agentErrorRecoveryBenchmark

#!/usr/bin/java --enable-native-access=ALL-UNNAMED -Dtools.permissions.default=allow --class-path=../zsmith/zbo/zsmith.jar:../zsmith/zbo/lightmetal.jar --source 25

import airhacks.zsmith.agent.boundary.Agent;
import airhacks.zsmith.benchmark.boundary.ErrorRecoveryBenchmark;
import airhacks.zsmith.tools.control.WriteClipboardTool;
import airhacks.zsmith.tui.control.CommandLine;

void main(String... args) {

    var faults = CommandLine.firstInt(args, 3);

    var benchmark = new ErrorRecoveryBenchmark(faults);

    // The serial walk needs one turn per hop plus one retry turn per injected fault plus one
    // turn to emit the answer. The extra slack absorbs stray turns — a "thinking" turn with
    // no tool call, or an extra retry — so an occasional wasted turn does not trip a false FAIL.
    var maxIterations = benchmark.depth() + faults + 20;

    var agent = new Agent("error-recoverer", benchmark.systemPrompt())
            .withTool(benchmark.tool())
            .withMaxIterations(maxIterations);

    var result = benchmark.score(agent.chat("go"));

    // one normalized markdown row on stdout and in the clipboard — ready to paste into the
    // summary table in README.md; the recovery count goes to stderr so it never ends up in the table
    var row = result.markdownRow(agent.modelName());
    IO.println(row);
    System.err.println(WriteClipboardTool.write(row));
    System.err.println(result.diagnostics());
}