Run one of the supported recipes against the shared, content-pinned
Terminal-Bench 2.0 subset. The launcher requires Bash, Python 3.12+,
uv, Git 2.25+, and a running Docker daemon.
git clone https://github.com/simple-agent-lab/RSIHub.git
cd RSIHub
# API authentication is the default. Keep credentials out of recipe YAML.
cat > .env <<'EOF'
OPENAI_API_KEY=replace-me
# OPENAI_BASE_URL=https://your-openai-compatible-endpoint/v1
EOF
docker infoChoose a recipe, download and verify the pinned dataset, build that recipe's
pinned mutate-runner image (skipped for hyperagents_dsh, which is dataset-only
because mutate is runner: local), and launch one generation:
RECIPE=ahe
./scripts/setup_terminal_bench.sh "$RECIPE"
./scripts/run_recipe_demo.sh "$RECIPE"Supported values are aevolve, ahe, ahe_codex, gepa, hill_climb,
hill_climb_codex, hyperagents, hyperagents_codex, and
hyperagents_dsh (see its
recipe README
for the extra dsh SDK setup and the pre-experiment checklist: model pin / gen/0
retag, expected_trials vs tasks_per_round, non-TTY docker-exec doctor probe,
and Harbor vs DSH_TASK_TIMEOUT_SEC). Codex-capable
profiles may use CODEX_AUTH_JSON_PATH=/absolute/path/to/auth.json instead of
an API key. Use WORKSPACE, TASKS, GENERATIONS, ENV_FILE, or
EVOLVE_ASSET_DIR to override launcher defaults. See the
recipe guide and
operations guide for the full configuration and
recovery workflow.
For evolution results on Terminal Bench 2 and Tau³ Banking, see the benchmark results in the README.