Instructions and permissions
Read what the skill asks the agent to do. Check requests for secrets, unrelated file access, hidden actions, or instructions that conflict with your task.
A useful description is a starting point. Review the instructions and supporting files before trusting a community skill.
We inspect the text of available SKILL.md files for a small set of patterns including remote shell execution, reverse shells, instruction overrides, sensitive credential paths, and destructive root deletion. Code blocks are checked too. We do not execute the skill. Results include the scan date, scanner version, and a content hash; the hash identifies the file inspected and is not a repository commit.
Results are partial: the scan does not inspect supporting files, scripts, remote content, or behavior at runtime. Patterns may be benign examples or may miss harmful instructions. A result with no findings is not a safety guarantee. Sources that cannot be fetched are marked unavailable; listings not yet processed are marked not scanned.
Read what the skill asks the agent to do. Check requests for secrets, unrelated file access, hidden actions, or instructions that conflict with your task.
Review executable files, package requirements, network destinations, and setup commands. Understand their effects before running them.
Check the author, repository history, and exact folder. Record a revision and compare updates before replacing a local installation.
Check reuse terms, required tools, and operating-system support. A matching SKILL.md format alone does not prove the workflow works in your agent.
Reachability, file structure, security findings, and real-world compatibility are different checks. A link that loads does not prove a skill is safe or works correctly.
Read the installation checklist →