The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs extensive discovery and management of OCI resources using shell and Python scripts to run OCI CLI and kubectl commands. Mutating actions are secured via a dedicated confirmation gate (confirm_gate.sh) that uses nonces to prevent command replay and ensures the user approves the exact command string.
[EXTERNAL_DOWNLOADS]: The install_fn_linux.sh script downloads the Fn CLI binary from the official Fn Project GitHub repository. This is a trusted source for OCI Functions tooling. The script implements robust security checks, including SHA-256 digest verification and version validation, before installing the binary.
[PRIVILEGE_ESCALATION]: Operational scripts such as node-doctor-run.sh execute high-privilege diagnostics by creating debug pods that use chroot /host and sudo. These capabilities are essential for node-level troubleshooting and are explicitly gated by instructions requiring user confirmation for every execution.
[INDIRECT_PROMPT_INJECTION]: The skill's troubleshooting components ingest untrusted data from OCI and Kubernetes environments.
Ingestion points: System state collected via kubectl get/describe and oci CLI outputs (e.g., in oke-pod-network-check.sh).
Boundary markers: Present; the skill uses automated summarization and normalization logic to isolate system signals from the agent's context.
Capability inventory: High; includes the ability to mutate cloud infrastructure and cluster resources upon user approval.
Sanitization: Scripts such as oke-object-correlator.sh use structured JSON parsing and regex filters to extract specific metadata while ignoring irrelevant or potentially malicious text in command outputs.

oci