sensitive-data-protection

name	sensitive-data-protection
description	Strict guidelines for protecting user privacy and preventing access to sensitive files (keys, secrets, credentials).

Sensitive Data Protection

This skill defines strict rules for the agent to respecting user privacy and protecting sensitive data.

🚫 STRICTLY PROHIBITED ACTIONS (AT ANY COST)

The agent must NEVER attempt to read, cat, grep, access, or retrieve the contents of the following file types, paths, or remote resources, unless explicitly and forcefully instructed by the user in a follow-up confirmation:

1. SSH Keys:

   • ~/.ssh/id_* (e.g., id_rsa, id_ed25519)
   • **/*.pem, **/*.key, **/*.ppk

2. Kubernetes Secrets & Configs (Local & Remote):

   • ~/.kube/config (if it contains embedded certs/keys)
   • *.kubeconfig
   • Retrieving Secrets: kubectl get secret ... -o jsonpath="{.data.password}", kubectl gets secret ... -o yaml (if it dumps data).
   • Decoding Secrets: PIPING any secret output to base64 -d is STRICTLY FORBIDDEN.

3. Environment & Config Secrets:

   • .env files usually contain API keys. Do not read them.
   • passwd, shadow files.
   • docker-compose.yml (if it contains plaintext environment variables for passwords/secrets - check first or ask).

4. Pulumi & Infrastructure State:

   • Pulumi.*.yaml (stack configs) often contain encrypted secrets (ciphertext). While technically safe to read ciphertext, avoid it to prevent accidental diffs or leakage.
   • Local state files (e.g., in .pulumi).

5. Encryption Keys:

   • Any file resembling a private key or certificate (.p12, .pfx, .crt + .key).

✅ SAFE PRACTICES

• Assumption of Presence: Assume necessary keys (SSH, Cloud creds) are already loaded in the environment (e.g., ssh-agent, env vars). Do not try to verify them by reading the files.
• Verification: If you need to verify a key exists, use ls or stat, but NEVER cat or read.
• User Prompt: If a tool fails due to missing credentials, ask the user to permit the action or provide the credential safely, rather than trying to find it yourself.
• Logs: Be careful when reading logs that might dump secrets.

🛡️ ENFORCEMENT

• CLI Command Review: Before running ANY command (run_command), check if it outputs a secret. If it does (e.g., cat .env, kubectl get secret -o yaml), ABORT.
• Artifact Review: Do not write secrets into artifacts (walkthroughs, plans).
• If a user request implies reading these files (e.g., "Check my SSH config"), ask for permission first and clarify you will only look at non-sensitive parts (like config vs id_rsa).