Mise Tasks Reference
This document provides a reference for all available mise tasks in the Morphir Moonbit monorepo.
Pinned compiler and core
The compiler is the stock 0.10.20+65efd5b93 release, and every command uses the
core that ships with it, as consumers of published modules do. That core is the
upstream source commit 156637b5d9149a4d2980bf9522c060f829667e1b with two later
upstream commits applied, 6f8725f3 and 858dc4a4 (the merge of the array core
into builtin/array.mbt); no single upstream commit has the same files. It includes the
fix for long decimal integers on Wasm GC (moonbitlang/core#4345). The release
archives, their SHA-256 values and the complete source digest are recorded in
the core pin. Consumers need moonc 0.10.20 or later.
mise install and entering an activated mise shell complete the toolchain and verify
the source bytes of $MOON_HOME/lib/core against the pin. No core override is used:
the checks refuse a set MOON_CORE_OVERRIDE. In a shell without mise activation, use
mise exec -- moon …. The project selects mise's pinned compiler directory as
MOON_HOME. Setup updates its Mooncakes registry before compiling the portable checker,
including on a fresh installation. Windows first installs and bundles the stock core.
From this toolchain, a successful moon check also rewrites each checked package's
pkg.generated.mbti. Use moon check --no-mbti in this workspace: the SDK source
tree is an audited identity, and generated interface changes would alter it.
The setup does not replace the core in a shared MOON_HOME. An absent or altered
workspace core fails validation. The separate LLVM qualification keeps its own
compiler/core pair.
Task Categories
Linting Tasks
| Task | Description | Platform |
|---|---|---|
mise run lint |
Run all linting tasks | All |
mise run lint:yaml |
Lint YAML files | All |
mise run lint:engine-boundaries |
Refuse host imports in synchronous engine manifests, including tests | All |
mise run lint:moonbit |
Check Moonbit code formatting | All |
Formatting Tasks
| Task | Description | Platform |
|---|---|---|
mise run format |
Format all Moonbit code | All |
Build Tasks
| Task | Description | Platform |
|---|---|---|
mise run build |
Build all packages for all targets | All |
mise run build:wasi |
Build all packages for WASI target | All |
mise run build:browser |
Build all packages for browser (WASM-GC) target | All |
mise run build:js |
Build all packages for JavaScript | All |
mise run build:workbench |
Build the shared Rabbita browser frontend and runtime worker | All |
mise run workbench:artifact-source-check |
Verify the pinned upstream artifact decoder, Unicode data and licenses; also runs before Workbench builds/tests | All |
mise run build:native |
Build native release executables | All |
Test Tasks
| Task | Description | Platform |
|---|---|---|
mise run test |
Run tests for all packages | All |
mise run test:workbench |
Test the local pipeline, connected v1 protocol, loopback proxy and browser workflows | All |
mise run test:workbench-live-host -- --host-bin /absolute/path/to/morphir |
Test authenticated connected v1 workflows against an installed Rust UI host | All |
mise run test:workbench-artifacts |
Test exact generated file and ZIP bytes through a real installed SDK process and the connected UI | All |
mise run test:workbench-artifact-client-mck |
Qualify production artifact client traces in Bun, Node and Chromium against an explicit shared draft28 client kit | All |
mise run test:workbench-native-session |
Test desktop session authority and revocation without CEF | All |
mise run test:workbench-native-workers |
Qualify real owned native workers, exact values, bounded I/O, cancellation and shutdown without CEF | macOS/Linux |
mise run test:workbench-native-client-mck |
Qualify production client draft.17 scenarios in Node/Chromium with debug/release cores and independent report checks | Bun, Node, Chromium and explicit parent runner/kit |
mise run test:workbench-native-owner-mck |
Qualify production owner draft.18 traces in native debug/release with independent report checks | Bun and explicit parent runner/kit |
mise run test:workbench-native-policy-mck |
Qualify exact native checking-policy boundary draft.20 in Node/native and owner draft.22 in native debug/release with independent report checks | Explicit parent runner/kit |
