The Agent Skills Directory

[NO_CODE]: The skill is composed entirely of natural language instructions, markdown tables, and structured diagnostic templates. It does not include any Python, Node.js, shell scripts, or other executable components.
[SAFE]: No malicious patterns such as prompt injection, persistence mechanisms, or obfuscation were detected. The logic follows a standard academic analysis framework (Toulmin model) and explicitly excludes out-of-scope tasks like content validation or language polishing.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided academic text, which is an ingestion point for untrusted data. However, since the skill has no access to sensitive files, network operations, or shell execution tools, this surface does not present a practical risk. The structured output phases provide clear boundaries for the agent's operations.

argument-auditor