The Agent Skills Directory

[DATA_EXFILTRATION]: Accesses sensitive local files including 'profiles.yml' and '.env' files to determine project configuration and environment settings (documented in 'SKILL.md' Step 2 and 'kb/core/1_4_005.yaml'). These files frequently contain database credentials, access tokens, and other sensitive information.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill reads dbt project files, including SQL models, macros, and YAML configurations ('SKILL.md' Step 2).
Boundary markers: The skill instructions ('SKILL.md' Rule 8) explicitly warn the agent to treat project files and command output as untrusted and to ignore instructions embedded within them.
Capability inventory: The skill has extensive shell and tool capabilities via 'Bash' (git, dbt, uvx, uv), enabling it to perform file edits and command execution.
Sanitization: No explicit sanitization or filtering of project content is described before the agent processes it.
[REMOTE_CODE_EXECUTION]: Downloads and executes Python packages and CLI tools at runtime using 'uvx' and 'uv run --with', as seen in the local execution profile ('references/exec-local.md').
[COMMAND_EXECUTION]: Extensive execution of shell commands for project migration and state management, including git operations and dbt project parsing. Static analysis identifies the use of 'subprocess.run()' in the skill's supporting scripts.
[DYNAMIC_EXECUTION]: Automatically generates and manages virtual environments to run specific versions of 'dbt-core' and adapter plugins for verification purposes ('references/exec-local.md').

upgrading-dbt