mise run test:workbench-native-policy-client-mck |
Qualify exact native checking-policy client draft.21 in Node/Chromium with debug/release cores and independent report checks | Bun, Node, Chromium and explicit parent runner/kit |
mise run test:workbench-native-workspaces |
Qualify real read-only grants, traversal/link/replacement refusals, original-byte chunks and revocation without CEF | macOS/Linux |
mise run test:workbench-capability-policy-mck |
Opt-in exact draft.12 capability-policy qualification through the shared Rust runner in Node/Chromium/native debug/release; requires explicit runner/kit and Workbench Chromium dependencies | All |
mise run test:workbench-capability-execution-mck |
Opt-in exact draft.19 immutable Scheme execution traces in Node/Chromium/native debug/release; requires explicit parent runner/kit and Workbench Chromium dependencies. Actual isolated-worker acceptance is separate | All |
mise run test:conformance |
Compare versioned independent Ion expectations, live Scheme and explicitly pinned optional Rust; retain separate coverage receipts | All |
mise run test:cli |
Run CLI target and npm package smoke tests | All |
mise run test:cli-extension-elm -- --json |
Qualify the pinned Elm extension executables (v0.4.1, elm-classic v0.5.1 on MEP 0.1 and v0.6.0 on MEP 0.2.0-draft.1) and the Elm v0.5.0-preview.2 WASM module on the worker host from the native and npm CLIs: describe, compile of the Elm fixtures, generate (WASM), and no Node on the native run's PATH. See qualification |
macOS, Linux; manual |
mise run test:cli-extism-python -- --json |
Qualify the pinned Python extension 0.4.1 (one document, multi-document refusal), 0.4.2 (IR 3 and 4 compile, generate, consumer, recompile), and 0.5.0 and 0.6.0 on MEP 0.2.0-draft.1 as WASM and process workers, from the native and npm CLIs; --python PATH names Python 3.12 or later for the consumer |
macOS, Linux, Windows; manual |
mise run test:cli-native-prerequisites -- --json |
Qualify the native MoonBit guest as a process worker from both CLIs with the shared kit's source, consumer and JSON artifacts, and run the shared MEP kit against it | macOS, Linux; manual |
mise run test:portable-mep-cli -- --json |
Qualify the portable MoonBit guest (morphir-bytes-v1) from both CLIs, and run the shared MEP kit against it through the native CLI's worker mode |
macOS, Linux; manual |
mise run test:extension-conformance -- --json |
Run the morphir-extension-sdk fixture guest in apps/morphir/fixtures/sdk-guest from both CLIs: the shared generic MEP kit through the native CLI's worker mode, a comparison with the kit's expected output, imports, claims, compile, generate and host functions |
macOS, Linux; manual |
mise run test:cli-extension-model-line -- --json |
Qualify the pinned avro 0.4.0, openapi 0.5.0 and rust 0.4.0 WASM modules on MEP 0.2.0-draft.1 from the native and npm CLIs: describe, generate at IR 4 from plain IR and from a model envelope, rust compile, IR 3 on the legacy MEP 0.1 fallback with its warning, the refusals, MEP 0.1 in a direct session, same output from both CLIs and exit within a bound. See qualification |
macOS, Linux, Windows; manual |
mise run test:extension-qualification -- --json |
Run all six suites above and write one receipt for each | macOS, Linux; manual |
mise run test:moonbit-frontend |
Compare original Boolean MoonBit, Scheme and generated libraries across four targets in debug and release; requires pinned MOON_HOME and retains identities |
All |
mise run test:supervision |
Verify process-tree cleanup, cancellation, deadlines and bounded diagnostics on Linux, macOS and Windows | All |
mise run test:embedding |
Install embedding dependencies and Chromium, then test Node, installed npm and browser hosts | All |
mise run test:assembly |
Generate a real multi-module Morphir library, assemble its source SDK/pinned core graph, persist binary Ion artifacts and consume them in fresh hosts | Pinned MoonBit, Bun and Node |
mise run test:wasm-hosts |
Opt-in MoonBit component boundary checks in Wasmtime, jco/Node and Chromium; optional pinned native Wasmoon checkout. See scope and evidence | Platforms with the pinned tool releases and Chromium |
Utility Tasks
| Task | Description | Platform |
|---|---|---|
mise run check |
