The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the execution of the answerrank CLI tool (e.g., answerrank record, answerrank report) and system utilities to read the clipboard, such as pbpaste (macOS), xclip (Linux), and Get-Clipboard (Windows). These commands are used to store and process the data retrieved from browser sessions.
[INDIRECT_PROMPT_INJECTION]: The skill processes questions from a local directory and instructs the agent to paste them directly into authenticated AI chat sessions without sanitization.
Ingestion points: Data is ingested via the answerrank questions command from local files like queries.json (SKILL.md).
Boundary markers: No boundary markers or delimiters are specified; the agent is instructed to paste the question "exactly as written."
Capability inventory: The agent has access to shell execution (answerrank), browser automation (Chrome DevTools/Playwright), clipboard access tools, and local file writing (/tmp/answer.txt).
Sanitization: No sanitization or filtering of the question content is mentioned before it is input into the browser.
[DATA_EXFILTRATION]: The skill accesses sensitive data from the user's authenticated web sessions (ChatGPT, Claude, Gemini, etc.), including potentially private history or memory, and moves this data through the system clipboard to local files. While this is the intended purpose, it represents a sensitive data flow involving authenticated sessions.
[DYNAMIC_EXECUTION]: The instructions direct the agent to generate and execute JavaScript snippets within the browser to manipulate the DOM (creating a <textarea>) and trigger clipboard events. This is a routine use of browser automation tools but involves runtime code generation.

answerrank-browser