Skip to content

Latest commit

 

History

History
36 lines (30 loc) · 1.92 KB

File metadata and controls

36 lines (30 loc) · 1.92 KB

TrustSignal AI Engineering Guardrails

Compliance posture

This repository is operated for SOC 2 readiness and Vanta-aligned evidence collection. Every change must preserve security controls, auditability, least privilege, and deterministic CI evidence.

Hard rules

  • Never hardcode secrets, API keys, OAuth tokens, refresh tokens, passwords, private keys, or session cookies.
  • Never emit sensitive values or raw PII in logs, stack traces, telemetry, or test fixtures.
  • Use environment variables or an approved secret manager for all sensitive configuration.
  • Use standards-based auth (OAuth 2.0/OIDC/JWT with vetted libraries); do not invent custom cryptography.
  • Require TLS 1.2+ for all network paths carrying sensitive data.
  • Preserve structured audit logs for security-relevant events (timestamp, actor/service, action, target, request_id, result).
  • Enforce least privilege for IAM roles, service accounts, DB permissions, and API scopes.
  • Do not disable auth checks, logging, tests, linting, or security controls to make a task pass.
  • Prefer maintained dependencies and pin versions where practical.

Required workflow for AI-generated changes

Before coding:

  1. Identify trust boundaries.
  2. Identify secrets/PII touched by the task.
  3. Identify compliance-sensitive surfaces (auth, storage, logging, third-party integrations).
  4. Propose the safest minimal implementation.

During implementation:

  • Keep diffs small and reviewable.
  • Add or update tests for security-critical behavior.
  • Preserve backward compatibility unless a migration is explicitly included.

When finishing:

  • Summarize files changed.
  • Summarize security impact.
  • List validation commands run and outcomes.
  • Call out unresolved risks or control gaps.

Refusal rule

If a request is insecure or non-compliant (for example: exposing secrets, bypassing auth, logging PII, or weakening controls), refuse the unsafe approach and provide a safer alternative immediately.