The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external files (.hwp, .hwpx, .pdf) and renders their contents into Markdown for the agent.
Ingestion points: fs.readFileSync(filePath) in SKILL.md reads content from user-provided file paths.
Boundary markers: No specific delimiters or safety instructions are provided to the agent to treat the parsed table content as untrusted data.
Capability inventory: The skill possesses file system read capabilities and executes shell commands via npm exec.
Sanitization: The skill lacks sanitization of the document content before converting it to Markdown.
[DYNAMIC_EXECUTION]: The Node.js script dynamically computes the file path to the kordoc library at runtime by inspecting the environment's PATH variable and then uses a dynamic import() call to load the module.
Evidence: const kUrl = "file:///" + p.join(p.resolve(nmBin, ".."), "kordoc", "dist", "index.js").split(p.sep).join("/"); import(kUrl).then(...) in SKILL.md.
[COMMAND_EXECUTION]: The skill uses npm exec to set up a package context and run a shell script that executes arbitrary Node.js code.
Evidence: npm exec --yes --package=kordoc --package=pdfjs-dist -- node <<'SCRIPT' in SKILL.md.
[DATA_EXFILTRATION]: The skill accesses local files using absolute paths provided by the user or resolved by the agent.
Evidence: fs.readFileSync(filePath) in SKILL.md where filePath is derived from user input.

kordoc-parse-table