The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses internal control logic like "語言判斷（最優先執行）" (Language determination, top priority) and "門檻指令（不可妥協）" (Threshold instructions, non-negotiable). These are legitimate instructional constraints designed to maintain the quality and focus of the AI's output within the skill's intended use case, rather than attempts to bypass agent safety filters or administrative rules.
[DATA_EXFILTRATION]: No instances of sensitive data access (e.g., credentials, environment variables, or SSH keys) or network operations (e.g., curl, wget, or fetch) were detected. The skill only references local markdown files within its own directory.
[REMOTE_CODE_EXECUTION]: The skill does not include any instructions for downloading, installing, or executing external scripts or packages. It is purely focused on text processing and prompt generation.
[COMMAND_EXECUTION]: No shell commands, subprocess calls, or dynamic code execution patterns were found in any of the skill files.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data (user-provided skill drafts), which is a potential injection surface. However, the skill lacks the critical capabilities (like network access or file system writes) necessary to transform an injection into a functional exploit. It merely outputs text for the user to review or for a separate tool to handle.

decision-tree-helper