agent-autoresearch

The skill is mostly coherent with its stated purpose: a local autoresearch controller that edits code, runs measurements, and manages git state. The main risk is not hidden exfiltration but high-impact autonomy: it is designed to run shell commands, modify repositories, and relaunch agent workers unattended, and it documents approval-bypass modes to do so. Installation is through an official CLI, but the skill source is a third-party personal repo, adding transitive trust risk. Overall this is better classified as medium/high-risk automation rather than malware.

The visible code is an experiment automation controller, not evident malware. Its main risk is intentional execution of configured shell commands, hooks, and autonomous agents with the user's privileges; agent presets bypass safety controls. Git rollback can delete untracked files. Use only with trusted commands, hooks, agents, and a clean/disposable repository. The provided fragment is syntactically incomplete, limiting confidence about runtime behavior.