The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes repository data like workflow content and job names which could contain malicious instructions. This risk is effectively mitigated by the use of explicit boundary markers and specific instructions for subagents to ignore any commands found within those boundaries.
[COMMAND_EXECUTION]: The skill invokes shell commands like git and gh and internal scripts such as scan.py and report.py to perform its scanning and reporting functions. These operations are necessary for the skill's primary objective of auditing security and do not exhibit signs of abuse.
[PROMPT_INJECTION]: Instructions within references/prompts.md contain defensive language designed to protect subagents from being manipulated by untrusted content. These are proactive security measures rather than malicious attempts to bypass constraints.

ci-secure