The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill directs the agent to execute npx -y @nomadamas/k-skill@0 instruct k-dart. This command downloads and runs code from a remote, third-party npm package at runtime to fetch primary instructions, allowing the skill's logic to be updated dynamically without changes to the local skill files.
[COMMAND_EXECUTION]: The skill uses shell commands including curl, grep, awk, and unzip on Linux/macOS, and Invoke-WebRequest and Expand-Archive on Windows to interact with the DART API and process downloaded files.
[EXTERNAL_DOWNLOADS]: The skill fetches a corporate code database (CORPCODE.xml) from the official DART API (https://opendart.fss.or.kr/api/corpCode.xml) and extracts it to the local temporary file system.
[CREDENTIALS_UNSAFE]: The API_K_DART authentication key is passed as a plaintext query parameter in curl and Invoke-WebRequest commands. This practice can lead to the key being exposed in command history, system logs, or network proxy logs.
[INDIRECT_PROMPT_INJECTION]: The skill ingests external data from the DART API and corporate registry files, creating a vulnerability surface for indirect instructions.
Ingestion points: Data is fetched from corpCode.xml and various JSON endpoints on opendart.fss.or.kr (e.g., list.json, company.json).
Boundary markers: The instructions do not define any delimiters to separate external data from agent instructions.
Capability inventory: The skill has capabilities for file system access (writing to /tmp or $env:TEMP) and shell command execution.
Sanitization: There is no evidence of sanitization or filtering of the external API content before it is processed by the agent.

k-dart