What problem does it solve? AI coding agents can accidentally print secrets into the conversation history when running shell commands like cat .env or kubectl get secret -o yaml, and once exposed, a credential persists in model context, caches, and logs, forcing costly rotation. This Skill enforces an absolute rule that secrets never pass through the agent window. ## Core Features & Use Cases - Prohibited command list: Blocks direct or indirect display of .env files, environment variables, keyrings, JWT tokens, cloud CLI credential output (aws/gcloud/az), and Kubernetes secrets. - Indirect verification patterns: Provides safe alternatives such as test -n presence checks, truncated SHA-256 digests, masked last-four-character display, and end-call health checks that report only success or failure. - Rationalization defense: Lists the common justifications agents use to bypass the rule ("it's just dev", "the user asked") and instructs the agent to stop when it catches itself reasoning this way. - Use Case: While debugging a failing deployment, the agent needs to confirm an API key is set; instead of running printenv, it runs test -n "${API_KEY:-}" and reports only that the key is present with its character count. ## Quick Start Ask the agent to verify that the API_KEY environment variable is configured without printing its value.
