agent-manager

SUSPICIOUS: the core tmux/Python lifecycle-management behavior is coherent with the stated purpose, but the footprint is not fully proportionate. The skill encourages launching agents with --dangerously-skip-permissions, installs through an unpinned third-party npm CLI before falling back to same-org git clone, and supports persistent cron jobs plus skill injection into managed agents. No clear credential theft or malicious exfiltration is shown, but the combination of safety-bypass guidance and third-party installer trust issues makes this higher-risk than a normal local orchestration skill.

The script appears to be a conventional installer for a GitHub-hosted skill and contains no direct evidence of malware or malicious intent. Its security risk is primarily supply-chain related: it runs an unpinned npm installer through npx, clones an unpinned GitHub repository, copies downloaded files into executable skill directories, and executes the downloaded main.py during verification. Use is safer with pinned versions or commits, integrity verification, and review of the fetched repository and openskills package.