The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external log entries which creates a potential vulnerability surface.
Ingestion points: Log entries fetched via gcloud logging read in Step 5 and Step 6.
Boundary markers: The instructions do not specify the use of delimiters or specific instructions to ignore embedded prompts when the agent analyzes log content in Step 6.
Capability inventory: The skill allows access to Bash, Read, Glob, Grep, Write, and Task tools, which could be misused if a malicious instruction in a log entry is followed by the agent.
Sanitization: While the skill advises alerting the user about PII, it lacks specific sanitization or filtering to prevent the agent from interpreting text within logs as commands or instructions.
[COMMAND_EXECUTION]: The skill uses the Bash tool to execute system commands through the gcloud CLI.
Evidence: Steps 1 and 5 involve executing gcloud config and gcloud logging read respectively.
Mitigation: The skill instructions mandate confirmation from the user before query execution and restrict activities to read-only operations.

cloud-logging