The Agent Skills Directory

Command Execution and Input Interpolation: The skill constructs Google Cloud CLI (gcloud) commands by interpolating user-provided variables such as bucket names, project IDs, and service account emails. This is a primary function of the diagnostic skill, but it represents a potential surface for command injection if inputs contain shell metacharacters. The skill addresses this by mandating that commands be printed and explicitly approved by the user before execution.
Indirect Prompt Injection Surface: The skill processes untrusted data from users which is then used in tool calls.
Ingestion points: Target resource names (bucket, object), principal identities, and project IDs are ingested from user prompts in SKILL.md.
Boundary markers: No specific delimiters or markers are used to isolate user-supplied variables within the command strings.
Capability inventory: The skill utilizes gcloud and curl to perform read operations (logs, IAM policies) and modification operations (policy bindings, VM settings) as seen in references/403_troubleshooting.md.
Sanitization: No explicit sanitization or escaping of the user-provided variables is specified before interpolation into shell commands.
Remediation Safety Controls: A significant security feature is the 'CRITICAL SAFETY MANDATE' which prevents the autonomous execution of state-modifying commands. This requirement for human-in-the-loop confirmation (Y/N) is a best practice for managing high-impact infrastructure changes.
Use of Official Infrastructure: All troubleshooting steps and external links are directed toward official Google Cloud services and documentation, ensuring the diagnostic process remains within a trusted ecosystem.

google-cloud-storage-diagnostic