SafeInstall CLI through version 0.10.1 contains a vulnerability where its agent guard shell parsing fails to detect some package-manager and registry-runner commands. This failure is caused by case-variant launcher names, leading file-descriptor redirections, and shell wrappers with options, which cause raw install commands to bypass the guard's approval decision. Remote project scaffolding commands can also avoid policy evaluation. Exploitation requires a coding agent to execute attacker-influenced crafted shell commands, potentially allowing malicious package installations or scaffolding to run with developer account permissions. The vulnerability affects only commands bypassing the guard interception; commands routed through the SafeInstall CLI remain protected. The issue is resolved in version 0.10.2 by normalizing launcher names, parsing redirections, handling wrapper options conservatively, and routing scaffolding commands through the approval path.

An attacker able to influence a coding agent's shell commands can bypass SafeInstall's intended deny or ask responses, causing unauthorized package installations or registry scaffolding commands to execute without policy checks or lifecycle script restrictions. This can compromise the confidentiality, integrity, and availability of local source code, credentials, and development resources by running code with the developer's permissions.

Upgrade to safeinstall-cli version 0.10.2 or later, which includes fixes for command normalization, redirection parsing, and approval routing. Until upgrading is possible, manually review all coding-agent shell commands and prevent the agent from invoking package managers or registry runners directly, as running an affected guard does not ensure raw package-manager execution is safe.