The Agent Skills Directory

[COMMAND_EXECUTION]: The skill contains numerous templates for shell-based automation and CI/CD pipelines.
Provides GitHub Actions YAML workflows that run 'pip', 'dbt', and 'sqlfluff' CLI tools.
Includes bash snippets for monitoring data quality via BigQuery ('bq query') and 'jq'.
Commands are appropriate for the intended DataOps purpose.
[EXTERNAL_DOWNLOADS]: Pipeline examples reference well-known tools and official repositories.
Uses standard GitHub Actions such as 'actions/checkout', 'actions/setup-python', and 'actions/download-artifact'.
Installs verified packages including 'dbt-core', 'dbt-bigquery', 'sqlfluff', and 'great-expectations' from official registries.
[INDIRECT_PROMPT_INJECTION]: The skill has an inherent attack surface for indirect injection as it is designed to process external dbt models and warehouse metadata.
Ingestion points include dbt project files ('models/**') and warehouse schema metadata.
The recommended usage of manual approval gates and multi-stage environment promotion acts as a mitigation against automated exploitation.
[METADATA_POISONING]: Multiple reference files contain over 100 sections of repetitive filler content.
Documentation is presented as 'extremely dense technical specifications' for 'Staff+ Engineers' but consists of a small set of technical paragraphs repeated hundreds of times.
This technique inflates the skill size and obscures the signal-to-noise ratio, though no malicious variant was detected in the repetitions.

devops-dataops