The Agent Skills Directory

[PROMPT_INJECTION]: The SKILL.md file contains mandatory rules that instruct the AI agent to override default behaviors, such as prohibiting the direct reading of memory credentials and mandating immediate execution of user-specified styles without internal deliberation. These instructions function as behavioral overrides.
[CREDENTIALS_UNSAFE]: Multiple files, including bi_utils.py, datasource_ops.py, and proc_ops.py, contain a hardcoded secret key (SIGNATURE_SECRET = 'dd05f1c54d63749eda95f9fa6d49v442a') used to sign API requests. While this is a vendor-owned resource (jeecgboot), it is a static secret embedded in the skill. Additionally, several documentation files provide example connection strings using default credentials like root/root.
[DYNAMIC_EXECUTION]: The skill is designed to generate and inject executable JavaScript code into component configurations (e.g., jsConfig in link_ops.py and customOption in JCommon components). It also enables the creation of SQL stored procedures. These dynamic code generation features are central to the skill's purpose as a dashboard and big-screen generator.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a significant attack surface as it ingests untrusted data from multiple external points, including SQL databases, YApi mock interfaces, and uploaded files (Excel/CSV). This data is directly interpolated into configurations and scripts without specified sanitization or boundary markers, creating a path for malicious data to influence agent behavior.
[COMMAND_EXECUTION]: The script gen_all_comps.py utilizes subprocess.run with shell=True to execute the system clip command for clipboard interactions. Similarly, proc_ops.py uses subprocess to orchestrate other local Python scripts.

jimubi-bigscreen