The Agent Skills Directory

[SAFE]: The skill proactively mitigates command injection risks when using the datahub CLI. It defines specific guidelines to reject shell metacharacters (such as backticks, dollar signs, and pipes) and requires the use of JSON variable files for complex identifiers to prevent unsafe shell interpolation.\n- [SAFE]: Protection against indirect prompt injection is integrated into the skill instructions. It explicitly identifies user-supplied metadata as untrusted and directs the agent to ignore any instructions embedded within fields like descriptions, tags, or glossary terms.\n- [SAFE]: A strict human-in-the-loop workflow is enforced for all write operations. The skill requires the agent to resolve entities, build a comparison plan, and obtain explicit user approval before executing any changes via GraphQL mutations.\n- [SAFE]: The skill handles data with integrity by verifying changes after execution and providing templates for rollback (undo commands) in the enrichment reports.

datahub-enrich