The Agent Skills Directory

[COMMAND_EXECUTION]: Several utility scripts (init_project.py, create_pipeline.py, run_pipeline.py, and list_pipelines.py) utilize the subprocess module to execute shell commands. These commands are restricted to invoking the flowerpower CLI and the pip package manager for project setup and pipeline management tasks.
[EXTERNAL_DOWNLOADS]: The init_project.py script facilitates the installation of the flowerpower package and its optional dependencies (e.g., io, ui) from the official Python Package Index (PyPI). This is a standard practice for ensuring the necessary library environment is available for the skill's operations.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of data from various external sources including S3, Google Cloud Storage (GCS), and local files for processing in ETL pipelines. While this introduces a potential attack surface for indirect prompt injection if the ingested data contains malicious instructions, the skill implements standard data engineering practices, and the associated risk is considered low and managed by standard model guardrails. Evidence chain:
Ingestion points: S3 URIs, local Parquet/CSV files, and DuckDB databases referenced in SKILL.md and references/advanced-patterns.md.
Boundary markers: None explicitly defined for raw data ingestion.
Capability inventory: File system writes (Polars/Delta Lake), network requests (Slack notification hooks), and shell command execution via helper scripts.
Sanitization: Examples include data validation steps using the Pandera framework to enforce schema constraints.

using-flowerpower