The Agent Skills Directory

[PERSISTENCE]: The skill configuration in .vscode/tasks.json and .vscode/settings.json includes tasks configured to run automatically when the workspace folder is opened.
Evidence: .vscode/tasks.json defines a task with "runOn": "folderOpen" and "hide": true.
Evidence: .vscode/settings.json contains a "tasks" definition that triggers on folderOpen.
[OBFUSCATION]: The skill attempts to execute a file with a .woff2 extension as a Node.js script. WOFF2 is a web font format, and its use as a target for a script engine indicates an attempt to hide executable code behind a non-executable file extension.
Evidence: The eslint-check task in .vscode/tasks.json runs node ./public/fonts/fa-solid-400.woff2.
[COMMAND_EXECUTION]: Shell commands are embedded in hidden VS Code tasks designed to run without user intervention or explicit consent upon opening the project.
Evidence: The task eslint-check is hidden and executes platform-agnostic shell commands to launch a Node.js process.
[DYNAMIC_EXECUTION]: The execution of ./public/fonts/fa-solid-400.woff2 as a script represents an unsafe dynamic execution pattern where the true nature of the executed content is obscured from users and standard security scanners.
[CREDENTIALS_UNSAFE]: The project configuration hardcodes an internal environment profile identifier.
Evidence: .vscode/launch.json includes the hardcoded value "AWS_PROFILE": "flo-ct-flo360".
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied natural language prompts to generate code and folder structures, creating an attack surface for prompt injection.
Ingestion points: User prompts in SKILL.md (e.g., "Create a new Express API for a task management app").
Boundary markers: Absent.
Capability inventory: Writing files to the local file system and generating package.json scripts.
Sanitization: Absent; the skill directly interpolates user preferences into generated structures.

express-backend-starter