The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions are strictly task-oriented and do not contain any patterns typical of prompt injection, such as attempts to override safety filters, bypass constraints, or extract system prompts.
[DATA_EXFILTRATION]: No network-enabled tools or commands (e.g., curl, wget) are used. The skill only processes and returns text to the user.
[REMOTE_CODE_EXECUTION]: There are no patterns involving the download or execution of external scripts or packages. The skill is entirely comprised of Markdown and YAML configuration.
[COMMAND_EXECUTION]: The skill does not invoke any shell commands or subprocesses. It lacks the 'allowed-tools' or configuration necessary to perform system-level actions.
[CREDENTIALS_UNSAFE]: No hardcoded secrets, API keys, or tokens were detected. The skill does not access sensitive credential paths.
[DATA_EXPOSURE]: The skill processes user-provided text as input for its 'Detection' and 'Rewrite' phases. While this is an ingestion point for potentially malicious instructions (Indirect Prompt Injection), the skill has no capabilities (like network access or file-writing) that would allow for an exploit.
[EXTERNAL_DOWNLOADS]: The skill contains a reference to a Wikipedia article for documentation purposes in 'detection-patterns.md'. Wikipedia is a well-known and trusted information service; this reference is informational and does not trigger runtime execution.

humanize