Run all checks (lint + test + build) | All |
mise run check:core |
Verify the toolchain core's source bytes and exact decimal parsing | All |
mise run clean |
Clean build artifacts | All |
mise run validate |
Run all validation checks | All |
mise run validate:packages |
Verify package structure is valid | All |
mise run list-tasks |
List all available mise tasks | All |
Setup Tasks
| Task | Description | Platform |
|---|---|---|
mise run setup:workbench |
Install workbench npm dependencies and Chromium; --ci adds Linux system dependencies |
All |
mise run setup:core |
Verify the pinned toolchain's own core (Windows: install and bundle it first) | All |
mise run setup:wasm-hosts |
Install the optional component qualification dependencies and Chromium | Platforms supported by Playwright |
mise run setup:hooks |
Install git hooks for pre-push validation (idempotent, auto-runs on directory entry) | All |
Site Tasks
| Task | Description | Platform |
|---|---|---|
mise run setup:site |
Install site build dependencies and Chromium for the smoke test | All |
SITE_MAIN=1 mise run build:site |
Build the project site into _site/ (SITE_MAIN=1 fails on draft pages) |
All |
mise run dev:site |
Serve _site/ at http://127.0.0.1:4173/morphir-moonbit/ and rebuild on content changes |
All |
mise run check:site |
Check links, titles, the Atom feed and forbidden paths in the built _site/ |
All |
mise run test:site |
Site unit tests, highlight tests and the browser smoke test against _site/ |
All |
Beads Tasks
| Task | Description | Platform |
|---|---|---|
mise run beads:plan -- <command> |
Store superpowers specs, plans and ledgers in beads and render them back, with bd-plan fetched from finos/morphir at a pinned commit (see AGENTS.md) |
All |
Task Structure
Task names are declared in .config/mise/config.toml. Their orchestration uses
pinned standalone MoonBit scripts under scripts/:
| Script | Task names |
|---|---|
repository-tasks.mbtx |
build, build:wasi, build:browser, build:js, build:native, test, format, lint, lint:yaml, lint:moonbit, validate, clean, list-tasks |
refresh-ir-fixtures.mbtx |
ir:fixtures, including optional --morphir and --morphir-elm checkout paths |
setup-hooks.mbtx / pre-push.mbtx |
setup:hooks and Git pre-push validation |
beads-plan.mbtx |
beads:plan, acquiring the pinned upstream helper with curl and invoking it with Bun |
check.mbtx |
check |
validate-packages.mbtx |
validate:packages |
The Bash/PowerShell task pairs and Python fixture generator have been removed.
Toolchain installation still uses the small bootstrap in tools/; application
compilers and tests retain their own languages. Run mise run test:ci for script
formatting and automation regression checks. The existing qualification setup matrix
also runs those regressions on Linux x64/ARM64 and Windows, and on macOS in the nightly run.
Common Workflows
First-Time Setup
# After cloning the repository
cd morphir-moonbit
mise install
Note: The setup:hooks task runs automatically when you enter the directory (via mise hooks), so git hooks are set up automatically. This is idempotent and can be run multiple times safely.
Pre-Push Validation (Automatic)
The pre-push hook automatically runs before every git push:
- Linting (
mise run lint) - Format Check (
mise run lint:moonbit) - Validation (
mise run validate)
If any check fails, the push is blocked with a clear error message.
Bypass (not recommended): git push --no-verify
Before Committing
# Run all checks
mise run check
Quick Development Cycle
# Make changes...
# Format code
mise run format
# Check formatting and lint
mise run lint
# Run tests
mise run test
Build for Specific Target
# Build only for WASI
mise run build:wasi
# Build only for browser
mise run build:browser
Clean and Rebuild
# Clean all build artifacts
mise run clean
# Rebuild everything
mise run build
CI/CD Integration
The lint, format-check, and configuration-validation workflows delegate to mise
tasks. On Linux x64, Linux ARM64 and Apple Silicon, the mise http:moonbit
postinstall runs the official installer for the pinned release, including the
matching moonbitlang/core standard library and moonx. Local mise install
and CI therefore use the same complete Unix toolchain. CI's setup action also
repairs older cached installations before running moon build / moon test.
