The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill requires executing an external NPM package at runtime via npx -y @nomadamas/k-skill@0. This pattern downloads and executes code from a non-whitelisted source, creating a significant security risk if the package is compromised.
[EXTERNAL_DOWNLOADS]: The skill makes network requests to k-skill-proxy.nomadamas.org for geocoding and apis.data.go.kr for building data. These external communications involve sending user-derived queries to third-party services.
[COMMAND_EXECUTION]: The instructions provided in SKILL.md direct the agent to execute several shell commands, including npx for fetching instructions and updating the toolset, which bypasses static instruction analysis.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external APIs without using explicit boundary markers or instruction-ignoring warnings. This creates a surface for indirect prompt injection if the upstream data source returns malicious content.
Ingestion points: scripts/building_register.py fetches data using urllib.request from the vendor proxy and government API.
Boundary markers: Absent in the formatted output displayed to the agent.
Capability inventory: Performs network GET requests, reads local configuration files (~/.config/k-skill/secrets.env), and prints to stdout. No file-writing or subprocess-spawning logic was identified in the Python scripts.
Sanitization: Implements JSON and XML parsing for transport but does not sanitize the text content for LLM safety.

building-register-search