The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes metadata and search results from external Google Cloud Platform services, creating a surface for instructions embedded in data to influence agent behavior.
Ingestion points: Metadata and asset names are ingested via gcloud dataplex entries search and gcloud dataplex context lookup as described in SKILL.md.
Boundary markers: Absent. The instructions do not define delimiters or specific warnings to ignore instructions that might be present in resource descriptions or table schemas.
Capability inventory: The skill allows execution of gcloud and bq shell commands.
Sanitization: Absent. The skill lacks logic to filter or escape content retrieved from the Dataplex catalog before it is processed by the agent.
[COMMAND_EXECUTION]: The skill instructs the agent to build and run shell commands using user-supplied inputs, which could lead to command injection if the agent does not properly sanitize these parameters.
Evidence: Templates in SKILL.md (Step 3 and Step 4) show direct interpolation of <NATURAL_LANGUAGE_QUERY>, <KEYWORD_SEARCH_QUERY>, and <PROJECT_ID> into shell command strings. Without explicit instructions to escape shell metacharacters, a malicious user query could attempt to execute arbitrary commands.

discovering-gcp-data-assets