The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill provides explicit instructions and commands to extract administrative credentials, including Personal Access Tokens (PAT) and Private Key JWTs, from Kubernetes secrets (iam-admin-pat, iam-admin-jwt).\n- [COMMAND_EXECUTION]: The skill contains commands that use kubectl exec to perform direct SQL queries against the ZITADEL PostgreSQL database, specifically targeting application client IDs and client secrets (projections.apps7_oidc_configs).\n- [PRIVILEGE_ESCALATION]: The skill requires and demonstrates the use of high-privilege Kubernetes operations (reading secrets across namespaces and executing commands inside pods) to perform management tasks.\n- [DYNAMIC_EXECUTION]: A Python script for signing JWT assertions is dynamically generated and executed via a shell heredoc, using the jwt library.\n- [INDIRECT_PROMPT_INJECTION]: The skill has a broad attack surface for indirect prompt injection. It ingests untrusted data such as organization names, project names, and user profiles, which are then interpolated directly into shell-based curl commands without evident sanitization. This is documented across multiple files including the creation of organizations, projects, and users.

zitadel-api