The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes hotel descriptions, amenities, and policies retrieved from an external API. This data is interpolated into agent responses and translated, creating a surface for indirect injection if the API source were compromised. The instructions lack explicit boundary markers for this untrusted data.
[EXTERNAL_DOWNLOADS]: The skill implements a self-update mechanism that polls a vendor API for new versions. If an update is available, the agent is instructed to fetch new files from a provided release URL and overwrite local skill files, subject to user confirmation.
[DATA_EXFILTRATION]: The skill collects and transmits user information, such as full guest names and contact emails, to a remote API to complete bookings. This is a functional requirement of the skill but involves the transfer of PII to an external service.
[COMMAND_EXECUTION]: The skill utilizes shell commands in the test suite to verify repository state (e.g., ensuring secrets are not tracked by Git) and may use command-line tools like git during the update process to synchronize files.
[PROMPT_INJECTION]: The instructions contain directives to suppress specific technical outputs, such as hiding failures of the background update check and omitting notices when specific fee or tax metadata is unavailable from the API, to maintain the intended user experience.
[DYNAMIC_EXECUTION]: The update workflow involves the agent dynamically updating its own instructions and scripts by downloading and overwriting files from a remote source at runtime, which effectively modifies the agent's behavior based on external input.

tourmind-booking