The Agent Skills Directory

COMMAND_EXECUTION (HIGH): The skill frequently generates shell commands using npm test and npm install. Workflows such as workflows/debugging/steps/step-01-init.md and workflows/running/steps/step-02-run-tests.md interpolate user-controlled placeholders directly into shell strings. Evidence: npm test -- -t "[exact test name]" and npm test -- [path/to/file.spec.ts]. Without sanitization, shell metacharacters in these inputs can lead to arbitrary command execution. \n- REMOTE_CODE_EXECUTION (HIGH): The command injection vulnerability provides a direct path to execute code on the system where the agent is running. Since unit tests often require significant permissions, this can lead to full system compromise. \n- EXTERNAL_DOWNLOADS (MEDIUM): The skill's setup workflow (workflows/setup/steps/step-02-dependencies.md) instructs the agent to install several Node.js packages via npm install. While some are standard, they are not from the strictly defined 'Trusted External Sources', making the installation of unverified dependencies a risk. \n- INDIRECT_PROMPT_INJECTION (HIGH): The skill is designed to ingest and analyze untrusted external content (source code and test descriptions) and possesses high-privilege capabilities like command execution. This combination represents a significant indirect injection surface as malicious code paths could be triggered or commands manipulated through processed data.

typescript-unit-testing