Full qualification checks uncached mise installs on Linux x64 and ARM64, and on Apple Silicon in the nightly and release runs.
Main CI Workflow (.github/workflows/ci.yml)
- Lint runs YAML lint once. Validate Configuration Files lists the task configuration and runs
mise run validate:packages. Localmise run validatestill includes YAML lint. - TypeScript installs frozen dependencies, checks all projects and proves invalid types fail checked builds once on Linux.
setup-typescriptacceptsbuild-runtimes: 'false'for static consumers; runtime consumers retain JavaScript emission by default. - Build retains Wasm, Wasm GC and JavaScript release CLI smoke tests and JavaScript Workbench checks. Test retains Wasm, Wasm GC, JavaScript and native workspace tests. Native qualification retains the separate native CLI artifact and MEP process checks.
- Process supervision retains Linux and Windows execution whenever selected, and macOS execution in the nightly and release qualification. Installed and signed-provider qualification retain their platform, trust and debug/release checks.
- ci-ok requires every selected job to succeed and refuses any failed or cancelled dependency. A selected job that is missing or skipped fails the gate.
Component selection
mise run ci:changes uses exact event revisions. The tested rules in scripts/ci-changes.mbtx select each job and produce the aggregate gate's required-job list from the same selection.
| Changed inputs | Independently selected baseline jobs | Other retained checks |
|---|---|---|
| Workbench app, native Workbench packages and Workbench scripts | TypeScript | Build/test backends, native CLI/MEP, installed Workbench, site when its inputs change |
| Workbench sample generator or generated sample inputs | TypeScript, Elm/Scheme freshness | Workbench and backend checks |
| Host embedding directory | TypeScript, embedding including Chromium | Build/test backends, native CLI/MEP; site for documentation changes |
| MoonBit MEP directory or extension app | TypeScript | Build/test backends, native CLI/MEP |
| MoonBit generator, bindings, SDK or frontend | TypeScript, Elm/Scheme | Build/test backends, native CLI/MEP, selected transitive acceptance and site |
| Native telemetry adapter/scripts | TypeScript | Build/test backends, native CLI/MEP, telemetry and site when affected |
| Shared IR, engine, host, tooling, locks or unknown code paths | All baseline jobs | Conservative acceptance dependency rules |
| Documentation or site only | TypeScript when a site input changes | Site, lint, format and package validation |
Shared inputs and combinations union their dependencies. Renames count both the old and new paths, including code moved to documentation. A manual run selects everything. Main pushes use a two-commit diff and PR/merge-queue events use the merge base. An unreachable push base selects everything. The separate Site workflow owns main deployment.
CI policy, shared setup, workspace and mise toolchain changes still exercise full qualification. The known site workflows exercise the site build without native qualification. Further workflow narrowing requires a tested mapping. Uncached Linux x64/ARM64 and macOS toolchain installs, installed OS matrices and exact-candidate release qualification remain in qualification.yml, which has no trigger of its own. Pull request, merge-queue and main runs call it with the macOS jobs skipped. The scheduled macOS runs are in nightly-qualification.yml. It skips the macOS jobs when trusted evidence from a passing nightly run on main in the last 7 days has the same input fingerprint (moonx scripts/ci-evidence.mbtx fingerprint); CONTRIBUTING_DEV.md has the trust rules.
Evidence command output
Use --json with scripts/ci-evidence.mbtx when a program needs its result.
The existing fingerprint, lookup and mode JSON objects keep their schemas
and platform framing. JSON help performs no lookup or file write:
moonx scripts/ci-evidence.mbtx record --help --json
JSON record, help and self-test results use morphir/command-result, version
0.1.0, with command, state, exitCode, diagnostics and either result
or error. A successful record result names output, artifact and the exact
evidence object written to disk. Success is emitted after the write finishes.
