The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands via curl and jq to interact with the Discord REST API. It also utilizes a local Python script (scripts/discordbot.py) to manage complex logic. Input validation is enforced for all Discord IDs (Snowflakes) using regular expressions to ensure they conform to the platform's expected 17-20 digit format.
[DATA_EXFILTRATION]: All network operations are directed to the official Discord API at https://discord.com/api/v10. The skill uses the DISCORDBOT_TOKEN environment variable for authentication, which is the standard and safe method for providing credentials to agent-based tools. No sensitive local files or environment variables beyond the intended bot token are accessed or exfiltrated.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection because it reads message content from external Discord channels.
Ingestion points: SKILL.md (Reading recent messages) and scripts/discordbot.py (fetching a specific consent message and reading DM history).
Boundary markers: The instructions explicitly warn the agent to "confirm the exact destination and final content with the user" before sending any data, which acts as a robust human-in-the-loop mitigation.
Capability inventory: The skill can read channel history and send messages/DMs.
Sanitization: All outbound payloads are constructed using jq --arg or Python's json.dumps, ensuring that any malicious content in the message strings cannot escape the JSON structure or inject additional API parameters.

discordbot