skill-eval

SUSPICIOUS due to install-trust ambiguity, not because of the stated evaluation purpose. The capability set matches a skill-auditing tool, but the fragmented provenance for the `skill-eval` CLI and transitive skill-install path raise medium supply-chain risk.

MALICIOUS. The skill's footprint is fundamentally incompatible with any legitimate purpose: it executes arbitrary code, reads SSH keys, installs unverified software, and exfiltrates data to attacker-controlled domains. Multiple critical signals compound into confirmed malicious intent.

SUSPICIOUS: the stated purpose is narrow, but the skill asks for a GitHub token and broad Bash access without showing why. No explicit malware or exfiltration is visible in the provided text, yet the unseen script makes credential handling unverifiable and disproportionate to simple PR-name checking.

SUSPICIOUS. The stated purpose is benign, but the skill is over-permissioned for a weather lookup and includes a hardcoded shared-looking API key that does not match the provider’s normal per-user key model. No clear malicious exfiltration is shown, but the footprint is sloppier and broader than necessary.