outlook-calendar
Warn
Audited by Gen Agent Trust Hub on Aug 19, 2026
Risk Level: MEDIUMCREDENTIALS_UNSAFEEXTERNAL_DOWNLOADSPROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [DATA_EXPOSURE_AND_EXFILTRATION]: The skill accesses a sensitive configuration file located at
~/.outlook/config.jsonwhich contains the user's plaintext email and password. Although the instructions state this file should be created manually by the user, the agent's scripts read these credentials directly, which constitutes a data exposure risk. Furthermore, session cookies and Bearer tokens are stored and read from the same directory, creating a surface for session hijacking if the local environment is compromised.- [PROMPT_INJECTION]: TheSKILL.mdfile includes a directive stating that the agent 'must' call this skill for any query involving a wide range of common keywords such as 'tasks,' 'work,' or 'matters.' This is an aggressive attempt to override the agent's default tool-selection logic and influence its reasoning behavior.- [UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill setup requires runningplaywright install chromium, which downloads and installs binary browser components from external Microsoft-controlled servers. While the source is a well-known service provider, this represents an unverified external binary dependency.- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the Outlook API and provides it to the agent without sanitization. Ingestion point: Meeting subjects are retrieved byowa_calendar.py. Boundary markers: There are no delimiters or 'ignore' instructions used when presenting this data to the agent. Capability inventory: The skill has permissions to read local sensitive files and perform network operations. Sanitization: None. A malicious actor could send a meeting invite with a subject containing instructions designed to hijack the agent's session or exfiltrate data.- [CREDENTIALS_UNSAFE]: The skill's primary authentication mechanism relies on the storage and retrieval of a plaintext password from a local file, which is a significant security weakness compared to using secure token storage or environment-specific secret management.
Audit Metadata