The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a wrapper script (scripts/run.py) to execute the uv command-line tool. This is the intended mechanism for running the exmergo-dex-core engine and managing its dependencies.
[DYNAMIC_EXECUTION]: The skill dynamically installs and updates its core engine and optional feature extras (such as scikit-learn for clustering or MetricFlow for semantic analysis) using the uv package manager. This process is restricted to pinned versions and official or bundled sources.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data and metadata from warehouses (table names, column descriptions, query results). While this creates an attack surface for indirect prompt injection if the processed data contains malicious instructions, the skill mitigates this by using structured JSON output envelopes and providing specific instructions to the agent on how to parse the results safely. Evidence:
Ingestion points: Metadata and data fetched via inventory, map, profile, and query commands.
Boundary markers: Tool output is encapsulated in structured JSON envelopes.
Capability inventory: Shell execution via uv, file writing to the .dex/ cache, and network operations through established database connectors.
Sanitization: The underlying engine implements PII detection, masking, and cost-guarding logic.

explore