The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses the sensitive .env file to retrieve the WEBHOOK_URL. While it specifically targets the WEBHOOK_URL key, accessing the environment file is a sensitive operation.
[EXTERNAL_DOWNLOADS]: The skill uses curl to transmit data to an external Discord webhook URL. As the URL is configured in the .env file, the agent interacts with external network endpoints.
[COMMAND_EXECUTION]: The instructions provide Bash scripts that utilize grep, cut, cat, and curl to process files and interact with the network. The skill uses quoted heredocs ('PAYLOAD') and double-quoted variables to prevent shell injection within these scripts.
[PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it ingests untrusted user-supplied messages that are subsequently processed and sent to an external service.
Ingestion points: The user-provided message passed as an argument to the /discord-webhook command in SKILL.md.
Boundary markers: The skill contains explicit instructions to JSON-escape message content to prevent injection into the API payload.
Capability inventory: The skill utilizes the Bash tool to perform file reads (grep on .env), write temporary files (cat), and execute network operations (curl).
Sanitization: The instructions mandate JSON-escaping of all message content and the use of temporary files for payloads to avoid shell escaping vulnerabilities.

discord-webhook