AI-Written GitHub Actions Workflows Require Mandatory Permission Checks

AI-generated GitHub Actions YAML files can pass syntax validation while still containing dangerous permission configurations, such as write access or exposed secrets. Developers are advised to treat any model-generated workflow as an untrusted patch rather than a verified configuration. A two-layer approach is recommended: running actionlint for syntax validation, followed by a lightweight Python script that rejects workflows requesting write permissions or referencing secrets. The Python gate catches common failures like workflow-wide write-all permissions or steps that expose secret values, even when actionlint raises no errors. However, static checks have clear limits and cannot detect runtime data exfiltration, malicious shell commands, or indirect secret access via pull_request_target triggers.