pytest-obligation 0.8.0__tar.gz

pytest_obligation-0.8.0/.agents/plugins/marketplace.json

@@ -0,0 +1,20 @@
+{
+  "name": "pytest-obligation",
+  "interface": {
+    "displayName": "pytest-obligation"
+  },
+  "plugins": [
+    {
+      "name": "pytest-obligation",
+      "source": {
+        "source": "local",
+        "path": "../../plugins/pytest-obligation"
+      },
+      "policy": {
+        "installation": "AVAILABLE",
+        "authentication": "ON_INSTALL"
+      },
+      "category": "Developer Tools"
+    }
+  ]
+}

pytest_obligation-0.8.0/.claude/skills/find-upstream-weaknesses/SKILL.md

@@ -0,0 +1,91 @@
+---
+name: find-upstream-weaknesses
+description: Use to probe a real open-source project (an application first, or a framework) with due-work-harness, find background work that can be lost, repeated or misrecorded, and disclose it upstream as issues plus a PR that carries the harness contract. Triggers on "find problems in <project>", "probe <project>", "pick another app or library to test", "add a demo", "open an issue and a PR upstream", "disclose this finding", "adopt the harness in <project>", or any request to repeat the find-and-disclose cycle. Not for adopting the harness in the user's own repository (see ADOPTING.md).
+---
+
+# Find weaknesses in a real project, then disclose them
+
+The full method is [docs/upstream-playbook.md](../../../docs/upstream-playbook.md); read the step you are
+on when you reach it. The copy-and-fill templates are in
+[docs/upstream-templates/](../../../docs/upstream-templates/). This file is the runbook: what to do, in
+order, and the rules that are not negotiable.
+
+The installable plugin's [probe-upstream-due-work](../../../plugins/pytest-obligation/skills/probe-upstream-due-work/SKILL.md)
+skill is the portable route for people working outside this repository. This repository-local skill
+retains the maintainer-specific release and disclosure workflow below; neither skill replaces
+gigaverse-backend's own due-work runtime and conformance instructions.
+
+## Rules that do not bend
+
+- **Expose, don't fix.** Deliver a test that fails for a stated reason and a contract that finds it. Never
+  propose or write the fix in an issue or PR.
+- **Every adoption PR MUST carry `ObligationContract` and a collected
+  `@due_work_contract_suite(CONTRACT)` class.** Follow the
+  [required adoption shape](../../../ADOPTING.md#required-adoption-shape): declare
+  all A–J profiles for the current source (or the installed release's matching catalog), put histories in the contract, execute the generated
+  cases, and link the declaration, suite and `--due-work-summary` report. Ordinary
+  tests and direct crash-helper calls are supporting evidence, never a substitute.
+  The decorator generates pytest cases, not Python files; use `--junitxml` when a
+  generated result file is needed and retain it as a CI artifact.
+- **Claim only what you ran.** A sentence in a report is either measured or marked as code reading. Agents'
+  findings (including your own from an earlier turn) are hypotheses until step 5.
+- **Posting is outward-facing.** Issues, PRs, comments and releases go out under the user's account.
+  Do it only with the user's authorization for this project in this session; if unsure, ask once, with
+  the draft. Follow the target's own contribution rules over these: some ask for a human-written PR
+  description or forbid AI-authored contributions, and then the user decides.
+- **Every comment you post opens with the self-identification line**
+  `> I'm {model}, AI can make mistakes. {a fresh 3-5 word joke}` (no joke on incident threads).
+- **Every probe strengthens the harness.** A gap the probe hit becomes a general capability in the harness,
+  released before the upstream PRs pin it.
+
+## The cycle
+
+Work in a scratch directory, not in a checkout that matters. Track the steps as tasks.
+
+1. **Choose.** Prefer an application whose failure a customer or money feels; take a framework when apps
+   depend on it and nobody has probed it. Count handoffs with `uv run due-work-harness check` (or the
