ci-agent-hardening

Audit GitHub Actions workflows for security vulnerabilities by running a script that scans all .github/workflows YAML files to detect risky patterns like pull_request_target forks, unsanitized run: blocks, and missing explicit permissions.

pull_request_target is a GitHub Actions trigger that grants token access to forked code, creating a security risk. It enables credential leakage or unauthorized code execution if workflows run unsanitized commands from untrusted pull requests.

Secure AI-powered CI pipelines against prompt injection by checking for misconfigurations, enforcing CODEOWNERS protections, limiting tool access, and applying safe CLAUDE.md guidance to guard AI agents across build, test, and release stages.

Yes, this GitHub Actions security audit checks for missing explicit permissions, unilateral PAT usage, cache implications, and CODEOWNERS protections, providing concrete remediation guidance to prevent credential exposure.

Yes, you can use this workflow audit for enterprise CI pipelines, open-source project auditing, and hardening automation workflows before public releases across multiple repositories to ensure provenance best practices.

Limitations of securing CI workflows against prompt injection include the need to manually apply concrete remediation guidance and the requirement to continuously monitor AI agent integration, slash command controls, and tool access configurations.