If you discover a vulnerability in this project, please report it privately. This template repository itself has minimal attack surface (it's mostly Markdown), but vulnerabilities can exist in:
- Code examples that contain security anti-patterns
- Instructions that could lead agents to make insecure choices
- Missing coverage of critical security domains
Do NOT open a public issue for security vulnerabilities.
Email: Create a security advisory via the GitHub Security tab.
The main branch is the only supported version. All security fixes are applied to main.
The AGENTS.md template covers extensive security guidance for projects that adopt it:
| Section | Security Domain |
|---|---|
| Section 13 | Security best practices — input validation, parameterized queries, rate limiting, secrets |
| Section 27.3 | Security constraints — SQL injection, eval/exec, path traversal, password hashing |
| Section 31.1 | Prompt injection defenses — instruction hierarchy, untrusted model/tool output, OWASP GenAI guardrails |
| Section 31.2 | Admin audit logging — IP, user agent, action tracking |
| Section 31.3 | API key encryption — Fernet-based at-rest encryption |
| Section 31.4 | Admin IP whitelist — network-level access control |
| Section 31.5 | Security CI/CD — gitleaks + pip-audit on every PR, plus scheduled scans |
| Section 36 | Explicit NEVER list — 9 categories including code, git, GitHub, security, and untrusted-input prohibitions |
| Section 37 | Pre-commit hooks — gitleaks, bandit, private key detection |
| Section 44 | Secrets management — tiered strategy, rotation, OIDC, scanning |
| Section 54 | MCP usage and guardrails — server vetting, tool-output trust, SSRF blocks |
| Section 55 | Prompt-injection defenses for agents — trust boundaries, guardrails |
| Section 58 | Supply-chain security — SBOM, provenance, signed artifacts, digest pinning |
We do not currently cover:
- Hardware security module (HSM) integration
- FIPS compliance requirements
- Zero-trust architecture patterns
If you identify a security domain not covered that leads to real-world vulnerabilities, please contribute (see CONTRIBUTING.md).