Skip to content

Security: LittleSongxx/SoulSearcher

Security

SECURITY.md

Security Policy

Reporting a Vulnerability

If you discover a security issue, please do not open a public GitHub issue.

Instead, report it privately:

  • Email the maintainer, or
  • Use GitHub Security Advisories (if enabled for this repository)

Please include:

  • A clear description of the issue
  • Steps to reproduce
  • Impact assessment (what an attacker can do)
  • Suggested fix (if you have one)

Secrets

  • Do not commit real API keys.
  • Keep .env local; use .env.example as a template.
  • This repo includes a lightweight secret scan for common key patterns:
python scripts/secret_scan.py

If you accidentally committed a real secret:

  1. Rotate the key immediately.
  2. Remove it from the codebase.
  3. Consider rewriting git history if it was pushed.

API Authentication

  • Set SOULSEARCHER_INTERNAL_API_KEY to enable Bearer token or X-API-Key authentication on most /api/* endpoints.
  • The SOULSEARCHER_AUTH_USER_HEADER (default: X-SoulSearcher-User) provides per-user session isolation behind an authenticated proxy.
  • For public deployments, enable rate limiting (RATE_LIMIT_ENABLED=true) to prevent abuse.

Sandbox Security

  • Remote sandboxes (E2B/Daytona): Code execution is fully isolated from the host. This is the recommended configuration for production.
  • Host bash (HOST_BASH_ENABLED=true): Opt-in only. When enabled, commands are audited and pipe-to-shell patterns are detected. Not recommended for untrusted environments.
  • Tool approval (TOOL_APPROVAL=true): Enable Human-in-the-Loop approval for potentially risky tool executions.

Skill Security

  • Installed skills undergo automatic security scanning via agent/skills/security_scanner.py.
  • Each skill declares allowed tools via allowed-tools in SKILL.md frontmatter; tool allowlisting is enforced at runtime.
  • Skill evolution (SKILL_EVOLUTION_ENABLED) is disabled by default. Enable only in trusted environments.

Dependency Updates

  • Dependabot is configured for weekly automated updates (pip, npm, GitHub Actions).
  • CI runs on every PR: linting, secret scanning, compilation checks, and tests.
  • Pre-commit hooks enforce code quality (trailing whitespace, YAML validation, merge conflict detection, Ruff linting + formatting).

Secure Development

# Run secret scan before commits
make secret-scan

# Full pre-commit check
make check          # lint + test + secret-scan

# Pre-commit hooks (auto-run on git commit)
pre-commit install

There aren't any published security advisories