agentLoopBenchmark

#!/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.PointerChasingBenchmark;
import airhacks.zsmith.tools.control.WriteClipboardTool;
import airhacks.zsmith.tui.control.CommandLine;

void main(String... args) {

    var depth = CommandLine.firstInt(args, 50);

    var benchmark = new PointerChasingBenchmark(depth);

    // The walk needs one agent turn per hop (the hops are serially dependent, so at most one
    // tool call resolves per turn) plus one turn to emit the answer: depth + 1 at minimum.
    // The extra slack absorbs stray turns — a "thinking" turn with no tool call, or a retry
    // after an unknown-key error — so an occasional wasted turn does not trip a false FAIL.
    var maxIterations = depth + 20;

    var agent = new Agent("pointer-chaser", 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; failure details go to stderr so they never end up in the table
    var row = result.markdownRow(agent.modelName());
    IO.println(row);
    System.err.println(WriteClipboardTool.write(row));
    var diagnostics = result.diagnostics();
    if (!diagnostics.isEmpty()) {
        System.err.println(diagnostics);
    }
}