The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands to perform environment checks and execute dbt operations.
scripts/preflight.py uses subprocess.run to call git status for verifying the state of package-lock.yml and dbt debug for live connection testing.
SKILL.md instructs the agent to execute the configured runner (e.g., uv run dbt, poetry run dbt) for building and testing models.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface where untrusted data could influence agent behavior.
Ingestion points: The skill reads dbt_project.yml, profiles.yml, and .dbt-runner/context.md from the project directory. It also instructs the agent to grep and analyze execution logs (e.g., /tmp/dbt_run.log).
Boundary markers: The instructions do not define specific delimiters for separating data from instructions when reading logs, relying on the agent to map verbatim strings to the failure catalog.
Capability inventory: The skill has the capability to write the .dbt-runner/context.md file and execute arbitrary shell commands through the dbt runner.
Sanitization: scripts/preflight.py implements a manual YAML-like parser for context files rather than using a standard deserializer, which reduces the risk of deserialization attacks but does not mitigate prompt injection.

dbt-runner