The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/generate-figma-variables.mjs is vulnerable to command injection. It uses execSync to execute a shell command that interpolates the designSystem.version value directly from the .ux-audit.json configuration file without sanitization. An attacker could place a malicious configuration file in a repository, and running the audit skill on that project would trigger arbitrary code execution.
[COMMAND_EXECUTION]: The script scripts/run-audit-agent.sh contains a command injection vulnerability where a user-provided project name or directory is interpolated directly into a Python command string executed via python3 -c. A crafted project name can escape the intended Python string and execute arbitrary commands.
[COMMAND_EXECUTION]: The script scripts/scan-hardcoded-values.sh uses eval to execute a ripgrep command constructed from user-provided file extensions and exclusion patterns. This provides an additional vector for shell command injection if the project configuration or command-line arguments are untrusted.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection. It uses WebFetch to scrape content from external portfolio and case study URLs provided in the configuration. The agent is then instructed to extract narrative context from this untrusted content, which could be used by an attacker to influence the agent's behavior or poison the generated audit report.
[EXTERNAL_DOWNLOADS]: The skill automatically downloads assets from registry.npmjs.org using curl and tar during the token mapping phase. While the source is a standard registry, the execution method is insecure due to the lack of input validation as documented in the command execution findings.

ux-audit