The Agent Skills Directory

[SAFE]: The skill implements secure configuration management practices by mandating that environment variables are only accessed at the application entry point. It requires that all configuration values, including absolute file paths, are resolved centrally and passed through a 'Universe' object, which prevents individual processing units (bUnits) from accessing sensitive environment data or performing unsafe path resolutions.
[PROMPT_INJECTION]: The skill defines a data pipeline architecture designed to ingest and process external data, establishing an inherent surface for indirect prompt injection. * Ingestion points: Untrusted data enters the agent context through Stage 1 (Ingest) adapters as described in the implementation order and delegation references. * Boundary markers: The provided templates and references do not specify the use of delimiters or 'ignore' instructions to isolate embedded data from the agent's core instructions. * Capability inventory: The pipeline framework includes capabilities for file system writes (Export adapters and 5r_reuse bUnits), logging operations (via the run_and_log_function decorator), and complex data transformations across multiple stages. * Sanitization: No explicit sanitization, validation, or escaping logic for external content is included in the base bUnit or orchestrator templates.

bclearer-pipeline-engineer