We release patches for security vulnerabilities. Which versions are eligible for receiving such patches depends on the CVSS v3.0 Rating:

We take the security of cogito seriously. If you have discovered a security vulnerability in this project, please report it responsibly.

Please DO NOT report security vulnerabilities through public GitHub issues.

Instead, please report them via one of the following methods:

1. GitHub Security Advisories (Preferred)

   • Go to the Security tab of this repository
   • Click "Report a vulnerability"
   • Fill out the form with details about the vulnerability

2. Email

   • Send details to the repository maintainer through GitHub profile contact information
   • Use PGP encryption if possible for sensitive details

Please include the following information (as much as you can provide) to help us better understand the nature and scope of the possible issue:

• Type of issue (e.g., prompt injection, data leakage, memory exposure, etc.)
• Full paths of source file(s) related to the manifestation of the issue
• The location of the affected source code (tag/branch/commit or direct URL)
• Any special configuration required to reproduce the issue
• Step-by-step instructions to reproduce the issue
• Proof-of-concept or exploit code (if possible)
• Impact of the issue, including how an attacker might exploit the issue
• Your name and affiliation (optional)

• Acknowledgment: We will acknowledge receipt of your vulnerability report within 48 hours
• Initial Assessment: Within 7 days, we will provide an initial assessment of the report
• Resolution Timeline: We aim to resolve critical issues within 30 days
• Disclosure: We will coordinate with you on the disclosure timeline

We prefer all communications to be in English.

When using cogito in your applications, we recommend:

1. Keep Dependencies Updated

   go get -u github.com/zoobzio/cogito

2. Protect LLM Credentials

   • Never hardcode API keys in source code
   • Use environment variables or secret managers
   • Rotate keys regularly

3. Validate LLM Outputs

   • Treat LLM responses as untrusted input
   • Validate extracted data before using in critical paths
   • Use typed extraction (Analyze[T]) for structured data

4. Memory Security

   • Be mindful of what data is persisted to Memory
   • Implement access controls at the application layer
   • Consider data retention policies for Notes

5. Context Management

   • Use Compress/Truncate to limit context exposure
   • Be aware that Notes accumulate across pipeline steps
   • Review what information flows between primitives

6. Resource Management

   • Use timeouts for all LLM operations
   • Implement rate limiting for LLM calls
   • Monitor token usage and costs

cogito includes several built-in security features:

• Type Safety: Generic types prevent type confusion in extracted data
• Context Support: Built-in cancellation and timeout support
• Error Isolation: Errors are properly wrapped and traced
• Observability: Full signal emission for audit trails via capitan
• Session Isolation: Separate LLM sessions prevent context leakage

When building LLM-powered systems:

• Prompt Injection: Validate and sanitize user inputs before including in prompts
• Data Exfiltration: Be cautious about what context is sent to external LLM providers
• Hallucination: Don't trust LLM outputs for security-critical decisions without validation
• Cost Attacks: Implement safeguards against requests designed to consume excessive tokens

This project uses:

• CodeQL: GitHub's semantic code analysis for security vulnerabilities
• Dependabot: Automated dependency updates
• golangci-lint: Static analysis including security linters (gosec)
• Codecov: Coverage tracking to ensure security-critical code is tested

• Security vulnerabilities will be disclosed via GitHub Security Advisories
• We follow a 90-day disclosure timeline for non-critical issues
• Critical vulnerabilities may be disclosed sooner after patches are available
• We will credit reporters who follow responsible disclosure practices

We thank the following individuals for responsibly disclosing security issues:

This list is currently empty. Be the first to help improve our security!

Last Updated: 2025-12-05