runcomposer

runcomposer is an open-source, tag-based test run composer & orchestrator: see your test corpus through a curated tag taxonomy, compose precise selections with a real filter language, and turn a selection into a reproducible, portable run spec that any executor can fulfill — with results flowing back from any transport into a run history that feeds new selections ("rerun what failed").

stochasticentropy.github.io/runcomposer — how it works, in one page.

Or watch it

• The 69-second teaser — narrated: why precise selection is awkward, and what changes when the selection becomes a document.
• The 6-minute explainer — seven chapters: the catalog, composing a selection, freezing it into a spec, the document itself, the three ways to execute it, results returning from any transport, and rerunning what failed.

Version 0.1.0. Read ADOPTING.md to connect it to your own corpus and machines, DESIGN.md for the architecture and the reasoning behind it, or CONTRIBUTING.md to work on it.

Quickstart

pip install runcomposer

runcomposer demo     # boot the neutral web-shop demo end-to-end
runcomposer serve    # web UI (EN/DE) + API at http://127.0.0.1:8100
# or: docker build -t runcomposer . && docker run -p 8100:8100 runcomposer

demo seeds a real store — five completed runs with per-item verdicts — into ./runcomposer-demo/, and prints the commands that read it back, including "rerun what failed". Everything it writes is in that one directory, so rm -rf runcomposer-demo undoes it.

What it does

Catalog. A test source enumerates your tests as items — an opaque stable id plus tags — and content-hashes the catalog. Two ship: manifest (a plain JSON or YAML list, zero dependencies, the adoption path for any framework) and robotframework (walks .robot files, ids are longnames).

Compose. Navigate a curated taxonomy, build a filter, watch the preview recompile. The filter language is small and lossless: a bare word is a literal tag, prefix:Checkout- is sugar for an anchored regex, regex: is the escape hatch, and the operators are AND, OR, NOT. Available in the web UI and from the CLI.

Freeze. The filter is compiled against the catalog snapshot and the resulting item list is written into a versioned run spec document — so a dispatched spec is self-sufficient, and the snapshot makes corpus drift detectable. Core sections are generic and closed; exactly one section, runner, is open, and the core never looks inside it. runcomposer validate checks a document against the published JSON Schema.

Execute — three ways, one document. In-process on a robot-pool (partition fan-out, duration-balanced chunking, listener-streamed live verdicts, drift refusal); on your own agent anywhere via runcomposer-exec; or by triggering an existing parameterized CI job with ci-trigger.

Ingest. Results return over a token-guarded HTTP push, a watched file-drop directory, or runcomposer ingest on the command line. Redelivery rules are explicit — a byte-identical bundle is a no-op, a different bundle replaces that shard. A bundle whose marker matches no dispatched run lands in a visible quarantine inbox rather than quietly entering history.

Reuse. Once runs accrue, history becomes a selection source: runcomposer runs --failed-in latest --label suite=nightly, spec --from-history 'failed@latest?suite=nightly', and a UI quick-pick. The label scope keeps "latest" from meaning somebody else's run, and the resolved reference run is recorded in selection.derived_from.

Hand off. runcomposer export <run> --format ctrf for tools that speak CTRF; robot-output-xml and junit-xml parse results coming the other way (both refuse documents carrying entity or DTD declarations).

The web UI ships pre-built inside the wheel in English and German, so evaluating it needs no Node toolchain. Persistence is sqlite by default, and runcomposer gc keeps runs, quarantine and artifacts bounded.

The export round trip

Compose a run spec, execute it anywhere with the vendorable single-file consumer, and ingest the results bundle back — no coupling between composer and executor beyond the spec document itself:

runcomposer spec 'Regression' --title "Nightly" --format json -o spec.json --export
runcomposer-exec spec.json --out results --simulate   # or --command "your-runner {ids_file}"
runcomposer ingest results                            # marker-correlated, idempotent
runcomposer runs                                      # → COMPLETE (PASS/FAIL)

runcomposer-exec is a single stdlib-only Python file — copy it next to any executor (CI checkout, air-gapped host) and it renders the spec's materialized item list, runs your command, and writes the runcomposer_run.json correlation marker beside the results.

examples/remote-agent turns that into a complete adopter kit for the remote round trip: a documented config, an agent that needs only python3 + robot on the executing machine, and a transport-agnostic driver whose local-directory default runs the whole loop — compose, carry, execute, carry back, ingest — on one machine. It is the neutral template a private adopter package copies.

Examples

Executing Robot Framework in-process needs one extra:

pip install "runcomposer[robot]"

Developing

pip install -e ".[dev]" && pytest         # Python 3.10–3.13
cd ui && npm ci && npm run dev            # UI dev server (proxies to :8100)
npm run build                             # rebuild src/runcomposer/ui_dist

See CONTRIBUTING.md for the boundaries that matter, and SECURITY.md to report a vulnerability.

License: MIT.