The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates reading untrusted data from external sources, specifically Microsoft Outlook emails, calendar events, and contact information. This creates an attack surface for indirect prompt injection, where an attacker could send an email containing malicious instructions designed to manipulate the AI agent's logic once the message is processed. 1. Ingestion points: The skill utilizes tools such as OUTLOOK_GET_MESSAGE, OUTLOOK_QUERY_EMAILS, and OUTLOOK_LIST_EVENTS to fetch external content. 2. Boundary markers: No specific delimiters or safety instructions are defined in the skill to separate untrusted email content from the agent's core instructions. 3. Capability inventory: The skill has the ability to manage mail folders, contacts, and calendar events, and interacts with attachments via OUTLOOK_DOWNLOAD_OUTLOOK_ATTACHMENT. 4. Sanitization: There is no evidence of content validation or sanitization before the agent processes retrieved data.
[SAFE]: The skill follows recommended security practices by using OAuth-based authentication via the Composio platform rather than requesting or storing hardcoded credentials.
[EXTERNAL_DOWNLOADS]: The skill references external resources including the Rube MCP server (rube.app/mcp) and official Composio documentation. These are standard components for the skill's stated purpose of providing an Outlook automation interface and are treated as safe service endpoints.
[DATA_EXFILTRATION]: No patterns of unauthorized data exfiltration were found. Access to sensitive Outlook data is scoped to the tools required for the automation tasks described in the skill metadata, and no third-party data-send operations were identified.

outlook-automation