The Agent Skills Directory

[COMMAND_EXECUTION]: The instructions guide the agent to execute wmill CLI commands (wmill app new, wmill generate-metadata) to automate project scaffolding and metadata updates. The skill specifies using non-interactive flags to ensure these commands run successfully in the agent's environment.
[PROMPT_INJECTION]: The skill includes instructions to run CLI commands directly rather than asking the user to perform them manually to avoid interactive shell prompts that would hang the agent. This is a functional requirement for the agent environment, and the instructions require the agent to obtain all necessary configuration values from the user before executing any commands.
[DATA_EXPOSURE]: The skill includes explicit instructions for the agent to use the data-wm-no-record attribute in generated UI code. This practice ensures that sensitive user data is excluded from session recordings and public demos.
[INDIRECT_PROMPT_INJECTION]: The skill mitigates potential code injection vulnerabilities by mandating the use of parameterized queries in all SQL examples and instructions provided to the agent for backend development. It explicitly warns against string concatenation of user input into SQL commands.

raw-app