Impulse AI Assistant Skills

Impulse provides agent skills that can be used with Databricks Genie Code or any tool that follows the Agent Skills open standard. Skills teach AI assistants how to use Impulse — the Databricks Labs library for analyzing large-scale time-series measurement data on Spark and Delta.

Skills

Each skill is a folder with a SKILL.md file that documents usage patterns. Start with impulse, which explains the core concepts and routes to the right skill for the task.

Skill	What it covers
impulse	Entry point. Core concepts (container, channel, event, aggregation), the three usage modes, setup, and a decision tree to the other skills.
impulse-tsal	The Time Series Analytics Language DSL — selecting channels, deriving virtual signals, and the four result types (SampleSeries, Intervals, PointsInTime, PointsInTimeSeries).
impulse-data-model	The silver-layer input tables Impulse reads, the gold-layer star schema it writes, landing your own data, and adapting an existing layout via column mappings.
impulse-config	The ImpulseConfig schema — source tables, sink, container filters, solver options, incremental processing, and sinkless mode.
impulse-events	Defining event windows: BasicEvent, ContainerEvent, SequenceOfEvents, PointsInTimeEvent, TimeWindowEvent.
impulse-aggregations	Computing results over channels: 1D/2D histograms (duration/distance/custom-weight), StatsAggregator, PointValueAggregator, and pages.
impulse-channels	Calculated (derived) channels — materializing a new signal from existing channels via CalculatedChannel and solve_calculated_channels.
impulse-reporting	The batch pipeline that persists events and aggregations to the gold-layer star schema with Report / Page, plus incremental runs.
impulse-analyze	Ad-hoc analysis — evaluating TSAL directly through the query engine and returning Spark or pandas DataFrames, no gold-layer write.
impulse-ml	Extracting event-scoped statistics as a flat feature matrix for MLflow / AutoML.

Install the skills

These skills are not discovered from this skills/ folder directly — an agent only reads them from its configured skills directory. Copy the skill folders into that directory once; do it again whenever you change a skill (the copy is a snapshot, not a live link back to this repo).

Databricks Genie Code

Genie Code discovers skills from a .assistant/skills/ directory. Choose one:

• User-level (just you): /Workspace/Users/<your-email>/.assistant/skills
• Workspace-wide (everyone in the workspace): /Workspace/.assistant/skills

Import this folder with the Databricks CLI, run from the repo root:

# user-level
databricks workspace import-dir skills /Workspace/Users/<your-email>/.assistant/skills

# or workspace-wide
databricks workspace import-dir skills /Workspace/.assistant/skills

Genie picks the skills up the next time you use it. Confirm with a prompt like "List the Impulse skills you can use."

Claude Code

Copy the skill folders to ~/.claude/skills/ on your machine, keeping each SKILL.md at the top level of its folder (not nested deeper):

cp -R skills/impulse* ~/.claude/skills/

Claude Code also reads the .claude-plugin/plugin.json in this folder, so the family can alternatively be installed as a plugin. Genie Code ignores that file and reads each SKILL.md directly.

Other Agent Skills tools

Place the skill folders in the tool's configured skills directory, keep SKILL.md at each folder's root, and restart the tool if it caches skills on startup.

Installing the Impulse library itself

The skills teach an agent to use Impulse; they do not install it. Inside a Databricks notebook or job (with an active spark session and a Databricks SDK WorkspaceClient), install the library one of two ways:

• Wheel — %pip install databricks-impulse[local-dev] (the local-dev extra pulls in pydantic, scipy, and the other libraries otherwise assumed pre-installed on Databricks Serverless / DBR ML).
• Git folder — clone this repository into a Databricks Git folder and add its src/ directory to sys.path, as the demo notebooks do.

Impulse requires Python 3.12 (Serverless Environment Version 2+), PySpark 4.0, and Delta Lake 4.0. See impulse for the full setup snippet.

Scope and guardrails

These skills are scoped to Impulse's public API — the classes and config a user imports and calls. They do not document internal solver stages or private helpers. Every example is self-contained and uses only the framework's public primitives.