Skip to content

Security Model

Nurul Hadi edited this page Jun 5, 2026 · 1 revision

Security & Trust Model

Amanah Blueprint is built with a "Security First" philosophy. We have implemented several layers of protection to ensure the safety of your codebase and system.

1. Human-in-the-Loop

The framework never executes shell commands without your explicit permission. During the /build process, the agent will present each command and wait for your approval.

2. Untrusted Input Handling

The framework treats all content read from your project repository as untrusted. This protects against "Indirect Prompt Injection," where malicious instructions could be hidden in source code to hijack the AI agent.

3. Quality Gates

Every generated blueprint passes through security-specific audits:

  • SQL Injection Check: Ensures all queries use parameterized statements.
  • Auth Enforcement: Verifies that all new endpoints have authentication dependencies.
  • Secret Scan: Checks for hardcoded keys, tokens, or credentials.
  • Tenant Isolation: Ensures all queries are properly scoped for multi-tenant environments.

4. Supply-Chain Integrity

Our official installation scripts are pinned to specific, audited commit hashes. This prevents "branch hijacking" and ensures you are always running the code that has been verified.

5. Local Processing

Amanah Blueprint operates locally on your machine. Your proprietary code and specifications are only processed by the AI models you have authorized through Claude Code or Gemini CLI.

Clone this wiki locally