The Agent Skills Directory

[METADATA_POISONING]: The skill displays conflicting identity information. While the platform metadata identifies the author as 'reason-machines', the internal content claims it is a skill by 'ara.so' and provides links to a GitHub repository under the user 'law-chain-hot'. This inconsistency in branding and source ownership can mislead users regarding the origin and safety of the software.
[EXTERNAL_DOWNLOADS]: The skill instructs users to install a browser extension from an untrusted GitHub repository (github.com/law-chain-hot/websocket-devtools.git) via Developer Mode. Browser extensions can access all website data, including passwords and cookies, and manual installation from unverified sources bypasses standard browser extension security audits.
[INDIRECT_PROMPT_INJECTION]: The skill allows monitoring and simulation of WebSocket messages. A malicious server could inject instructions into these messages to compromise an AI agent that is interpreting the traffic.
Ingestion points: Monitoring of WebSocket onmessage events in the DevTools background script and monitoring panel.
Boundary markers: The instructions lack delimiters or warnings for the agent to treat captured WebSocket content as untrusted data.
Capability inventory: The skill includes capabilities to monitor, simulate, and replay messages across multiple connections.
Sanitization: There is no mention of sanitization or escaping of the captured WebSocket message content before it is displayed or processed by the agent.

websocket-devtools-extension