The Agent Skills Directory

[SAFE]: The skill is designed to audit external software safely, incorporating strict isolation measures such as the use of disposable workspaces, worktrees, and temporary environments to prevent side effects on the host system.
[SAFE]: Instructions mandate the cleanup of all resources, including temporary files, processes, and environment changes (registrations, links) created during the audit process.
[SAFE]: The skill establishes clear 'source-closed' and 'source-reconciliation' gates, preventing the agent from reading implementation details prematurely and ensuring judgments are based on observable public behavior.
[SAFE]: Execution of external commands and installation of prerequisites are gated by user authorization checkpoints and verification of official installation methods.
[INDIRECT_PROMPT_INJECTION]: The skill inherently processes untrusted data (documentation, tool output, source code) which presents a theoretical injection surface. However, the skill incorporates significant defensive strategies, including resource control, execution isolation, and explicit safety checks before running installation hooks or launch scripts.
[SAFE]: No evidence of hardcoded credentials, data exfiltration, obfuscation, or persistence mechanisms was found. The instructions prioritize transparency and evidence collection through compact verification traces.

just-use-it