The Agent Skills Directory

[SAFE]: The skill establishes a robust DevSecOps workflow, directing the agent to use established security tools like npm audit, pip-audit, tfsec, and trivy for scanning vulnerabilities and misconfigurations in development and deployment pipelines.
[SAFE]: The instructions promote secure secret management, including the use of dedicated secret managers (e.g., AWS Secrets Manager, HashiCorp Vault) and strict avoidance of storing credentials in version control systems or environment files.
[SAFE]: The skill includes detailed guidance on supply chain security, specifically addressing risks like typosquatting and dependency confusion through best practices such as lockfile pinning, registry scoping, and signature verification.
[SAFE]: Guidance for CI/CD security emphasizes the principle of least privilege, recommending the use of OIDC for cloud authentication and restrictive branch protection rules to prevent unauthorized changes.
[SAFE]: Indirect Prompt Injection Attack Surface Analysis:
Ingestion points: Dependency lockfiles (e.g., package-lock.json), package manifests (package.json), and IaC manifests (Terraform/Kubernetes files) in the SKILL.md process description.
Boundary markers: The skill does not explicitly define prompt boundaries for data ingestion, but relies on specialized scanning tools to process these files.
Capability inventory: Tools like Bash, Edit, and WebFetch are authorized for auditing, scanning, and reporting tasks.
Sanitization: Not explicitly mentioned, but the focus on structured auditing tools rather than direct LLM parsing of complex untrusted logic significantly reduces the risk of malicious instruction execution.

security-devsecops