These new results are compact UTF-8 JSON followed by one LF on stdout. Notices
and self-test notes use stderr. Expected JSON refusals exit normally with status
1; default text invocations retain their existing runtime failure behavior.
The JSON flag controls presentation and does not suppress the operation's writes
or GitHub reads. Use help to inspect the command before running it.
CI measurements and rollback
Run mise run ci:measure -- RUN_ID --logs > .dev/ci-measurement.json on a completed run. It uses authenticated gh, paginates all jobs in the exact run attempt, and records per-job queue seconds, execution seconds, per-OS runner occupancy, individual step costs and cache log observations. Missing logs mean unknown cache state. GitHub retains successful jobs on retries and resets their creation timestamps. The command marks those results as retained, leaves their original queue delay unknown, and excludes them from the current attempt's execution totals. Effective totals include retained results for workload comparisons. Save each attempt before retrying and sum current-attempt execution totals to measure cumulative runner cost. These seconds are unweighted and are not billed minutes. Queue sums overlap, and run elapsed time includes dependency waits. The API does not expose a complete dependency DAG; use the final jobs and workflow dependencies to inspect the critical path.
Compare the same workload and dependency locks, report warm/cold/unknown cache observations, and distinguish elapsed time from runner occupancy. Do not turn one pair of runs into a general percentage claim. Optimization evidence and the HTML coverage report are stored in beads epic morphir-ukvn.5.65.
Bun/package download caches and additional JavaScript artifact transfers require a measured benefit. Keep frozen installs on misses. Never cache trust decisions, signing state, mutable fixtures or qualification receipts. Existing native artifact admission verifies source, run, target, compiler/core and dependency identity; absent or incompatible artifacts fail closed. The scheduled installation checks remain uncached.
For a material regression, revert the affected optimization commit. Restore broad baseline job conditions and their required-job selection together, or revert the complete selector slice. Keep the TypeScript job ID in ci-ok; restoring its OS matrix does not require a protection change. Keep the runtime-setup default enabled. Reverting these optimizations must preserve the qualification workflow and exact-candidate gate.
Merge queue
Pull requests to main merge through a merge queue. Click Merge when ready (or run
gh pr merge <number> --squash), and GitHub tests the PR on top of main and the PRs ahead of it in the queue
(the merge_group trigger). The PR merges only when ci-ok passes on that combination. When you add a job to
ci.yml, add it to the needs list of ci-ok.
A queue run does not repeat work that the PR's own green run proved. scripts/ci-reuse.mbtx verifies the PR run's
receipt through the GitHub API, and scripts/ci-changes.mbtx then selects only the checks that the commits landed on
main since the receipt's base affect. A PR that changes CI policy (workflows, actions, scripts/ci-*, mise
configuration or the toolchain pin) is trusted only when its queue commit has the tree that its PR run tested: the
same base, and Git's merge of that base and the PR head. Anything that cannot be proven selects the full set as
before. The Detect code changes job logs each decision with the prefix ci-changes: reuse:. To return to logging
only, set reuse_enforced to false in scripts/ci-changes.mbtx.
Branch sync
.github/workflows/branch-sync.yml keeps main here and moonshots in finos/morphir on one commit. The two
branches hold the same tree, so the workflow runs in both repositories. Each run fast-forwards the other
repository's branch to its own branch, but only when its own branch is ahead and ci-ok passed on its tip. Only
that run can push in its direction, so it reads a pending ci-ok again every 30 seconds for up to 30 minutes
(--wait-minutes 30). A schedule here repeats the run every 30 minutes in case a push event was missed.
A commit that lands on moonshots through its merge queue has ci-ok from the merge_group run, and a
migration pushed as an admin has ci-ok from its PR run on the same commit. CI does not run on pushes to
moonshots, because a commit forwarded from main here already passed. A commit pushed to moonshots without a
PR stops the sync with blocked; run gh workflow run ci.yml -R finos/morphir --ref moonshots to give it ci-ok.
The job never merges, rebases or force-pushes. When the branches have diverged, it pushes nothing, fails, and opens
an issue labelled branch-sync; the next run that finds them together closes the issue. To resolve a divergence,
merge one branch into the other through a PR with a merge commit. Land migrations from finos/morphir main on
moonshots the same way, as merge commits, never squashed.
