The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to integrate with data sources that contain content from third parties, which may include malicious instructions.
Ingestion points: Data is ingested through email messages (rules/email-messages.md), meeting transcripts (rules/notetaker-meetings.md), and contact details (rules/contacts-crud.md).
Boundary markers: The instructions in SKILL.md and rules/security-untrusted-content.md explicitly define "prompt-injection boundaries" and require keeping provider-originated values separate from agent instructions.
Capability inventory: The skill provides instructions for sending emails, creating calendar events, and managing account configurations across multiple rule files (e.g., rules/email-messages.md, rules/calendar-events.md, rules/agent-accounts.md).
Sanitization: A mandatory security module (rules/security-untrusted-content.md) explicitly instructs the agent to treat all grant-scoped API data as untrusted, forbids following instructions found within that data, and requires explicit human-in-the-loop confirmation for any mutations derived from external content.
[EXTERNAL_DOWNLOADS]: The skill references official SDKs for Python, Node.js, and Ruby.
Evidence includes instructions for npm install nylas, pip install nylas, and gem install nylas in rules/sdk-quickstart.md and SKILL.md. These are official vendor packages from Nylas.
[COMMAND_EXECUTION]: The skill provides example commands for interacting with the Nylas API.
Evidence includes standard curl examples in AGENTS.md and rule files (e.g., rules/agent-accounts.md) for API authentication and resource management. These are documented for developer guidance and do not involve silent or unauthorized execution.

nylas-api