The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill's primary setup and workflow rely on an external package from an untrusted source. Instructions in SKILL.md and instruction.md mandate executing npx -y @nomadamas/k-skill@0 to fetch instructions and to run helper scripts. The -y flag bypasses package installation prompts, resulting in the silent download and execution of arbitrary remote code.
[DYNAMIC_EXECUTION]: The script scripts/test_scholarship_filter.py utilizes dynamic module loading at runtime. Specifically, the load_module function uses importlib.util.spec_from_file_location and spec.loader.exec_module to execute Python code from arbitrary file paths.
[COMMAND_EXECUTION]: The skill involves shell command execution via subprocess calls. In scripts/test_scholarship_filter.py, the subprocess.run method is used to launch the Python interpreter and execute filtering scripts as separate processes.
[INDIRECT_PROMPT_INJECTION]: The skill defines a workflow for ingesting and processing data from external, untrusted web sources (scholarship announcements), which creates a surface for indirect prompt injection attacks.
Ingestion points: Scholarship records are retrieved from external domains like kosaf.go.kr and various *.ac.kr university sites as described in the search workflow.
Boundary markers: The instructions lack boundary markers or specific guidance for the agent to ignore potentially malicious instructions embedded within the retrieved scholarship content.
Capability inventory: The skill possesses command execution capabilities (via npx and subprocess) and file system access for processing JSON records.
Sanitization: No sanitization or filtering is performed on the extracted scholarship text before it is presented in the final report to the agent/user.

korean-scholarship-search