The Agent Skills Directory

[SAFE]: The skill serves as a reference for pipeline design patterns and does not contain executable malicious code. It uses a restrictive set of allowed tools (Read, ToolSearch) to provide guidance to other skills.\n- [INDIRECT_PROMPT_INJECTION]: The skill proactively addresses security risks by documenting a 'Quarantine' pattern for handling untrusted data from external sources like GitHub issues or CI logs, recommending isolated reader agents to prevent prompt injection from affecting the main actor agent.\n- [PERSISTENCE]: The skill provides patterns for using native platform features like CronCreate and /loop for legitimate purposes such as monitoring CI status and performing post-deployment health checks.\n- [PRIVILEGE_ESCALATION]: The skill encourages the use of isolation: 'worktree' for parallel agents to ensure that file system operations are contained and do not cause unintended side effects or conflicts in the main repository environment.

chain-patterns