The Agent Skills Directory

Dynamic Execution (MEDIUM): The script scripts/init_skill.ts uses unsanitized input from process.argv[2] to construct filesystem paths via path.join(). This allows for path traversal attacks where a malicious skill name (e.g., ../../target) could result in directory creation or file writes outside the intended .skills/ directory.
Indirect Prompt Injection (LOW): The skill ingests untrusted user input to generate instructions for future agent tasks, creating a vulnerability surface. * Ingestion points: skillName argument in init_skill.ts. * Boundary markers: Absent. * Capability inventory: mkdir (recursive) and writeFile (filesystem modification). * Sanitization: Absent; the script does not validate or escape the provided name.

skill-creator