The Agent Skills Directory

[SAFE]: The skill provides well-structured templates for Docker, Kubernetes, and Google Cloud Run deployment. It follows security best practices by utilizing environment-specific secrets and least-privileged runtime identities.
[CREDENTIALS_UNSAFE]: The documentation uses truncated placeholder strings (e.g., sk-ant-api03-...) to illustrate command syntax. The skill instructions explicitly prohibit hardcoding real credentials in manifests, images, or command history, recommending external secret managers instead.
[INDIRECT_PROMPT_INJECTION]: The skill includes instructions for generating deployment receipts based on monitoring data, creating a potential data processing surface.
Ingestion points: Monitoring logs and SLO metrics as described in SKILL.md.
Boundary markers: Explicit directives to verify that logs and probes cannot expose prompts or responses (sensitive_content_logged=0).
Capability inventory: Access to deployment tools via Bash (including gcloud, kubectl, and docker).
Sanitization: A mandatory requirement to exclude API keys, prompts, responses, and customer identifiers from the final deployment output.
[SAFE]: The skill references official developer resources and service status pages from Anthropic to assist in deployment and maintenance workflows.

anth-deploy-integration