cai/docs/continue_mode.md

332 lines
9.9 KiB
Markdown

# CAI Continue Mode
## Overview
The `--continue` flag enables CAI agents to operate autonomously by automatically generating intelligent continuation prompts when they would normally stop and wait for user input. This feature uses AI-powered analysis to provide contextual advice based on the conversation history, allowing agents to work on complex tasks without manual intervention.
## Quick Start
```bash
# Tell jokes continuously
cai --continue --prompt "tell me a joke about security"
# Analyze code autonomously
cai --continue --prompt "find all SQL injection vulnerabilities in this codebase"
# Run security audit
cai --continue --prompt "perform a comprehensive security audit"
```
## Example: Security Jokes with Continue Mode
Here's what happens when you run `cai --continue --prompt "tell me a joke about security"`:
```bash
$ cai --continue --prompt "tell me a joke about security"
🤖 Processing initial prompt: tell me a joke about security
Agent: Why did the hacker break up with their password?
Because it wasn't strong enough! 💔🔐
🤖 Auto-continuing with: Tell another cybersecurity joke or pun.
Agent: Why don't cybersecurity experts tell secrets at parties?
Because they're afraid of social engineering! 🎉🕵️
🤖 Auto-continuing with: Tell another cybersecurity joke or pun.
Agent: What's a hacker's favorite season?
Phishing season! 🎣💻
[Continues until interrupted with Ctrl+C]
```
## How It Works
### 1. Intelligent Context Analysis
When an agent completes a turn, the continuation system analyzes:
- **Original request**: The initial task or prompt from the user
- **Conversation history**: Recent messages and responses
- **Tool usage**: Which tools were used and their outputs
- **Error states**: Any errors encountered and their types
- **Task progress**: Current state of task completion
### 2. AI-Powered Continuation Generation
The system uses the configured AI model (default: alias1) to generate contextual continuation prompts:
```python
# The system creates a detailed context summary
context_summary = """
ORIGINAL TASK: Tell me a joke about security
CONVERSATION FLOW:
User: Tell me a joke about security
Agent: Why did the hacker break up with their password? Because it wasn't strong enough!
CURRENT STATUS:
- Last action: Told a cybersecurity joke
- Tools used: None
- Errors: No
Generate a specific continuation prompt...
"""
```
### 3. Smart Fallback System
When the AI model is unavailable, the system provides intelligent fallbacks based on context:
| Scenario | Fallback Continuation |
|----------|----------------------|
| Security joke told | "Tell another cybersecurity joke or pun." |
| File not found | "Search for the correct file path or create the missing resource." |
| Search completed | "Examine the search results in detail and investigate the most relevant findings." |
| Security analysis | "Analyze the code for security vulnerabilities like injection flaws or authentication issues." |
| Permission denied | "Check permissions and try accessing the resource with appropriate credentials." |
## Common Use Cases
### 1. Automated Security Audits
```bash
cai --continue --prompt "perform a security audit of the authentication system"
```
The agent will:
- Search for authentication-related files
- Analyze code for vulnerabilities
- Check for common security issues
- Generate a comprehensive report
### 2. Continuous Bug Hunting
```bash
cai --continue --prompt "find and document all XSS vulnerabilities"
```
The agent will:
- Search for user input handling code
- Identify potential XSS vectors
- Document findings
- Suggest fixes
### 3. Extended Code Analysis
```bash
cai --continue --prompt "analyze this codebase for OWASP Top 10 vulnerabilities"
```
The agent will:
- Systematically check for each vulnerability type
- Provide detailed findings
- Continue until all categories are covered
### 4. Entertainment Mode
```bash
cai --continue --prompt "tell me cybersecurity jokes and fun facts"
```
The agent will:
- Tell jokes about security topics
- Share interesting security facts
- Continue entertaining until stopped
## Configuration
### Environment Variables
```bash
# Use a different model for continuation generation
export CAI_MODEL=gpt-4
cai --continue --prompt "analyze this code"
# Set a fallback model if primary fails
export CAI_CONTINUATION_FALLBACK_MODEL=gpt-3.5-turbo
cai --continue --prompt "test application security"
# Configure API keys for custom models
export ALIAS_API_KEY=your-api-key
cai --continue --prompt "perform penetration testing"
```
### Combining with Other CAI Features
```bash
# Use specific agent with continue mode
CAI_AGENT_TYPE=bug_bounter_agent cai --continue --prompt "test example.com"
# Set workspace for file operations
CAI_WORKSPACE=project1 cai --continue --prompt "audit all Python files"
# Enable streaming for real-time output
CAI_STREAM=true cai --continue --prompt "monitor security events"
```
## Advanced Features
### Continuation Decision Logic
The system decides whether to continue based on:
1. **Completion indicators**: Stops if agent says "completed", "finished", "done"
2. **Active work detection**: Continues if tools are being used
3. **Error recovery**: Attempts to resolve errors automatically
4. **Task progress**: Evaluates if the original goal is achieved
### Context-Aware Prompts
The continuation prompts adapt based on:
- **Task type**: Security analysis, testing, code review, etc.
