The Agent Skills Directory

[REMOTE_CODE_EXECUTION] (HIGH): The 'useScript' hook (rules/useScript.md) enables the loading and execution of external JavaScript files. This allows for arbitrary code execution in the browser context if an agent is directed to use an attacker-controlled script URL.\n- [PROMPT_INJECTION] (HIGH): The skill exposes a broad attack surface for Indirect Prompt Injection (Category 8). 1. Ingestion points: 'useWebSocket' (rules/useWebSocket.md), 'usePostMessage' (rules/usePostMessage.md), 'useClipboard' (rules/useClipboard.md), and 'useUrlSearchParams' (rules/useUrlSearchParams.md). 2. Boundary markers: No delimiters or ignore-instructions warnings are provided in the documentation or usage examples. 3. Capability inventory: The skill allows for script execution (useScript), modifying local storage/cookies (useLocalStorage), and making network requests (useWebSocket). 4. Sanitization: There is no guidance or built-in logic for sanitizing or escaping ingested content before it influences the agent's state or actions.\n- [DATA_EXFILTRATION] (MEDIUM): The library facilitates access to sensitive data and sensors (Category 2), including geolocation (rules/useGeolocation.md), bluetooth (rules/useBluetooth.md), and clipboard content (rules/useClipboard.md). These capabilities, when paired with the skill's network hooks (useWebSocket, useEventSource), provide the necessary primitives for unauthorized data exposure and exfiltration.

reactuse