v0.5.1 — Good CItizen

Changelog: 0.5.1

Release Date: 2026-07-10

---

ShiftLefter 0.5.1 makes sl run a first-class citizen of your CI pipeline: JUnit XML your CI ingests natively, tag-based subset selection, live per-scenario output, parallel execution, and a self-contained HTML report — one release, one theme.

v0.5.0 locked the core surface and launched publicly. v0.5.1 is about everything around the run that a pipeline needs: machine-readable results, picking the subset that matters, seeing progress as it happens, and honest wall-clock speed.

---

What's New

JUnit XML for CI

sl run --junit-xml PATH (config mirror [:runner :report :junit-xml]; the flag wins) writes a CI-ingestible JUnit XML report alongside the console/EDN output — verified live against GitLab's artifacts:reports:junit ingest. Test-case identity is stable across runs, failures are attributed through macro expansion to the step that actually failed, and per-scenario and per-step durations are captured.

Two behaviors to know (full detail in ERRATA E009):

• The XML contains a <failure>/<error> iff the run's exit code is nonzero — the report never contradicts the exit code.
• A planning error (exit 2) writes no file at all. Gate CI on the process exit code, not on the report's presence.

Tag-Subset Selection

sl run --tags TAGS / --skip-tags TAGS runs a tagged subset of the suite: comma-separated, repeatable, @ optional; include is OR, exclude wins; feature/Rule/Examples-block tags inherit per the Gherkin spec.

Filtering happens at planning time — deselected scenarios are never bound, and counts, JUnit, and EDN output reflect only the selection. --dry-run with a filter previews it: the console appends ; M filtered out by tags, and the EDN gains an additive :filtered-out key. No filter = behavior unchanged.

Live Per-Scenario Output

sl run now prints each scenario's result as it finishes, instead of staying silent until the whole suite completes. Same lines, same order — you just see them while the run is still going.

Parallel Scenario Execution

sl run --max-parallel N (config mirror [:runner :max-parallel]; the flag wins; default 1 = sequential) runs scenarios on a bounded pool. Report output is identical in content and order to a sequential run at any N.

• @serial marks a scenario as exclusive: it runs alone, after the pool drains. It is an exclusivity mechanism, not an ordering one; a feature-level @serial marks each scenario individually, not the feature as an atomic block.
• Scenarios that wear costumes or touch a shared-impl interface auto-serialize the same way, with a notice line saying so.

Self-Contained HTML Report

sl run --html PATH (config mirror [:runner :report :html]; the flag wins) writes a single-file HTML run report: works from a double-click, no network requests. Failures come expanded by default, with step-by-step transcripts, durations, tag chips, and dark/light rendering.

The report embeds its own run data as a machine-readable EDN island (<script type="application/edn">) — agents read the data, humans read the page, one artifact. Runs with thousands of scenarios build a big DOM and will scroll sluggishly.

CLI Polish & Fixes

• sl --help now lists the daemon subcommands and repl --clj (previously undocumented in the help text).
• sl verify outside a ShiftLefter framework checkout now exits 2 with a notice (was: exit 0 — a consumer CI script could silently pass). Use sl orient and --dry-run to validate a project.
• close-browser handed a wrong-shape capability now returns a structured error instead of silently claiming :closed (both browser adapters).

Internal

• Two-plane reporting architecture: a synchronous Reporter protocol for user output, an observer event bus for everything else — the groundwork the features above stand on. A throwing reporter fails the run immediately.
• Run-event envelopes gain :scenario/id and :seq; the observer bus is offer!-based (a slow observer can never stall or crash a run; drops are counted and logged) and drains on close, so accepted events are always delivered — killing a long-standing race that dropped each run's final events.
• A generated runner-mechanics corpus (seeded generator + JUnit-vs-manifest verifier) now exercises tags × parallel acceptance, and generative round-trip properties lock the envelope purity contract.

Behavior Changes

• sl verify in a directory that isn't a ShiftLefter framework checkout exits 2 (was 0). A CI step that relied on the old silent green must switch to sl orient / sl run --dry-run.
• sl run console output now appears per-scenario during the run (content and order unchanged).

Stats

1710 tests, 5593 assertions, 0 failures
Cucumber compliance: 46/46 good, 11/11 bad (100%)

Installation

One-line installer (recommended):

curl -fsSL https://raw.githubusercontent.com/SHIFT-LEFTER/shiftlefter/main/release/install.sh | bash
export PATH="$PWD/sl:$PATH"
sl --version

Manual: download shiftlefter-v0.5.1.zip from releases, unzip, add to PATH. Java 11+ is the only requirement — no Clojure toolchain needed.

Links

• README
• Fit check — would ShiftLefter work for my project?
• Why it's built this way

v0.5.0

See CHANGELOG.md for the full 0.5.0 entry.