The Agent Skills Directory

[SAFE]: No executable code or scripts were detected in the provided files. The skill consists entirely of markdown-based prompts, YAML evaluation cases, and configuration metadata.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-supplied data such as release scopes, change lists, and system logs. It mitigates injection risks by instructing the agent to audit input completeness, separate facts from assumptions, and link conclusions directly to evidence. Mandatory evidence chain: ingestion occurs through user-provided materials specified in prompts/production-verification.md; boundary markers are enforced by separating facts, assumptions, and gaps; the skill definition contains no tool-calling or subprocess capabilities; sanitization is achieved through rigorous evidence verification requirements.
[SAFE]: The core constraints and execution rules explicitly mandate security-conscious behaviors, including performing read-only checks before writes, using masked data for privacy, and defining clear stop/escalation conditions for production environments.

production-verification