The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill includes hardcoded credentials and insecure security configurations.
In SKILL.md (Step 8), the instructions provide hardcoded default credentials (root/root) for a local MySQL database (jeecgboot3) to be used in automated menu setup commands.
In scripts/onlreport_api.py, the Python script explicitly disables SSL certificate verification by setting ssl.verify_mode = ssl.CERT_NONE. This configuration exposes the communication with the backend API to man-in-the-middle (MITM) attacks, which could result in the theft of the user-provided X-Access-Token.
[COMMAND_EXECUTION]: The skill contains instructions for the agent to execute arbitrary shell commands.
It mandates using powershell -Command on Windows to execute Python scripts to ensure synchronous execution.
It provides a template for executing mysql CLI commands directly against the local host to insert permission records.
[DYNAMIC_EXECUTION]: The skill instructions involve the creation and execution of scripts at runtime.
Step 6 in SKILL.md guides the agent to write Python code into temporary files and execute them, which presents a risk if the generated logic is influenced by malicious input or lacks proper validation.
[INDIRECT_PROMPT_INJECTION]: The skill acts as a surface for indirect prompt injection by processing untrusted user data (SQL statements and report descriptions) to drive its operations.
Ingestion points: User-provided SQL strings and natural language report requirements in SKILL.md.
Boundary markers: The instructions suggest using Velocity template markers (${paramName}) to identify parameters within SQL queries.
Capability inventory: The skill can perform network operations via API calls, write to the file system, and execute shell commands.
Sanitization: The skill relies on a parseSql API to extract fields and parameters, but this provides structure rather than security sanitization, allowing potentially malicious SQL or data to be processed.

jeecg-onlreport