Each repository has a branch-sync environment that only its paired branch can use. Its secret
BRANCH_SYNC_SSH_KEY is the private half of a write deploy key on the other repository, and both rulesets let
deploy keys bypass the merge queue.
Run mise run branch:sync -- --repository finos/morphir-moonbit --dry-run to see the decision without pushing.
With --json, stdout is one morphir/command-result document (command branch.sync) and progress goes to
stderr. The command exits 1 when the branches have diverged or ci-ok failed or is missing on the tip to push.
Adding New Tasks
To add a new task:
- Create a standalone MoonBit script in
scripts/<name>.mbtx. - Pin each import version; use
moonbitlang/asyncfor processes, files and stdio. - Declare the mise task in
.config/mise/config.tomlwithmoon run scripts/<name>.mbtxormoonx scripts/<name>.mbtx. - Update this documentation.
scripts/beads-check.mbtxis the reference example.
Older task categories use bash and PowerShell pairs. Migrate a task to MoonBit
when changing it, as required by .github/AGENTS.md.
For task categories (like lint or build), the main task should be named _default.
Tips
- Use
mise tasksto list all available tasks - Use
mise run <task> --helpfor task-specific help (if implemented) - Tasks can call other tasks using
mise run <task-name> - Environment variables from
.config/mise/config.tomlare available in all tasks
IR fixtures
mise run ir:fixtures regenerates portable MoonBit tests from the vendored IR JSON corpus and formats the workspace. Pass -- --morphir /path/to/morphir --morphir-elm /path/to/morphir-elm to refresh from the pinned commits in local clones. See the IR module for coverage and migration limits.
mise run test runs library tests on wasm, wasm-gc, js and native, followed by the CLI smoke tests.
Workbench
mise run setup:workbench, mise run build:workbench, and mise run test:workbench prepare, build and verify the shared browser frontend. The JavaScript build job runs the browser workflow in Chromium. The optional Proton host has its own workspace to keep CEF out of default builds; see desktop setup.
mise run check:workbench-package-pins checks target, runtime and checksum
refusals without acquiring CEF. mise run build:workbench-packaged -- linux-x64
and mise run build:workbench-packaged -- windows-x64 build only on the named
actual host and architecture. These opt-in commands acquire the pinned Proton
CLI and CEF, verify Linux packaging-tool downloads, build a fresh package, and
record a source-bound build receipt under .dev/packaged-targets/<target>/.
Install the documented platform compiler and CEF/display dependencies first.
The command refuses existing package evidence. A build receipt does not qualify
the CEF browser-worker workflow, OS dialogs or native worker authority.
mise run test:workbench-packaged -- <target> then launches that exact package
and accepts its real CEF browser-worker fallback. Linux runs under xvfb-run -a.
The application acceptance reuses the installed CLI's owned process-group and
Windows Job supervisor for its package fixture. It checks native authority,
strict typed fields/JSON/Ion, standalone mode, real worker exit and real window
closure. The MoonBit aggregate refuses either missing, failed, partial or
wrong-source target receipt. Full qualification requires both host lanes;
macOS native worker and OS-dialog evidence remain separate.
