What problem does it solve? Opening a cloned or downloaded project in an IDE, agent, or package manager can silently execute attacker-controlled code through hooks, MCP servers, lifecycle scripts, or compromised dependencies. This Skill performs a read-only static audit to classify whether a repository is weaponized before you trust, install, build, or run it. ## Core Features & Use Cases - Bounded hostile-repository audit: Reviews agent instructions, Git hooks, IDE tasks, package lifecycle scripts, and CI paths for concealed execution, prompt injection, credential theft, and persistence. - Dependency compromise screening: Parses existing lockfiles as inert data and batch-checks public package identities against OSV and GitHub malware advisories without invoking package managers. - Strict safety contract: Never executes target code, never writes to the target, and issues a verdict of NO_HOSTILE_INDICATORS, SUSPICIOUS, HOSTILE, or INCONCLUSIVE with evidence-backed findings. - Use Case: Before opening a GitHub project you just cloned, run this audit to confirm its hooks, configs, and dependencies contain no credible attack chain against your workstation or coding agent. ## Quick Start Audit the cloned repository at ~/Downloads/some-project for signs it is weaponized against developers or coding agents before I open it.
