The Agent Skills Directory

[SAFE]: The skill implements standard restaurant booking functionality using an official connector (opentable). Analysis of the instructions and metadata revealed no malicious patterns or security risks.
[DATA_EXPOSURE_AND_EXFILTRATION]: Instructions strictly follow data minimization principles. It explicitly directs the agent to only include relevant special requests (e.g., allergies) and prohibits including unrelated personal data in the booking. It also manages contact details (name, email, phone) through authorized profiles or direct user input.
[PROMPT_INJECTION]: No evidence of prompt injection or instructions to bypass safety guidelines was found. The instructions use clear, natural language focused on the booking flow.
[REMOTE_CODE_EXECUTION]: There are no patterns of remote code execution, external script downloads, or dynamic code generation. The skill relies on predefined connector methods.
[INDIRECT_PROMPT_INJECTION]: The skill processes restaurant and availability data. It includes boundary-aware instructions to summarize results in plain language and avoid quoting raw provider-internal status fields, reducing the risk of indirect injection from external data sources.
[PRIVILEGE_ESCALATION]: The skill operates within the scope of the opentable connector permissions defined in manifest.yaml and does not attempt to gain unauthorized system privileges.

opentable