The Agent Skills Directory

[COMMAND_EXECUTION]: The skill contains instructions to download a shell script from a remote repository and pipe it directly to the shell for execution. This pattern bypasses local verification of the script content before execution.
Evidence: curl -sSfL https://github.com/cloudnative-pg/cloudnative-pg/raw/main/hack/install-cnpg-plugin.sh | sudo sh -s -- -b /usr/local/bin in SKILL.md and references/kubectl-cnpg.md.
[PRIVILEGE_ESCALATION]: The plugin installation command uses sudo to gain administrative privileges for installing the binary into /usr/local/bin.
Evidence: ... | sudo sh -s -- -b /usr/local/bin in SKILL.md.
[CREDENTIALS_UNSAFE]: The skill includes hardcoded default credentials for a MinIO backup storage service in its configuration examples.
Evidence: ACCESS_KEY_ID: "minioadmin" and ACCESS_SECRET_KEY: "minioadmin" in references/backup-s3.md.
[EXTERNAL_DOWNLOADS]: The skill fetches resources and installation scripts from external sources, including GitHub and Helm repositories.
Sources: https://cloudnative-pg.github.io/charts, https://github.com/cloudnative-pg/cloudnative-pg/raw/main/hack/install-cnpg-plugin.sh, and container images from ghcr.io.
[INDIRECT_PROMPT_INJECTION]: The skill processes and generates Kubernetes manifests, PostgreSQL configurations, and diagnostic outputs, which serves as an attack surface for indirect prompt injection from untrusted cluster data.
Ingestion points: Reads cluster status, pod logs, and event descriptions via kubectl and cnpg plugin commands in references/troubleshooting.md and references/kubectl-cnpg.md.
Boundary markers: None explicitly defined in the provided instruction snippets to delimit external data from agent instructions.
Capability inventory: File writes (YAML manifests), network operations (Helm, Curl), and shell command execution (Kubectl, Psql) across all documentation files.
Sanitization: No specific sanitization or validation of the ingested cluster data or logs is mentioned before processing.

cloudnative-pg