The Agent Skills Directory

[SAFE]: The skill's primary function is to perform a beat-by-beat analysis of talking-head video scripts to identify points where viewers might lose interest. The logic is based on a defined 'listener model' (waiting vs. holding information) which is applied transparently.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided scripts, which constitutes a potential injection surface. However, the instructions include strict guidelines for handling user content, specifically the 'Step 4'改稿协议 (Edit Protocol) which mandates the use of clear markers (~~delete~~, 🆕 add) and explicitly forbids erasing original text, reducing the risk of the agent being misdirected by content within the script.
[DATA_EXPOSURE]: The skill is configured to read from local reference directories (e.g., aikey/references/) and a memory vault ({vault}/08-技能记忆/aikey-flow/). This behavior is documented and appropriate for a writing assistant designed to maintain consistent style and rules across sessions. No access to sensitive system paths (like .ssh or .aws) or hardcoded credentials was found.
[COMMAND_EXECUTION]: The skill does not contain any shell commands, subprocess calls, or scripts that execute outside of the LLM's prompt environment. Analysis is performed strictly through text processing.
[EXTERNAL_DOWNLOADS]: No remote code execution or unauthorized external downloads were detected. References to the author's GitHub repository are for documentation and user support purposes.

aikey-flow