E2 installed lifecycle and native telemetry
setup:llvm: explicitly acquire hash-pinned compiler/core archives into a newMORPHIR_LLVM_HOMEand writeMORPHIR_LLVM_PIN; darwin/arm64 only. Default stable toolchain remains unchanged.test:installed-lifecycle: installed npm acceptance forMORPHIR_REQUIRED_TARGETS(defaultjs,wasm-gc,native), JSON/Ion text/binary, debug/release, output modes, local sinks and deliberate mismatch rejection. Includes installed MoonBit source lowering to Ion and four Boolean rows against original compiler, Scheme and generated execution on each required lane. Source receipts and instrumentation-neutrality evidence are inmoonbit-source/. Installed library receipts are inmoonbit-library/. Naming qualification inmoonbit-naming/compares 20 independently authored source rows and three pre-change checkpoint rows per target/mode against the original compiler, Scheme and generated execution. Direct consumers verify 26 rows per lane in readable and legacy modes, including exact target requests through a simulated explicit-capability host. Checks cover quiet structured warnings, both Ion resumptions, opaque metadata, JSON loss guards and preserved publication after strict or unavailable-provider failure. That lifecycle task uses simulated provider capabilities. The separatetest:installed-providertask exercises real signed restoration. For focused diagnosis, runnode apps/morphir/scripts/test-installed-lifecycle.ts --naming-onlywith the same environment. The required task always runs the complete lifecycle. SetMOON_HOME; LLVM additionally requires home/pin. SetMORPHIR_INSTALLED_RECEIPTSto retain evidence.setup:native-otel: build the optional host adapter from every accepted archive hash, outside the portable workspace. SetMOON_HOMEand optionallyMORPHIR_OTEL_OUTPUT. Requires Node and Python 3 for verified archive setup.test:native-telemetry: installed CLI OTLP HTTP/JSON capture, correlation, metrics, flush failure and paired overhead baseline. RequiresMORPHIR_OTEL_BINARY,MORPHIR_OTEL_PIN,MOON_HOME; setMORPHIR_TELEMETRY_RECEIPTSfor artifacts. Runs separately from portable tests.
These tasks use standalone MoonBit scripts. The native adapter and supported hosts/transports are documented in its guide.
test:installed-provider qualifies the packed npm CLI and native executable
against a clean, explicit MORPHIR_PROVIDER_RUST checkout at the pinned shared
revision. Set MOON_HOME; optionally set MORPHIR_PROVIDER_RECEIPTS for binary
Ion receipts and their JSON inspection view. It provisions fresh disposable test
trust per case and verifies independent generated API calls, authentication
refusals, context boundaries, publication preservation and audit evidence.
See provider operations
for setup, exact scope and remaining qualification work.
mise run test:workbench-live-host -- --host-bin /absolute/path/to/morphir verifies single-use launch authentication, native Gleam compilation/generation and workspace-model opening through the loopback proxy against a real Rust host. Supply a host with native Gleam available; see Workbench verification for setup and version evidence. This optional check is separate from the default test suite.
TypeScript qualification
mise run check:typescript installs every Bun dependency root from its frozen
lockfile, checks all four TypeScript projects, runs the intentional type-error
and environment-isolation gates, and emits portable runtime JavaScript. Existing
Workbench, CLI, compiler, supervision and embedding tasks use the same portable
scripts/typescript-tooling.mbtx entry point. Append --ci to setup tasks when
Chromium system dependencies must also be installed.
Engine dependency boundary
mise run lint:engine-boundaries checks engine module dependencies and package
imports, including black-box and white-box test imports. It parses both native
and legacy JSON manifests, skips nested modules and generated dependencies, and
fails on hosting, async, or Morphir host adapter imports. Pure Moon-Lander
configuration, logging, mediator and notebook packages remain available.
mise run lint:engine-boundaries -- --json
moonx scripts/engine-boundaries.mbtx --root . --json
moonx scripts/engine-boundaries.mbtx --self-test --json
moonx scripts/engine-boundaries.mbtx --help --json
JSON mode writes one compact morphir/command-result document, contract version
0.1.0, with command lint.engine-boundaries to stdout. Progress and human
failure details use stderr. Success includes state: "succeeded", exitCode: 0,
diagnostics: [] and result: {manifestsChecked, importsChecked}. Help replaces
the result with {help}; self-tests use {checks} and exercise actual command
invocations as well as manifest fixtures. Failure has state: "failed",
exitCode, error: {code, message}, and diagnostics, without a result.
Diagnostics have code, severity, path and message. Boundary codes are
engine.host_import, engine.manifest_syntax and engine.missing_root.
Exit 1 reports a refused boundary or inspection failure; exit 2 reports invalid
arguments. Top-level error codes are engine_boundary_failed,
inspection_failed and invalid_arguments respectively. Default text mode
retains those statuses. --root owns its next value; a literal -- ends option
parsing. The lint task includes this check; test:ci includes its self-tests.