The Agent Skills Directory

[PROMPT_INJECTION]: Use of redundant content for context manipulation. Eight files in the references directory (architecture-patterns.md, code-organization.md, deployment-pipelines.md, error-handling.md, performance-optimization.md, security-best-practices.md, state-management.md, and testing-strategies.md) consist of exactly 150 sections of repetitive boilerplate text. This massive redundancy serves no functional purpose and is a known vector for prompt bombing and context window manipulation.
[PROMPT_INJECTION]: Indirect prompt injection vulnerability surface. The skill ingests untrusted data from multiple sources including transformation tool configurations, data models, and environment definitions (SKILL.md). It lacks explicit boundary markers or instructions to ignore embedded commands within these inputs. Given the skill's capability to execute shell commands (dbt, soda, data-diff), this represents a surface for malicious inputs to influence agent behavior.
[COMMAND_EXECUTION]: Automated execution of data tooling via shell. The skill generates and executes shell commands for dbt, soda scan, and data-diff (SKILL.md, data-comparison-tools.md). While these are consistent with the skill's primary purpose, they operate on data models provided in the agent context without prior sanitization.
[EXTERNAL_DOWNLOADS]: Fetching external dependencies. The skill installs several Python packages including dbt-postgres, dbt-unit-testing, and soda-core-postgres during its lifecycle (.github/workflows/data-tests.yml). It also references official dbt packages like equalexperts/dbt_unit_testing and dbt-labs/dbt_utils.

data-testing