
}

Command Line Override

cyberstrike --model anthropic/claude-opus-4-5-20251101

In-Session Switching

In the TUI, run /models to open the model picker (bound to the “Switch model” command).

Extended Thinking

Enable extended thinking for complex security analysis:

{
  "provider": {
    "anthropic": {
      "options": {
        "thinking": {
          "type": "enabled",
          "budgetTokens": 10000
        }
      }
    }
  }
}

When to Use Extended Thinking

• Complex vulnerability chains
• Multi-step attack planning
• Code review with deep analysis
• Architecture security review

Token usage

In the session TUI header, CyberStrike shows context usage for the most recent assistant message with non-zero output tokens: a token total (input, output, reasoning, and cache read/write) and, when the model defines a context limit, a percentage of that limit. It also shows estimated session cost in USD as the sum of cost on assistant messages.

Prompt caching

For direct Anthropic models (providerID / SDK npm @ai-sdk/anthropic), CyberStrike adds Anthropic ephemeral cache control to selected system and recent messages when building the request. There is no separate cacheControl flag in provider.anthropic.options for this behavior.

Rate limits

Rate limits depend on your Anthropic account and product; refer to Anthropic’s documentation for current quotas.

For retryable API errors, CyberStrike uses exponential backoff (starting at 2s, doubling per attempt, capped at 30s when no Retry-After headers are present) and honors Retry-After / Retry-After-Ms when the provider returns them.

Best Practices

Model Selection

Cost Optimization

1. Use Haiku for repetitive tasks
2. Use /compact (or /summarize) for long sessions
3. Limit context with focused prompts

Security

1. Never commit API keys to repositories
2. Use environment variables
3. Rotate keys periodically
4. Monitor usage on Anthropic dashboard

Troubleshooting

Invalid API Key

Error: Invalid API key

Verify your key:

• Matches the format described when connecting Anthropic in the app (API keys typically start with sk-ant-api…)
• No trailing whitespace
• Key is active in console.anthropic.com/settings/keys

Rate Limit Exceeded

Error: Rate limit exceeded

Solutions:

• Wait and retry
• Upgrade your tier
• Use smaller context

Model Not Available

Error: Model not found

Check model ID is correct and available in your region.

Related Documentation

• Providers Overview - All supported providers
• Configuration - Full configuration options
• Authentication - Setup guide