The Agent Skills Directory

[SAFE]: The skill implements a robust safety architecture by defining 'Action-Level Gateways' (动作级阶段门禁) and a mandatory checkpoint system (🔴 CHECKPOINT · 🛑 STOP). This prevents the agent from autonomously performing sensitive operations such as overwriting user files, publishing data, or submitting resumes without explicit, real-time user authorization.\n- [PROMPT_INJECTION]: The instructions contain defensive directives against adversarial user requests aimed at data fabrication. It explicitly forbids the agent from 'filling in the blanks' or manufacturing facts, metrics, or professional tools not present in the user's source materials, thereby maintaining the integrity of the generated output.\n- [DATA_EXFILTRATION]: The skill identifies potentially risky network-facing operations (uploading, public posting, sending) and strictly marks them as 'BLOCKED' by default. This ensures that career data remains within the local context unless a user explicitly grants permission for a specific transfer action.

career-hub