scripts/dev/run_circle_packing_persistent_smoke.sh

#!/bin/bash
# Smoke test: Circle packing with eval service and persistent agents
#
# Prerequisites:
#   OPENHANDS_LOG_COMPLETIONS=1 ENABLE_FULL_TRAJECTORY_LOG=1 \
#   uv run eval_agent/ev2_service_standalone.py --host "0.0.0.0" --port 8733

set -e

EXPERIMENT_NAME_PREFIX="circle_packing_persistent_smoke"
NUM_GENERATIONS=20
MAX_PARALLEL_JOBS=4
META_INTERVAL=10
PERSISTENT_AGENTS="--persistent-agents"
PERSISTENT_CONTEXT_REFRESH_INTERVAL=10

LLM_MODELS="native-gemini-3-flash-preview"
LLM_SELECTION="ucb1"
LLM_TEMPERATURES="0.7 1.0"
LLM_MAX_TOKENS=65536

USE_EVAL_SERVICE="--use-eval-service"
EVAL_SERVICE_URL="http://localhost:8733"
EVAL_TRIGGER_MODE="periodic"
EVAL_TRIGGER_INTERVAL=5

EXPERIMENT_NAME=${EXPERIMENT_NAME_PREFIX}_g${NUM_GENERATIONS}_${EVAL_TRIGGER_MODE}${EVAL_TRIGGER_INTERVAL}

.venv/bin/python tasks/circle_packing/run_experiment.py \
    --experiment-name "$EXPERIMENT_NAME" \
    --num-generations "$NUM_GENERATIONS" \
    --max-parallel-jobs "$MAX_PARALLEL_JOBS" \
    --meta-interval "$META_INTERVAL" \
    $PERSISTENT_AGENTS \
    --persistent-context-refresh-interval "$PERSISTENT_CONTEXT_REFRESH_INTERVAL" \
    --llm-models $LLM_MODELS \
    --llm-selection "$LLM_SELECTION" \
    --llm-temperatures $LLM_TEMPERATURES \
    --llm-max-tokens "$LLM_MAX_TOKENS" \
    $USE_EVAL_SERVICE \
    --eval-service-url "$EVAL_SERVICE_URL" \
    --eval-trigger-mode "$EVAL_TRIGGER_MODE" \
    --eval-trigger-interval "$EVAL_TRIGGER_INTERVAL" \
    --verbose \
    --trajectory-log