The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to download and execute a shell script directly from a remote GitHub repository using a piped command pattern.
Evidence: curl -sSfL https://github.com/cloudnative-pg/cloudnative-pg/raw/main/hack/install-cnpg-plugin.sh | sudo sh -s -- -b /usr/local/bin found in SKILL.md and references/kubectl-cnpg.md.
[COMMAND_EXECUTION]: The skill uses elevated privileges via sudo to perform system-wide installation of the kubectl-cnpg plugin.
Evidence: sudo sh -s -- -b /usr/local/bin in SKILL.md.
[EXTERNAL_DOWNLOADS]: The skill downloads external configuration and binary installation scripts from domains associated with the project.
Evidence: https://cloudnative-pg.github.io/charts and https://github.com/cloudnative-pg/cloudnative-pg in SKILL.md.
[CREDENTIALS_UNSAFE]: The skill includes hardcoded default credentials in its documentation examples for S3-compatible storage (MinIO).
Evidence: ACCESS_KEY_ID: "minioadmin" and ACCESS_SECRET_KEY: "minioadmin" in references/backup-s3.md.
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it is designed to ingest and process outputs from the Kubernetes environment (logs, status commands) which could potentially contain attacker-controlled data.
Evidence: Commands like kubectl cnpg status and kubectl logs in references/troubleshooting.md ingest external data without explicit sanitization or boundary markers.

cloudnative-pg