The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides code templates for executing external shell commands and binaries using Node.js child_process modules.
Evidence includes instructions for using exec, execFile, and promisify(exec) to interact with system tools like sqlite3 for database access.
[DATA_EXFILTRATION]: The skill documents methods for accessing sensitive system-level information and managing data transfers.
Evidence shows patterns for reading the active browser URL (Chrome and Safari) and the frontmost application name using AppleScript via runAppleScript.
Evidence includes templates for reading from and writing to the system clipboard using the @raycast/api Clipboard utility.
[PROMPT_INJECTION]: The skill facilitates the creation of tools that ingest untrusted data from external sources, creating a surface for indirect prompt injection.
Ingestion points: Data enters the extension context through Clipboard.readText(), AppleScript browser URL retrieval, and sqlite3 database queries (SKILL.md).
Boundary markers: The provided templates do not include explicit delimiter-based isolation or instructions to ignore embedded commands in processed data.
Capability inventory: Extensions built with these templates have access to network fetching (fetch), shell command execution (exec/execFile), and clipboard manipulation (SKILL.md).
Sanitization: The code samples do not demonstrate sanitization or validation of the retrieved external content before processing.

raycast-extension