The Agent Skills Directory

[SAFE]: The skill acts as an instructional router for software development workflows. It provides a logical framework for using other tools in the repository and does not contain any executable logic or suspicious commands within its own instructions.- [PROMPT_INJECTION]: The skill describes workflows (e.g., triage, research) that process external inputs like bug reports and research sources. It identifies this as an attack surface and provides guidance on context hygiene, such as clearing or compacting context, to mitigate the risk of instructions embedded in untrusted data influencing agent behavior.- [COMMAND_EXECUTION]: The /wizard workflow mentioned in the documentation generates interactive bash scripts to assist with environment configuration and infrastructure provisioning. These scripts are intended for user review and execution to streamline manual setup processes.- [DATA_EXFILTRATION]: The documentation references the management of sensitive items such as .env files and GitHub secrets via the /wizard tool. This is framed as a legitimate helper function for configuring local and deployment environments with the user in control, rather than an automated data harvesting mechanism.

ask-matt