Access cheat sheets, secure-coding checklists, and IaC scanning scripts.
Integrate them into CI/CD to raise security and visibility across your SDLC.
Zero-Trust Prompting: Preventing Prompt Injection
·
AI SafetySecure SDLCCheat Sheets
Keep sensitive data out of prompts; validate and constrain outputs.
Treat model prompts and outputs like any other untrusted data flow. Redact sensitive inputs, validate responses,
and put AI assistants behind explicit policy checks. Below is a compact cheat-sheet you can drop into your team wiki.
AI Safety Cheat-Sheet (excerpt)
Prompt & Output Handling
• Do not include secrets/PII in prompts. Use masked test data only.
• Validate outputs (schema/regex/type) before using them in code or tools.
• Keep tools/extensions on an allowlist; restrict filesystem/network access.
• Rate-limit and log prompts/responses securely; set retention & access controls.
• Prefer retrieval-augmented generation (RAG) with sanitized sources.
• Never directly execute generated commands; require human confirmation or sandboxes.
PR Security Checklist (add to your repo)
.github/pull_request_template.md
### Security checklist
- [ ] Input validated/encoded (XSS/Injection)
- [ ] AuthZ checks present for sensitive operations
- [ ] No hard-coded secrets / tokens
- [ ] TLS/HTTPS enforced for endpoints
- [ ] Tests updated for risky flows
- [ ] If AI used: no sensitive data in prompts; outputs validated
Terraform Pitfalls & How to Scan Your IaC
·
TerraformKubernetesIaC Scanning
Scan for permissive IAM, public storage, and weak network rules before they ship.
Catch misconfigurations early. Use tfsec for Terraform, Trivy for files/containers, and
Semgrep for policy rules. The following workflows plug into GitHub Actions and GitLab CI.
From CI to CD: SBOM, Secrets Scanning & Policy Gates
·
CI/CDVisibilityCompliance
Build visibility with SBOMs, secrets scanning, and fail-on-severity policies.
Add local bootstrap scripts for developers, then enforce build-time controls that are visible to the whole org.
Start with a one-liner, add SBOM & secrets scanning, then gate on severity.