The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The shell script scripts/run-taze.sh contains a command injection vulnerability where the include_flag variable is expanded without quotes in the execution path (taze major $recursive $include_flag --include-locked). Shellcheck is explicitly disabled for this line to allow word splitting, which enables arbitrary command execution if a package name contains metacharacters such as ;, &, or |.
[REMOTE_CODE_EXECUTION]: The SKILL.md file instructs the agent to construct shell commands by directly interpolating package names (e.g., taze major --write --include <pkg1>,<pkg2>,<pkg3>). This creates a vulnerability where an attacker-controlled package.json file containing malicious package names could trigger arbitrary code execution when the agent attempts to apply updates.
[COMMAND_EXECUTION]: The skill performs shell command execution to run its internal script, the taze dependency tool, and the ni package manager wrapper.
[EXTERNAL_DOWNLOADS]: The skill recommends and utilizes external CLI tools taze and ni which are fetched from the npm registry.
[DATA_EXFILTRATION]: The identified command injection vulnerabilities could be leveraged to exfiltrate project files, environment variables, or other sensitive data from the execution environment.

bump-deps