The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides numerous command-line templates for the gcloud, bq, and python tools to manage Dataflow jobs, network configurations, and BigQuery tables. These are standard administrative actions for the target platform.
[EXTERNAL_DOWNLOADS]: Fetches pipeline templates, metadata, and JSON schemas from official, public Google Cloud Storage buckets (gs://dataflow-templates) and references the authoritative GoogleCloudPlatform GitHub repository. These sources are considered well-known and trusted for this vendor's ecosystem.
[INDIRECT_PROMPT_INJECTION]: The skill analyzes external data including Dataflow job messages, execution events, and Cloud Logging entries. This represents an ingestion surface for potentially untrusted data that could contain malicious instructions.
Ingestion points: dataflow_diagnostics_reference.md (fetches job messages via the Dataflow REST API and logs via gcloud logging read).
Boundary markers: The SKILL.md file defines a "Universal Execution Workflow" which mandates a pre-launch confirmation step where the agent must present the entire command for user review before execution.
Capability inventory: The skill has capabilities to execute shell commands (gcloud, bq), make network requests to Google APIs (curl), and write local files (JSON schemas).
Sanitization: There is no explicit mention of sanitizing or escaping the content retrieved from logs or job messages before the agent processes it.
[DYNAMIC_EXECUTION]: The skill dynamically constructs shell commands and JSON schema files based on project metadata, user inputs, and template requirements retrieved during execution.

gcp-dataflow