Credential Rotation After a Malicious Skill: Step-by-Step Playbook

If you suspect a malicious skill ran in your environment, assume credentials may be exposed.

This guide is a pragmatic playbook for what to do next.

Step 1: Contain first (stop the bleeding)

• Stop running skills immediately.
• Disconnect network if you suspect active exfiltration.
• Preserve evidence (logs, command history) before cleanup.

Use case: /use-cases/incident-response

Step 2: Identify the credential surface area

List what the skill could have accessed:

• .env files and config files
• process.env variables
• ~/.ssh, ~/.aws, cloud SDK config
• Git credentials / tokens

Step 3: Rotate in priority order

Rotate the credentials that create the largest blast radius first:

1. GitHub/GitLab tokens (repo access)
2. Cloud provider keys (AWS/GCP/Azure)
3. LLM provider keys (OpenAI/Anthropic/etc.)
4. Database credentials
5. Slack/email integrations

Step 4: Revoke sessions and audit access

• revoke tokens/sessions in provider dashboards
• check recent access logs if available
• look for new SSH keys or deploy keys

Step 5: Prevent repeat incidents

After rotation, lock things down:

• sandbox by default: /guides/sandbox-setup
• no-network default: /guides/network-allowlist
• verify skills: /verifier