docs/agent_examples.md

# Agent prompt examples

These are starting prompts that work well with the simulation assistant.
The agent will use its tools (read docs, list examples, submit runs, start
optimizations) to act on the request.

## Quick smoke test

> Run the test_100 example and tell me when it finishes.

## Set up a Young-Laplace validation

> I want to validate surface tension. Run the 01_Young-laplace case for
> 1000 production steps with images on every 100 steps. Once it's done,
> tell me what the final pressure was.

## Single-parameter sweep

> Find the value of dh in [0.18, 0.25] (step 0.01) that minimizes the
> absolute final pressure for test_100. Use 8 trials and TPE.

## Multi-objective trade-off

> Optimize test_100 for two objectives: minimize abs(Press_last) and
> minimize Temp stability (std-dev of the second half). Vary dh in
> [0.18, 0.25] and sigmao in [0.5, 2.0]. Run 12 trials with NSGA-II.
> Show me the Pareto front when done.

## Resolution study

> Run the same example (01_Young-laplace, prodRun=2000) at three different
> values of dh: 0.15, 0.20, 0.25. Tell me which one converges to the
> tightest pressure value.

## Inspect a finished run

> What were the parameters of run abc12345? Show me the thermo trace and
> tell me whether it converged.

## Diagnose a failure

> Run 8a3f4e1c failed. Look at its log and tell me what went wrong.

## Compare two runs

> Compare runs abc12345 and def67890. What parameters differ between them?
> Which one has a more stable pressure trace?

## Set up your project context

> I'm studying droplet break-up under shear. Save in your memory that I
> usually use prodRun=20000, dh=0.2, sigmao=1.0 unless I tell you
> otherwise. Also remember that I want images every 1000 steps by default.

## Educational

> Explain what fluid_type does in this code, with reference to the
> documentation.

## Working on a paper

> Reproduce the parameters from Figure 1(a) of the BCAM paper for a
> droplet of r = 25·dh. Don't run yet — just tell me which example to
> use and what parameters to override.