The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from an external platform, which could contain malicious instructions designed to influence the agent's behavior. \n
Ingestion points: Tools like listChannelMessages, getMessage, searchMembers, and getUser retrieve data such as message content, user biographies, and server metadata that are controlled by external users. \n
Boundary markers: The SKILL.md file provides instructions for the agent to disambiguate between multiple matches and explicitly defines unsupported actions (like moderation) to prevent unintended execution. \n
Capability inventory: The connector has extensive capabilities including sending/editing messages, managing roles, creating channels, and executing webhooks, which could be abused if an injection is successful. \n
Sanitization: The skill provides mechanisms like allowed_mentions (documented in references/discord-formatting.md) to allow the agent to control whether mentions in content actually trigger notifications, providing a layer of protection against ping-spamming attacks. \n- [COMMAND_EXECUTION]: The cli.js file uses the node:child_process module to spawn a subprocess that executes the connector's internal logic. This is an expected architectural pattern used by this vendor to manage runtime environments and ensure that import.meta.main is correctly evaluated for dispatching tool calls. \n- [EXTERNAL_DOWNLOADS]: The skill's documentation directs users to install dependencies and the connector package itself using standard package managers (npm, npx). These downloads target the official @zapier/discord-connector package on the npm registry, which is the intended distribution method for this skill.

discord