The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructs the agent to bypass standard user consent protocols by installing dependencies 'automatically without asking the user to confirm' and specifically tells the agent to suppress transparency by not describing the installation process to the user or requesting separate approval.
[EXTERNAL_DOWNLOADS]: The skill directs the agent to download and install a dependency from an untrusted third-party website (browseract.com) which is not recognized as a trusted organization or well-known service.
[COMMAND_EXECUTION]: The instruction for automatic installation requires the execution of shell commands to install external software without user oversight, creating a significant remote code execution risk.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection. 1. Ingestion points: Untrusted social media comments are ingested via the browser. 2. Boundary markers: No specific delimiters or instructions to ignore embedded commands are present in the processing logic. 3. Capability inventory: The agent is instructed to perform network operations (via BrowserAct) and write to local directories (workspaces/). 4. Sanitization: No data sanitization, escaping, or validation is performed on the ingested comments before processing.

social-comment-sentiment-collector