Examples¤

Every script under examples/ is a runnable skill. Start with deterministic examples, then move to model leaves, then tool-using react leaves.

uv run jdsl run examples/<name>.py
uv run jdsl run examples/<name>.py -i key=value

Example	What it teaches	Needs a key
`greeter.py`	Minimum tree: `root -> seq -> act`.	no
`gate.py`	`optional` and `invert` decorators with deterministic access control.	no
`triage.py`	`predict` writes a category and `sel` branches on it.	yes
`pipeline.py`	Multi-output `predict`, guarded routing, then a second `predict`.	yes
`reason.py`	Two `predict` leaves: reasoning first, final answer second.	yes
`refine.py`	`repeat` loops critique and revision until a guard passes.	yes
`react.py`	Model-driven tool calling with native function calls.	yes
`trip.py`	A compact `react` example for chained calculations.	yes
`shop.py`	Tool-heavy ordering flow with search, comparison, and arithmetic.	yes
`db.py`	Schema discovery and array arguments in `react`.	yes
`wiki.py`	Search, model selection, and later tool call wired by `ref`.	yes

Choosing an Example¤

Use greeter.py or gate.py when you are learning tree semantics. Use triage.py when you want to see a model make a local decision that deterministic tree code consumes. Use trip.py, shop.py, or db.py when the model should choose and chain tools inside one leaf.

Harness examples live under examples/harness/:

File	Purpose
`retail_mcp_server.py`	A tiny MCP server that produces structured retail traces.
`retail_tools.py`	Host-tool bindings for running the sample compiled package.
`retail.jdsl`	A compiled behavior-package fixture.

The harness fixture is intentionally structured: customer and order ids are discrete fields, so the compiler can prove exact value flow instead of scraping text.