The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to run npx -y @nomadamas/k-skill@0 instruct korean-transit-route to obtain its primary operating instructions. This pattern executes code downloaded from an external third-party registry at runtime.
[EXTERNAL_DOWNLOADS]: The skill requires multiple downloads of the @nomadamas/k-skill package for functions such as fetching instructions, listing helper files, and performing updates.
[DATA_EXFILTRATION]: User queries containing addresses or place names are sent to an external third-party proxy server (https://k-skill-proxy.nomadamas.org) for geocoding. While this is part of the functional design, it involves transmitting potentially sensitive user location data to a non-standard external endpoint.
[COMMAND_EXECUTION]: The skill uses shell commands to source environment variables from a local configuration file (~/.config/k-skill/secrets.env) and uses curl to make API calls, incorporating these secrets into command-line arguments.
[INDIRECT_PROMPT_INJECTION]: The skill processes transit route data and geocoding results from external APIs, creating a potential surface for indirect prompt injection if the upstream data sources are compromised.
Ingestion points: Data is received from the ODsay API and the Kakao geocoding proxy as documented in instruction.md.
Boundary markers: The skill does not implement explicit boundary markers or instructions to ignore embedded commands in the data it processes.
Capability inventory: The skill has the ability to perform network requests (curl, urllib.request) and read local configuration files.
Sanitization: Input queries are URL-encoded before being sent to external endpoints, which provides basic protection against malformed URI structures but not against higher-level instruction injection.

korean-transit-route