The Agent Skills Directory

[COMMAND_EXECUTION]: Static analysis identified a potentially destructive command rm -rf / within the SKILL.md file. Upon manual review, this command is used as a technical example to illustrate a 'poisoned pipeline execution' attack via expression injection. It is an educational example of a vulnerability to be avoided, not an instruction for the agent to execute.
[EXTERNAL_DOWNLOADS]: The skill references several external security tools and documentation sources, including zizmor, actionlint, OpenSSF Scorecard, and cosign. All referenced URLs and tools originate from trusted organizations or well-able well-known service repositories (e.g., GitHub, GitLab, Google, Amazon, OpenSSF, and the Center for Internet Security). These references are used to provide the agent with authoritative context for performing CI/CD security reviews.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data in the form of CI/CD configuration files (e.g., .github/workflows/*.yml) provided by users or from repository forks.
Ingestion points: Reads workflow definitions and configuration files (SKILL.md).
Boundary markers: The skill does not explicitly instruct the agent to use boundary markers or delimiters when processing these files.
Capability inventory: The skill primarily performs analysis and generates a report. It suggests hand-offs to other specialized skills for tasks like SAST or secret scanning.
Sanitization: There are no explicit instructions for the agent to sanitize the content of the files before processing.
[PROMPT_INJECTION]: No attempts to bypass safety filters or override system instructions were detected. The skill uses standard instructional language to define its scope and methodology.

cicd-hardening