| name | benchbox-workflow |
| description | Use when the user asks to "test TPC-H", "check compliance", "review architecture", "run quality checks", "check binaries", "test dialect translation", "compare implementations", "run live platform tests", or "plan and execute" a benchmark feature. |
| version | 0.2.0 |
| tools | Bash, Read, Write, Edit, Glob, Grep, Task |
BenchBox Workflow
Route benchmark, platform, quality, and architecture work to one action.
Before acting, read the file in the Read column for the selected action. Auto-
detect the runner, honor non-interactive mode, and produce human plus JSON/file
artifacts when supported.
Actions
| Action | Read |
|---|
test | references/test.md |
quality | references/quality.md |
compliance | references/compliance.md |
dialect | references/dialect.md |
binary | references/binary.md |
compare | references/compare.md |
live | references/live.md |
architecture | references/architecture.md |
plan | references/plan.md |
Aliases remain accepted: benchmark-test, quality-check, qa,
tpc-compliance, dialect-translation, sql, binary-check, binaries,
compare-impl, live-test, platform-test, arch, arch-review,
plan-execute, and implement.
Hard rules
- Use
uv run -- for Python; Makefile wrappers are fine.
- Propagate phases via
--phases: generate, load, power, throughput,
and maintenance. Default smoke scale is 0.01; scale >=1 is a whole
integer.
- Do not run live cloud tests without explicit approval and credentials.
- Use
benchbox.utils.clock.mono_time() / elapsed_seconds() for durations.
- Register adapter DDL rewrites under
benchbox/sql_compat/rules/ddl_optimize/.
- UAT/long-run sweeps follow AGENTS.md "Long-Running UAT" and
docs/operations/uat-framework.md.
Output
Report commands, pass/fail, artifacts, key findings, and next steps. For
failures include a root-cause hypothesis and the narrowest next check.