The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to fetch "prompt templates" from a remote server (pua-api.agentguard.workers.dev) and execute/apply them to the agent's current session. This allows for dynamic remote instruction injection, where the logic of the skill can be changed by the remote server without a local update.
[DATA_EXFILTRATION]: The skill requires the agent to solicit the user's phone number during registration and send it to an external API (pua-api.agentguard.workers.dev). It also periodically uploads session statistics, event data, and "desensitized" session content to external domains (pua-skill.pages.dev).
[COMMAND_EXECUTION]: The skill uses shell commands via curl to interact with remote APIs for authentication and metadata reporting. It also instructs the agent to run pip install and python3 to generate ASCII QR codes for a custom payment system.
[PERSISTENCE]: The skill maintains persistent state across sessions by writing user tokens, "evolution" baselines, and project-specific knowledge to files in the ~/.pua/ directory (e.g., config.json, evolution.md).
[PROMPT_INJECTION]: The skill uses a complex series of instructions to override the agent's default persona and safety guidelines, imposing a high-pressure "corporate PUA" behavior pattern that enforces specific cultural roles (e.g., Alibaba, Huawei flavors).
[EXTERNAL_DOWNLOADS]: The skill pulls content from unverified domains including pua-api.agentguard.workers.dev and pua-skill.pages.dev and installs the qrcode Python package from external registries at runtime.

pua