The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external market data from the pandadata-api to maintain local Parquet and DuckDB datasets, creating a vulnerability surface for indirect prompt injection via untrusted data sources. However, the skill provides strong defensive measures to mitigate this risk.
Ingestion points: Data enters the agent's context through external API responses defined by the pandadata-api dependency mentioned in SKILL.md.
Boundary markers: The references/warehouse-playbook.md establishes a mandatory validation checklist, including sample comparisons against fresh API results, to identify discrepancies.
Capability inventory: The skill utilizes scripts/market_bar_envelope.py for strictly controlled data conversion and file writing to the ~/.pandadata/warehouse/ directory, alongside DuckDB for local SQL queries.
Sanitization: The scripts/market_bar_envelope.py script implements exhaustive JSON validation, including type checking, finite number verification, and logic checks for OHLCV (Open-High-Low-Close-Volume) data consistency. It specifically rejects non-standard JSON constants such as NaN or Infinity to prevent parsing exploits.

pandadata-warehouse