The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions are focused entirely on the summarization task. There are no attempts to bypass safety filters, extract system prompts, or override the core behavior of the AI agent with administrative or 'jailbreak' commands.\n- [DATA_EXFILTRATION]: No sensitive file access or unauthorized network requests were detected. The skill instructions specifically mention that the skill does not fetch external media directly from URLs, relying instead on user-provided transcripts or files.\n- [COMMAND_EXECUTION]: The skill provides an installation command using npx, which points to the author's own GitHub repository (peter-tu-zynkr/zynkr-skill-builder). This is a standard practice for skill distribution and does not constitute a security risk in this context.\n- [PROMPT_INJECTION]: (Indirect) The skill ingests untrusted external data in the form of transcripts and video content (Step 1: SOURCE). While it lacks explicit boundary markers to encapsulate this data, its primary capability is limited to formatted text output in a recap document, which presents a negligible security risk regarding instruction following from the input data.

training-lecture-recap