The Agent Skills Directory

[COMMAND_EXECUTION]: The start_server.sh script automates dependency installation by executing pip3 install -r requirements.txt. This allows for the execution and installation of third-party code at runtime without explicit user verification of the packages.
[PROMPT_INJECTION]: The skill's web interface is vulnerable to Indirect Prompt Injection and Cross-Site Scripting (XSS). 1. Ingestion points: User answers entered into the text area on the web UI. 2. Boundary markers: Absent; user input is directly appended to the conversation history. 3. Capability inventory: File writing (report generation) in server.py and API calls to Anthropic and Groq in server_web.py. 4. Sanitization: Absent; the frontend in templates/index.html uses the .innerHTML property to render user-supplied content, allowing for arbitrary JavaScript execution in the browser context.
[DATA_EXPOSURE]: Multiple core files including server.py, SKILL.md, and prompt.md contain a hardcoded absolute file path (/Users/glebkalinin/Brains/brain) for storing automation reports. This discloses information about the developer's local environment and may cause functional errors or write files to unintended locations on systems with different directory structures.

automation-advisor