The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructions in SKILL.md and docs/skill-usage-guide.md contain hardcoded database credentials (-uroot -proot) within command templates used to query and update the local MySQL database.
[COMMAND_EXECUTION]: The agent is instructed to execute various shell commands, including mysql, find, grep, ls, and date, to interact with the local file system and database. This behavior is central to the skill's function but involves executing commands with user-provided or inferred parameters on the local environment.
[INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection as it processes untrusted data from external sources.
Ingestion points: The skill reads database schemas (via mysql commands), configuration files like application-dev.yml, and natural language requirements provided by the user.
Boundary markers: While it uses markers like update-begin/update-end for code organization, it lacks explicit security boundary markers or instructions to disregard malicious content embedded in ingested DDL or requirements.
Capability inventory: The agent has the ability to write and edit files across the project directory and execute SQL commands on the database.
Sanitization: There is no evidence of sanitization or strict validation of the ingested DDL or user requirements before they are incorporated into generated code or SQL migration files.
[DYNAMIC_EXECUTION]: The parallel-generation-mode.md documentation describes a beta mode where the main agent dispatches tasks to two SubAgents to generate frontend and backend code in parallel. This dynamic task delegation increases the complexity of the execution environment and relies on the main agent to validate output for consistency and safety.

jeecg-codegen