code-quality-audit

SUSPICIOUS. The skill's capabilities largely match its stated code-quality/security-audit purpose, and the referenced tools/endpoints are mostly official. The main concerns are supply-chain hygiene (curl|sh from mutable branches, unpinned installs), hook-driven execution in untrusted repos, and offensive-style DAST guidance that can affect external systems. No confirmed credential theft, covert exfiltration, or pre-execution malware behavior is present.

The fragment is primarily a toolchain configuration/catalog and does not itself contain an obvious malware payload or obfuscated code. It does document high-risk installer behavior: executing mutable remote Trivy and Gitleaks installation scripts through a shell and writing machine-wide binaries. That is a significant supply-chain security weakness and should be replaced with pinned, verified releases or a trusted package manager. Composer plugins and git-hook installers also warrant explicit consent and review. Findings are limited by the truncated context and referenced files not being shown.

No explicit malware, credential theft, suspicious network exfiltration, persistence, or destructive behavior is present. The workflow is a legitimate PR quality-review workflow, but it has a significant GitHub Actions security boundary issue: it executes pull-request-controlled Composer and tool code while granting pull-requests: write permission. This can allow a malicious PR or compromised dependency to abuse the token or modify PR comments. Use least-privilege permissions, avoid write-capable tokens when executing untrusted code, isolate untrusted analysis, and invoke file paths with safe quoting/argument handling. The code itself is not intentionally obfuscated.

No direct malware or deliberately malicious payload is evident. The integration is security-sensitive because it checks out attacker-controlled PR code and runs cloud audit routines, while optional unrestricted branch pushes and comment-based asynchronous gating can increase impact if the routine, connector, or polling logic is compromised. Use a read-only token and least-privilege GitHub permissions, isolate or sandbox analysis, pin verdicts to the exact commit and session, reject stale or unauthorized comments, and avoid unrestricted branch pushes unless required.

The fragment contains no direct malware or executable malicious behavior. It documents a high-impact prompt-injection design in which repository authors control instructions supplied to automated review agents. This creates a supply-chain and review-integrity risk if untrusted users can modify REVIEW.md, particularly because they could suppress or downgrade security findings. The risk is conditional on the surrounding service trusting repository-controlled content and on attacker write access to the file.

The fragment is benign setup documentation, but it contains insecure installation practices: unpinned dependencies and direct curl-to-shell execution of remote installer scripts. These practices create supply-chain and host-code-execution risk, particularly when run with elevated privileges. No evidence of intentional malware or data theft appears in the provided fragment; review the referenced install script and pin or verify all tool versions and installer checksums before use.