The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions in SKILL.md use highly prescriptive language to override agent autonomy, demanding the agent run a specific shell command as its first action and explicitly forbidding clarifying questions or the use of native agent tools (e.g., browser_navigate). This behavior steering aims to force the agent into a specific, potentially less-supervised mode of operation.
[COMMAND_EXECUTION]: The skill is designed around the execution of complex shell commands via the terminal tool. This includes the primary search script (node bin/search.js), the browser daemon (node bin/daemon.js), and an installation script that executes npm install and npx commands.
[EXTERNAL_DOWNLOADS]: During the installation phase (install.sh), the skill downloads a Chromium binary via the patchright utility. At runtime, it performs fetches to external domains including open.er-api.com for currency exchange rates and the vendor's own API at api.opentravel.one.
[DATA_EXFILTRATION]: The skill initiates outbound network connections to external domains to retrieve travel and currency data. These operations represent a network communication surface where data is exchanged with third-party servers.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting and processing unstructured text data from external travel sites (Booking.com, Agoda, Google Hotels). This data is extracted and presented to the agent without explicit boundary markers or sanitization to prevent embedded instructions from influencing agent behavior.

pricewin-deal-finder