The Agent Skills Directory

[Command Execution via Shell]: The skill uses jar, dexdump, and strings to inspect .ppmplugin archives. While these are standard development tools, the skill executes them via shell commands (bash/powershell) on files that could be crafted to exploit tool-specific vulnerabilities.
[Indirect Command Injection Surface]: The skill takes file paths as input and resolves them to shell commands (e.g., jar xf "<abs path>"). Although the skill structure suggests path resolution, passing unvalidated file paths to shell commands is a potential area to review for injection vulnerabilities.
[Prompt Injection Pattern]: The documentation mentions 'Override' and 'Policy: resolve, don't punt' in the context of tool detection logic. While used here for configuration, these terms can sometimes be associated with prompt injection patterns designed to bypass instructions.
[Dynamic Path Resolution]: The skill dynamically locates Android SDK build-tools paths to find dexdump. This behavior involves navigating the local filesystem based on environment variables and directory structures, which is a common pattern for build tools but warrants standard security hygiene.

audit-ppmplugin