- **Current state**: Errors, findings, progress
- **Tool usage**: Different prompts for different tools
- **Conversation flow**: Maintains coherent task progression
## Best Practices
### 1. Clear Initial Prompts
```bash
# Good - Specific and actionable
cai --continue --prompt "find SQL injection vulnerabilities in user.py"
# Less effective - Too vague
cai --continue --prompt "check security"
```
### 2. Monitor Progress
- Check output periodically to ensure correct direction
- Use Ctrl+C to stop if needed
- Review logs for detailed execution history
### 3. Set Appropriate Limits
```python
# In code integration, use max_turns
run_cai_cli(
starting_agent=agent,
initial_prompt="analyze security",
continue_mode=True,
max_turns=10 # Limit to 10 turns
)
```
### 4. Error Handling
The system automatically:
- Retries failed operations with different approaches
- Searches for alternatives when files are missing
- Adjusts strategies based on error types
## Troubleshooting
### Issue: Generic Continuation Messages
**Symptom**: Always see "Continue working on the task based on your previous findings"
**Solution**:
- Check model configuration is correct
- Ensure API keys are valid
- Review debug logs for API errors
### Issue: Continuation Not Triggering
**Symptom**: Agent stops after completing a task
**Possible causes**:
- Agent explicitly said task is "completed" or "done"
- No recent tool usage detected
- Error in continuation module
**Solution**:
- Use more open-ended initial prompts
- Check logs for completion indicators
- Verify --continue flag is properly set
### Issue: Infinite Loops
**Symptom**: Agent keeps doing the same thing
**Solution**:
- Set max_turns limit
- Use more specific initial prompts
- Interrupt with Ctrl+C and refine the task
## Technical Implementation
### Core Components
1. **`src/cai/continuation.py`**: Main continuation logic
- `generate_continuation_advice()`: Creates AI-powered prompts
- `should_continue_automatically()`: Decides when to continue
2. **`src/cai/cli.py`**: Integration point
- `--continue` flag handling
- Continuation loop implementation
3. **Context Analysis**:
- Extracts conversation history
- Identifies tool usage patterns
- Detects error conditions
### API Integration
The continuation system uses LiteLLM for model calls:
```python
response = await litellm.acompletion(
model=model_name,
messages=[{"role": "user", "content": context_summary}],
temperature=0.3, # Low temperature for focused responses
max_tokens=150
)
```
## Examples Gallery
### Security Audit Continuation
```
Original: "Audit the login system"
→ "Search for authentication-related files in the codebase."
→ "Analyze the login function for SQL injection vulnerabilities."
→ "Check password hashing implementation for security best practices."
→ "Review session management for potential security issues."
```
### Bug Bounty Continuation
```
Original: "Test example.com for vulnerabilities"
→ "Perform initial reconnaissance to gather information about the target."
→ "Scan for exposed endpoints and services."
→ "Test authentication endpoints for common vulnerabilities."
→ "Check for information disclosure in error messages."
```
### Code Review Continuation
```
Original: "Review api.py for security issues"
→ "Analyze input validation in API endpoints."
→ "Check for proper authentication and authorization."
→ "Review error handling for information leakage."
→ "Examine data serialization for injection vulnerabilities."
```
## Example Scripts
Explore working examples in the `examples/` directory:
### Security Jokes Example
```python
# examples/continue_mode_jokes.py
# Demonstrates continuous joke telling with --continue flag
python examples/continue_mode_jokes.py
```
### Security Audit Example
```python
# examples/continue_mode_security_audit.py
# Shows autonomous vulnerability scanning with --continue
python examples/continue_mode_security_audit.py
```
These examples demonstrate:
- How to use --continue flag programmatically
- Handling continuous output
- Graceful interruption with Ctrl+C
- Practical security use cases
## Summary
The `--continue` flag transforms CAI into an autonomous cybersecurity assistant capable of:
- Working independently on complex tasks
- Recovering from errors intelligently
- Maintaining context across multiple operations
- Providing entertainment with continuous jokes
Whether you're conducting security audits, hunting for bugs, or just want some cybersecurity humor, continue mode keeps your agent working until the job is done.