The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it processes untrusted data from external emails. -- Ingestion points: scripts/outlook-mail.sh (read, inbox, search, and folder). -- Boundary markers: Absent. -- Capability inventory: scripts/outlook-mail.sh (send, draft, delete, move, attach) and scripts/outlook-calendar.sh (create, update, delete). -- Sanitization: HTML tags are stripped using jq's gsub function in scripts/outlook-mail.sh, but the resulting plain text content is not sanitized for potential instructional overrides.
[COMMAND_EXECUTION]: The skill uses a collection of shell scripts to perform its operations, executing system commands such as curl, jq, az, and pandoc across all script files. The scripts/outlook-setup.sh script automates Azure application registration and OAuth flows, which are necessary for the skill's intended functionality.

outlook