Skip to content

Security: AgenticInternet/agentic-internet

Security

docs/SECURITY.md

Security

Hard Invariants

  • Never commit real API keys, tokens, cookies, or browser session credentials.
  • Validate untrusted input before browser automation, code execution, or filesystem access.
  • Do not log secrets or raw provider credentials.
  • Keep .env local and use .env.example for safe placeholders.
  • Treat generated code, scraped content, and browser results as untrusted data.
  • Prefer least-privilege provider keys for local development and CI.

Auth and Secrets

This package is CLI/library-first and does not own end-user authentication. Provider authentication is environment-variable based:

  • HUGGINGFACE_TOKEN
  • OPENAI_API_KEY
  • ANTHROPIC_API_KEY
  • SERPAPI_API_KEY
  • OPENROUTER_API_KEY
  • BROWSER_USE_API_KEY

Secrets should be loaded at runtime through settings and never passed through README examples as real values.

Data Classification

Class Examples Handling
Public README text, docs, public search snippets Safe to store in repo
Internal Plans, architecture notes, local logs Review before publishing
User-provided Prompts, URLs, uploaded data, scraped pages Avoid unnecessary persistence
Secret API keys, tokens, cookies, browser sessions Environment/secret manager only

Sensitive Areas

  • agentic_internet/tools/code_execution.py
  • agentic_internet/tools/browser_use.py
  • agentic_internet/tools/mcp_integration.py
  • Provider/model routing in orchestration modules

Changes in these areas require focused tests and explicit review of side effects.

There aren't any published security advisories