+   `scan` API), require that the project already recovers some of them, that its own tests run on this
+   machine (Python floor, database, system packages), and that outside contributions are accepted. Read its
+   CONTRIBUTING, templates and AI policy now. Search its tracker for prior reports. If two targets fit,
+   pick one and say why; ask only if the choice needs the user's judgement (policy, risk to a relationship).
+2. **Set up.** Fork and clone; put Postgres/Redis in containers the way their CI does; **run their
+   unmodified suite first** so a red baseline is known before it is blamed on you.
+3. **Map.** `due-work-harness sites` and `due-work-harness in-transaction` list the candidates first
+   (a throwaway `[tool.due-work-harness]` table in the fork is enough). Then run the prompt in
+   [map-the-handoffs.md](../../../docs/upstream-templates/map-the-handoffs.md) through an agent (fill the
+   placeholders from the scan). Keep the result as ranked hypotheses.
+4. **Adopt in the fork.** A branch such as `test/<what-breaks>`, a `tests/due_work/` directory named the way
+   their runner collects it, a host (`django_host`, `redis_host`, or a new one), bindings that call the
+   real code (a string command that reaches no project code is refused: pass the command class), a
+   disposition for every profile with a reason, and one history per user-facing action whose observation
+   includes what users and external systems saw (not something the project's own cleanup deletes). A
+   management-command recovery is `management_command(name, after=...)`. Use `--due-work-summary` to read
+   the run.
+5. **Confirm each finding.** Reproduce the smallest version with a script that uses the project's own API
+   and no harness. Test the neighbouring claim before you make it, check server or library versions, and
+   search the tracker again. Drop what you cannot reproduce or what the project defends.
+6. **Declare and mark.** `Findings(delivered, {label: outcome})` on each history (generate it with
+   `--due-work-record-findings`, read it, then paste), `handoff_gaps` with the
+   issue link, an `expectedFailure` plain test that fails for the stated reason (remove the marker once and
+   read the failure), and the *this is where the magic happens* comment above each decorated class.
+7. **Improve the harness.** If the probe needed something the harness lacks, or you wrote a helper twice,
+   build it in the harness on its own branch (test it, in CI, on the stack it failed on), then run a DRY
+   round from demo to application to integration to core. Release it (see RELEASING.md) and raise the pins.
+8. **Disclose.** One issue per finding (template `issue.md`), then one PR (`pr.md`) that links the magic
+   point at the pushed commit and shows the `--due-work-summary` capture, then the plain-language comments
+   with ASCII art (`eli5-issue.md`, `eli5-pr.md`) on each. Fill their template and policy first.
+9. **Keep it green.** Watch CI. The usual causes are in the playbook's table: lockfile, an environment that
+   does not install the harness, a version-dependent table, a bot commit (merge, never rebase), workflows
+   waiting for approval, a `main` that is already red (prove it, say so, leave their test alone).
+10. **Record.** Update the demo and README tables if the project is a shipped demo, and keep the scratch
+    material until the threads close.
+
+## Before you say it is done
+
+- [ ] their own suite ran green here first; the baseline is known
+- [ ] each finding has an independent reproduction, and only measured claims are in the text
+- [ ] `ObligationContract`, its decorated collected class, `Findings`, gaps, plain test, and the magic comment are in the pushed commit
+- [ ] generated node IDs were collected and executed; the PR links the declaration, class, and summary/artifact, with gaps counted separately
+- [ ] the harness gap (if any) is fixed, tested in CI on that stack, released, and the pins raised
+- [ ] issues, PR and plain-language comments are posted with the self-identification line, no fix offered
+- [ ] CI is green, or its red is explained with evidence

