The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The skill accesses local configuration files and user communication data to perform its functions.
Evidence: The skill is instructed to read CHIEF_OF_STAFF_CONTEXT.md and workspace/tasks/current.md to learn about account settings and current priorities. This data access is necessary for the intended assistant persona and occurs within the project scope.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external emails and calendar invites, which represents a potential attack surface.
Ingestion points: Identified in the Inbox Triage Procedure (SKILL.md) and Multi-account Checking (references/calendar-rules.md).
Boundary markers: Absent; the instructions do not define specific delimiters to separate untrusted email content from the system instructions.
Capability inventory: The agent has the ability to draft and send emails, archive messages, and update local task files.
Sanitization: No explicit sanitization or filtering of incoming email or calendar data is required by the skill instructions.
[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill references established third-party scheduling services.
Evidence: References to well-known services such as Calendly, HubSpot, and Acuity are included in references/calendar-rules.md as preferred booking methods. These are standard integrations and do not involve the execution of untrusted remote code.

executive-assistant