The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external .hwp documents and converts them into structured text formats like Markdown or JSON. This functionality creates an attack surface for indirect prompt injection, where an attacker could embed malicious instructions in a document that the agent later interprets.
Ingestion points: The HWPProcessor module reads local files through methods like toMarkdown, toJSON, and toHTML in SKILL.md.
Boundary markers: There are no specified delimiters or system instructions to ignore embedded commands within the processed document content.
Capability inventory: The skill can execute various actions based on document content, such as sending KakaoTalk messages, booking train tickets, or writing to the local file system.
Sanitization: The skill documentation does not outline sanitization or validation steps for the extracted document text.
[EXTERNAL_DOWNLOADS]: The skill instructs users to install several third-party packages and command-line tools to enable its functionality.
Evidence: The documentation recommends npm install -g k-skill, pip install python-hwp, npm install -g hwp.js, and brew install kakaocli. These dependencies are fetched from standard public registries.
[COMMAND_EXECUTION]: The skill relies on the execution of multiple CLI tools and subprocesses. This includes using kakaocli for messaging, sops and age-keygen for secret encryption, and k-skill for environment management and file conversion.

k-skill-korean-ai-tools