QuantLake
GitHub
Documentation

AI agents

Install the Agent Skill so that coding agents can find, read and request data through the CLI.

QuantLake comes with an Agent Skill that teaches coding agents, such as Claude Code and Codex, how to use the lake.

Install the skill

npx skills@latest add supermomonga/quantlake --global

Alternatively, copy the skills/quantlake folder of the repository into your agent's skills folder, for example ~/.claude/skills/.

What an agent can do

With the skill installed, an agent can answer questions such as "Which files do I need to read BTCUSDT trades for October in DuckDB?". The skill tells it to:

  1. Check what is saved with quantlake saved --json.
  2. Get the file paths and the wildcard pattern with quantlake files --json.
  3. Read the files with DuckDB or Polars, filtering on the date column.
  4. If the data is missing, check what the plan includes with quantlake available and register a target with quantlake targets add.

It can also delete data and ask for downloads to be repeated, all through the CLI. See JSON output for the fields that agents read.

What agents cannot do

Agents cannot edit the data files directly. Any write to the lake fails with Operation not permitted. A note in the lake folder (AGENTS.md) and the skill both explain why, and tell the agent which command to use instead.

The skill also tells agents:

  • not to unlock the files with chattr or sudo, and not to edit metadata.db;
  • not to read or decrypt secrets, and to leave setting API keys to you;
  • not to change the Databento budget. If paid data is needed, the agent reports the over_budget days and leaves the decision to you;
  • to check free disk space before registering large order book downloads.