The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The scaffolded project is vulnerable to Cross-Site Scripting (XSS).
Ingestion points: Untrusted data enters the system through the API via the POST /api/items endpoint defined in app.py and is retrieved via GET /api/items.
Boundary markers: None present in the generated code templates to prevent command injection or data confusion.
Capability inventory: The frontend script static/js/main.js uses the innerHTML property to render data from the API.
Sanitization: The renderTable function in main.js does not perform any escaping or sanitization of the item.name field, allowing an attacker to execute arbitrary JavaScript in the context of the user's browser.
[DYNAMIC_EXECUTION]: The scaffolding script constructs file paths from user-provided arguments without validation.
Evidence: scripts/scaffold.py uses os.path.join on the name and dir arguments provided via the command line to determine where to create the project skeleton.
Risk: This pattern could allow for path traversal if an attacker influences the arguments, though it is typical for local scaffolding tools.
[EXTERNAL_DOWNLOADS]: The skill scaffolds projects that depend on external resources from well-known CDN providers.
The BASE_HTML template in scripts/scaffold.py includes script and style tags referencing cdn.jsdelivr.net for Bootstrap and ECharts libraries.
[COMMAND_EXECUTION]: The skill requires the user to execute local scripts and run a web server.
The instructions in SKILL.md guide the user to run python3 scripts/scaffold.py and start the Flask application with python app.py.

web-project-builder