The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources such as dbt job logs, run_results.json artifacts, and Git repository histories, which represents an indirect prompt injection surface.
Ingestion points: Processes debug logs, JSON artifacts, and Git diff outputs.
Boundary markers: Includes a dedicated 'Handling External Content' section that explicitly instructs the agent to treat all external content as untrusted and to ignore any instruction-like text.
Capability inventory: Performs file system operations (git clone), shell command execution (git, dbt, gh), and calls MCP tools to query APIs.
Sanitization: Instructs the agent not to execute any scripts, code, or commands found within the analyzed logs or repositories.
[COMMAND_EXECUTION]: Executes shell commands such as git log, git diff, and various dbt CLI commands to analyze the project's state and identify failure causes.
[REMOTE_CODE_EXECUTION]: The skill utilizes dbt parse and dbt compile on downloaded repository content. Because dbt projects execute Jinja-based macros during these phases, this operation involves executing code derived from external Git repositories.
[EXTERNAL_DOWNLOADS]: Clones remote Git repositories to a local temporary directory using git clone to facilitate code review and diagnosis.

troubleshooting-dbt-job-errors