CLI Reference

The entropy command exposes the full framework. Agents and datasets are referenced as module:attr (e.g. myproj.agents:agent) and file paths respectively.

Commands

Command	Description
`entropy init`	Scaffold a project (agent, dataset, `entropy.toml`).
`entropy run`	Run the Monte Carlo suite.
`entropy test`	Regression test against a stored baseline.
`entropy benchmark`	Compare multiple agent variants.
`entropy compare`	Diff two result JSON files.
`entropy replay`	Replay a saved trace / results file.
`entropy dataset`	Inspect a dataset file.
`entropy doctor`	Check optional dependencies & adapters.
`entropy simulate`	Run a simulator (user / env / adversarial).
`entropy chaos`	Run fault injection.
`entropy ui`	Build / serve the dashboard.
`entropy watch`	Live observability over rounds.
`entropy report`	Export a report from results JSON.

Examples

# Run the suite
entropy run --agent m:agent --dataset d.json --trials 100 --seed 42

# Regression gate (writes baseline with --update first run)
entropy test --agent m:agent --dataset d.json --baseline base.json [--update]

# Compare agent variants
entropy benchmark --agents m:a m:b --dataset d.json

# Fault injection
entropy chaos --agent m:agent --dataset d.json --faults api_fail,timeout

# Build & serve the dashboard
entropy ui --agent m:agent --dataset d.json --serve

# Live watching with a heatmap
entropy watch --agent m:agent --dataset d.json --rounds 10 --heatmap h.png

# Export a report
entropy report --results r.json --format html --out report.html

Common flags

--agent — module:attr callable.
--dataset — path to a JSON / JSONL / YAML / CSV dataset.
--trials — number of Monte Carlo runs per case (default 100).
--seed — RNG seed for reproducibility.
--out — write results to a file.

Scaffold then run

entropy init creates agent.py and dataset.json; edit them, then entropy run --agent agent:agent --dataset dataset.json.