forge-security-review

The skill’s purpose and stated review workflow are internally coherent for a Forge security-audit skill, but it still carries elevated risk because it equips an AI agent with offensive/security-scanning behavior, processes untrusted code, and executes an unspecified local analysis script that may call third-party services. This looks more like a high-risk but plausibly legitimate security-review skill than malware.

This Bash wrapper is not overtly obfuscated and shows no direct evidence of malware behaviors (e.g., exfiltration, backdoors, credential theft) within the wrapper itself. However, it creates a meaningful supply-chain execution risk: it downloads and executes a platform-specific executable from a remote GitHub Releases URL without integrity/authenticity verification or version pinning, and it extracts untrusted archives and selects the first matching fsrt/fsrt.exe for installation. The wrapper should be mitigated by pinning an immutable artifact/version and verifying a checksum/signature prior to execution; archive contents should also be hardened/validated before installing/executing.

This script is primarily a legitimate-looking bootstrapper for a security-review tool, but it performs a high-impact supply-chain operation: it downloads and executes an external binary from a fixed remote release without any integrity/authenticity verification, and it selects the executable purely by filename from extracted archive contents. If the remote release asset (or its upstream publishing pipeline) is compromised, the script will likely execute attacker-controlled code. No explicit malicious payload logic is visible in the script itself, but the trust model is sufficiently weak that the package should be reviewed/pinned (hash/signature) and the extracted file should be validated before execution.