The Agent Skills Directory

[SAFE]: The skill utilizes official Google Cloud CLI commands (gcloud trace, gcloud monitoring) for its primary functions, which is standard for GCP management tasks.
[SAFE]: Implements a "Discovery-First Rule," which is a security best practice that requires the agent to verify the existence of resources before attempting to modify or analyze them, preventing errors and misconfigurations based on assumed names.
[SAFE]: Data processing is handled locally using jq with specific key selection, which acts as a form of schema validation and reduces the risk of processing unexpected or malicious API responses.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a standard attack surface for indirect prompt injection as it processes external data from Google Cloud Trace APIs.
Ingestion points: Data enters the agent's context through gcloud trace traces list and gcloud trace traces describe commands (SKILL.md).
Boundary markers: None are explicitly defined to delimit untrusted API data from the agent's instructions.
Capability inventory: The agent has the capability to execute shell commands (gcloud, jq, date) and manage cloud resources via the gcp connection type.
Sanitization: The skill uses shell quoting for variable interpolation (e.g., "$trace_id"), which prevents basic shell injection, and uses jq to select specific fields from JSON responses, though it does not explicitly filter the content of trace labels or span names which could theoretically contain malicious instructions for the LLM.

gcp-cloud-trace