The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads the kordoc package from the npm registry using bunx at runtime.
[REMOTE_CODE_EXECUTION]: The execution protocol recommends installing the Bun runtime using a shell script from its official domain (bun.sh), which is executed directly in the terminal.
[REMOTE_CODE_EXECUTION]: The skill uses the bunx command to fetch and execute the kordoc utility from a remote package registry.
[COMMAND_EXECUTION]: The skill runs a local TypeScript script (flatten-tables.ts) and the kordoc command-line tool to perform document conversion and formatting.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted HWP, HWPX, and HWPML documents.
Ingestion points: Reads document files from the local filesystem (SKILL.md, execution-protocol.md).
Boundary markers: No delimiters or isolation instructions are present to separate document content from agent commands.
Capability inventory: The skill can execute shell commands via bunx and bun run, and performs file writes using Bun.write (flatten-tables.ts, execution-protocol.md).
Sanitization: While kordoc is reported to provide XSS and XXE defenses, there are no specific measures against indirect prompt injection.

oma-hwp