The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill's primary instructions in SKILL.md and instruction.md require the agent to download and execute code from an external, untrusted source using the command npx -y @nomadamas/k-skill@0.2.4. This executes unverified software from the npm registry at runtime, which is a significant security risk as the package content is not controlled or audited within the skill itself.
[EXTERNAL_DOWNLOADS]: The skill fetches instructions and execution logic from the @nomadamas/k-skill package on npm. This external dependency is not from a trusted organization, and the skill forces the agent to treat the remote output as the primary source of instructions.
[COMMAND_EXECUTION]: The workflow involves constructing and running shell commands that incorporate user-provided text (e.g., npx ... --text "$TEXT" or cat $FILE | npx ...). This presents a high risk of command injection if the agent interpolates untrusted user content directly into the shell command without rigorous escaping of shell metacharacters.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted user data (e.g., resumes, essays) which could contain malicious instructions.
Ingestion points: User text is ingested through the --text argument, file paths provided to --file, or data piped via --stdin as shown in instruction.md and handled in scripts/korean_character_count.js.
Boundary markers: The instructions do not specify any delimiters or boundary markers to separate untrusted data from the command logic.
Capability inventory: The skill utilizes shell execution (via npx) to run a Node.js script that performs file reading (fs.readFileSync) and standard output printing.
Sanitization: There are no sanitization or escaping protocols described for handling user-supplied text when constructing the execution commands, leaving the agent vulnerable to shell-based injection attacks.

korean-character-count