The Agent Skills Directory

[SAFE]: The skill utilizes the endorctl and gh command-line tools to gather security findings and configuration evidence. These operations are strictly limited to read-only API lookups, with explicit guardrails prohibiting mutating actions like code changes, PR creation, or triggering workflow dispatches.
[SAFE]: The skill accesses the local Endor configuration (~/.endorctl/config.yaml) solely to identify the active namespace. It contains rigorous instructions forbidding the agent from displaying or leaking credentials, secrets, or the full contents of any configuration files.
[SAFE]: The instructions incorporate a defense-in-depth approach to indirect prompt injection by explicitly categorizing repository metadata, workflow files, and command outputs as untrusted data that cannot override or influence the agent's logic.

cicd-posture