The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from the user's dbt project.
Ingestion points: The agent reads and parses profiles.yml, dbt_project.yml, packages.yml, source definitions, and SQL model files (SKILL.md, references/switching-targets.md).
Boundary markers: The skill includes a 'Handling External Content' section that explicitly instructs the agent to treat project files as untrusted and to ignore instructions embedded in comments or metadata.
Capability inventory: The agent has the capability to execute shell commands (dbtf, rm, jq, grep), modify local files (migration_changes.md, profiles.yml), and access environment variables (SKILL.md, references/switching-targets.md).
Sanitization: The instructions direct the agent to extract only structured fields and ignore any instruction-like text within project files.
[COMMAND_EXECUTION]: The skill requires the execution of multiple shell commands to perform the migration workflow.
Evidence: Commands include dbtf compile, dbt test, dbt show, and a cleanup command rm -rf target/. It also uses shell pipes with utilities like jq, grep, sort, and comm to analyze the project DAG (SKILL.md, references/generating-unit-tests.md).
[EXTERNAL_DOWNLOADS]: The skill references downloading and updating external software from a trusted vendor.
Evidence: Instructions guide the user to install 'dbt Fusion' from an official dbt Labs domain and use the dbtf system update command for tool maintenance (references/installing-dbt-fusion.md).
[METADATA_POISONING]: There is an inconsistency in the skill's attribution that could be misleading.
Evidence: The skill metadata lists the author as 'dbt-labs', whereas the provided context identifies the author as 'kilo-org'. This discrepancy could lead to a misjudgment of the skill's origin or authority (SKILL.md).

migrating-dbt-project-across-platforms