The Agent Skills Directory

[COMMAND_EXECUTION]: The templates and configuration files (setup.sh, app.conf) use standard package managers like uv and npm to install dependencies and run the applications. These are official patterns for the Keboola deployment lifecycle and do not execute arbitrary or malicious commands.- [CREDENTIALS_UNSAFE]: The skill manages sensitive environment variables like KBC_TOKEN and WORKSPACE_ID. It correctly instructs the agent to avoid insecure harvesting methods (e.g., grepping the filesystem) and enforces the use of .env.local and .gitignore to prevent accidental credential leakage in repository-based deployments.- [REMOTE_CODE_EXECUTION]: The skill suggests the installation of kbagent via a GitHub repository. As this is the recommended CLI tool for managing the platform's resources and the repository is linked to the vendor's ecosystem, it is classified as safe functionality. No unverified or piped remote script executions were found.- [DATA_EXFILTRATION]: While the skill facilitates network operations to Keboola APIs, these are restricted to well-known vendor domains (connection.*.keboola.com, query.*.keboola.com). There is no evidence of sensitive data being transmitted to unauthorized third-party servers.- [PROMPT_INJECTION]: The skill's instructions focus on application development logic and do not contain patterns typical of jailbreak attempts or behavioral overrides. Instructional language is limited to platform-specific rules.- [INDIRECT_PROMPT_INJECTION]: The skill allows apps to process external data from Storage tables, creating a potential attack surface. However, it provides robust remediation guidance in references/storage-access.md, including specific code for SQL validation and escaping to mitigate injection risks.

dataapp-development