The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill is designed to clone upstream git repositories and download files from cloud storage providers (S3/ADLS) as specified in a bundle_config.py configuration file. It also provides instructions to invoke build tools such as sbt, Maven, and Gradle to assemble and smoke-test Scala JARs. These high-privilege operations are mitigated by instructions requiring explicit user confirmation before any network fetch or build invocation, and the use of static AST parsing to inspect configuration files without executing them.
[COMMAND_EXECUTION]: The skill makes extensive use of the databricks CLI for operations including workspace management, job configuration updates, and filesystem operations. These are standard operations within the Databricks ecosystem for the intended migration task.
[DATA_EXFILTRATION]: A structured failure reporting protocol allows users to submit anonymized diagnostic data to a GitHub repository. The skill implements a robust safety framework for this process, including a detailed redaction checklist for PII/secrets, a requirement for manual user review of the JSON report, and a mandatory deterministic post-serialization scrub to detect and block potential data leakage before generating a submission link.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user notebooks and configuration files as part of its analysis phase. It includes specific linting rules (e.g., cell-magic boundary checks) and uses static analysis patterns to minimize the risk of the agent misinterpreting data as instructions.

databricks-serverless-migration