The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes third-party application source code and container images, which constitutes a surface for indirect prompt injection if the ingested content contains malicious instructions.
Ingestion points: The skill reads from local source paths (targetPath) and remote container registries (imageUrl).
Boundary markers: While it refers to security protocols for data protection, there are no specific prompt delimiters defined to isolate ingested code from agent instructions.
Capability inventory: The skill utilizes manageCloudRun for service deployment, configuration updates, traffic management, and resource deletion.
Sanitization: The instructions explicitly mandate the use of the sensitive-runtime-data-protection.md protocol to prevent credential leakage.
[EXTERNAL_DOWNLOADS]: The skill facilitates the download and deployment of container images from well-known registries like Docker Hub and the vendor's internal Container Registry (CCR).
[COMMAND_EXECUTION]: The manageCloudRun tool is used to execute administrative actions on the CloudBase platform, including building source code, deploying versions, and managing canary releases.
[CREDENTIALS_UNSAFE]: The documentation mentions handling sensitive connection strings (e.g., DATABASE_URL) and platform credentials (x-cloudbase-context). However, it provides defensive guidance against echoing these values in responses or logs, representing a safe practice for credential management.

cloudrun-development