The Agent Skills Directory

[SAFE]: The skill functions as a security-centric framework for CI/CD configuration. It provides templates and instructions that prioritize hardening, such as explicitly instructing the agent to pin actions to immutable 40-character SHAs to prevent supply-chain attacks.
[COMMAND_EXECUTION]: The skill mentions and utilizes established development and security tools (e.g., go, docker, cosign, syft, trivy, actionlint, zizmor, goreleaser) for their standard operational purposes. The instructions include security mitigations for command execution, such as using environment variables instead of shell interpolation to prevent command injection.
[EXTERNAL_DOWNLOADS]: The skill references official GitHub Actions and binaries from reputable and well-known organizations including Sigstore, Anchore, Aqua Security, and the GitHub Actions official repository. These references are used for standard pipeline tasks like signing, SBOM generation, and security scanning.
[DATA_EXFILTRATION]: No patterns of unauthorized data access or exfiltration were detected. The skill includes explicit guidance on protecting credentials, such as setting persist-credentials: false on checkouts and using OIDC for identity verification to avoid long-lived secrets.
[PROMPT_INJECTION]: The skill contains no instructions intended to bypass safety guardrails, override system instructions, or extract system prompts. All instructional content is focused on the intended goal of CI/CD auditing and setup.
[SAFE]: While the skill involves a surface for processing untrusted data (reading a repository's existing CI/CD configuration to audit or reshape it), it provides clear defensive instructions to mitigate risks. The agent has file-write capabilities to update configurations, but the skill mandates sanitization practices (e.g., avoiding ${{ }} interpolation in shell blocks) and human-in-the-loop verification for significant changes.

cicd-fy