The Agent Skills Directory

[DATA_EXFILTRATION]: The CLI tool includes a --deliver webhook:<url> parameter that enables the agent to POST command results to any external URL. This feature represents a direct exfiltration vector if the agent is manipulated into processing and sending sensitive information.
[EXTERNAL_DOWNLOADS]: The skill instructions require installing binaries via go install from a third-party repository (github.com/mvanhorn) and npx for a package (@mvanhorn/printing-press), which executes code from external sources during the setup process.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface because it processes flight data from multiple external APIs.
Ingestion points: Data is retrieved from FlightAware AeroAPI, Google Flights, and Kayak.
Boundary markers: No explicit delimiters or warnings to ignore instructions within the API data are present in the skill.
Capability inventory: The skill can perform network operations (POST via webhooks), write to the file system, and execute shell commands.
Sanitization: No evidence of input/output sanitization or validation for external API content was found.
[COMMAND_EXECUTION]: The skill operates by driving a command-line interface (flight-goat-pp-cli), which is necessary for its functionality but expands the attack surface if the agent is tricked into injecting malicious arguments into the shell environment.

pp-flight-goat