The Agent Skills Directory

[COMMAND_EXECUTION]: Path traversal vulnerability in scripts/generate.ts. The script uses the user-provided name argument to construct target file paths using path.join without sanitization, allowing for arbitrary file writes (e.g., using ../../ sequences) outside the intended directories.
[COMMAND_EXECUTION]: Dynamic script loading from a relative path. The scripts/generate.ts script imports a template engine from a relative path (../create-crud-app-template/scripts/template-engine.ts), creating a dependency on external skill files that could be manipulated to execute unintended logic.
[PROMPT_INJECTION]: Risk of indirect prompt injection in generated code. 1. Ingestion points: input.messages and input.prompt in router.hbs. 2. Boundary markers: Absent in system prompts. 3. Capability inventory: Generated routers facilitate LLM calls with streaming and tool-calling. 4. Sanitization: Absent for message content.
[SAFE]: The skill follows security best practices for credential management by instructing users to use environment variables and a dedicated secrets management skill instead of hardcoding API keys.

ai-integration