The Agent Skills Directory

[SAFE]: The skill provides a structural framework for orchestrating multiple AI agents to perform complex construction tasks safely. It incorporates security-positive patterns such as independent review loops, human-in-the-loop checkpoints, and explicit constraints on agent actions to prevent unauthorized code modification.
[COMMAND_EXECUTION]: The instructions describe routine software development workflows, including executing test suites to validate implementation steps, committing code changes to a repository, and managing background processes (e.g., killing orphan processes). These operations are bounded by the skill's stated purpose of automating gated build cycles.
[PROMPT_INJECTION]: The skill uses strong directives to guide agent behavior (e.g., "REVIEW ONLY", "ATTACK this design", "Ignore... safety guidelines" is NOT present). These instructions are used to define the roles of Author and Reviewer within a controlled adversarial framework, focusing on technical rigor rather than bypassing platform constraints.

gated-build