pytest_obligation-0.8.0/.claude-plugin/marketplace.json

@@ -0,0 +1,12 @@
+{
+  "name": "pytest-obligation",
+  "description": "Generate ObligationContract tests and investigate lost or duplicate work in Python projects.",
+  "owner": {"name": "Gigaverse"},
+  "plugins": [
+    {
+      "name": "pytest-obligation",
+      "source": "./plugins/pytest-obligation",
+      "description": "Generate ObligationContract tests in your app or verify due-work bugs in open-source projects."
+    }
+  ]
+}

pytest_obligation-0.8.0/.gitattributes

@@ -0,0 +1 @@
+* text=auto eol=lf

pytest_obligation-0.8.0/.github/workflows/ci.yml

@@ -0,0 +1,470 @@
+name: CI
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+
+permissions:
+  contents: read
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+# Every job installs from uv.lock with --locked: a lock that no longer matches
+# pyproject.toml fails CI instead of silently resolving something new.
+
+jobs:
+  lint:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v7
+      - uses: astral-sh/setup-uv@v7
+        with:
+          enable-cache: true
+      - run: uv sync --locked --only-group lint
+      - run: uv run --no-sync ruff check
+      - run: uv run --no-sync ruff format --check
+      - name: actionlint
+        run: docker run --rm -v "$PWD:/repo" -w /repo rhysd/actionlint:1.7.7 -color
+
+  typecheck:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v7
+      - uses: astral-sh/setup-uv@v7
+        with:
+          enable-cache: true
+      # The demos type-check against the upstream code they run (Saleor's in its own job).
+      - run: uv run --no-project python demos/fetch_upstream.py procrastinate dbos-demo-apps
+      - run: uv sync --locked --all-extras --group demos
+      - run: uv pip install -e demos/.upstream/procrastinate
+      - run: uv run --no-sync pyrefly check
+
+  core:
+    # The core must work with no framework installed: the package, pytest and pytest-xdist only.
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        python: ["3.11", "3.12", "3.13", "3.14"]
+    steps:
+      - uses: actions/checkout@v7
+      - uses: astral-sh/setup-uv@v7
+        with:
+          enable-cache: true
+          python-version: ${{ matrix.python }}
+      # xdist is no framework: it lets the verify option's self-tests split a session across workers.
+      - run: uv sync --python ${{ matrix.python }} --locked --no-default-groups --group xdist
+      - run: uv run --python ${{ matrix.python }} --no-sync python scripts/check_core_is_framework_free.py
+      - run: uv run --python ${{ matrix.python }} --no-sync python -m compileall -q src/pytest_obligation
+      - run: uv run --python ${{ matrix.python }} --no-sync pytest tests/core
+
+  minimum-models:
+    # Exercise nested generic declarations on the supported minimum, without Hypothesis or frameworks.
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v7
+      - uses: astral-sh/setup-uv@v7
+        with:
+          enable-cache: true
+          python-version: "3.12"
+      - run: uv sync --locked --no-default-groups --group xdist
+      # This lane deliberately checks the declared lower bound, not uv.lock's current release.
+      - run: uv pip install 'pydantic==2.11.0'
+      - run: uv run --no-sync python scripts/check_core_is_framework_free.py
+      - run: uv run --no-sync pytest tests/core
+      - run: uv run --no-sync pytest examples/adopter/tests --due-work-require-assessed
+
+  exploration:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v7
+      - uses: astral-sh/setup-uv@v7
+        with:
+          enable-cache: true
+      - run: uv sync --locked --no-default-groups --extra exploration --group xdist
+      - run: uv run --no-sync pytest tests/core
+
+  django:
+    runs-on: ubuntu-latest
+    services:
+      postgres:
+        image: postgres:16
+        env:
+          POSTGRES_PASSWORD: postgres
+        ports: ["5432:5432"]
+        options: >-
+          --health-cmd pg_isready --health-interval 5s --health-timeout 5s --health-retries 10
+    env:
+      PGHOST: localhost
+      PGUSER: postgres
+      PGPASSWORD: postgres
+      PGPORT: "5432"
+      PGDATABASE: pytest_obligation
+    steps:
+      - uses: actions/checkout@v7
+      - uses: astral-sh/setup-uv@v7
+        with:
+          enable-cache: true
+      # pytest-django gives each xdist worker its own test database.
+      - run: uv sync --locked --no-default-groups --extra django --extra celery --group xdist
+      - run: uv run --no-sync pytest tests/django --ds=tests.django.settings -n 4
+
+  # The same suite on psycopg2, which Django uses when psycopg 3 is not installed: pretix and many older
+  # projects run on it, and the host's autocommit-write probe once assumed psycopg 3.
+  django-psycopg2:
+    runs-on: ubuntu-latest
+    services:
+      postgres:
+        image: postgres:16
+        env:
+          POSTGRES_PASSWORD: postgres
+        ports: ["5432:5432"]
+        options: >-
+          --health-cmd pg_isready --health-interval 5s --health-timeout 5s --health-retries 10
+    env:
+      PGHOST: localhost
+      PGUSER: postgres
+      PGPASSWORD: postgres
+      PGPORT: "5432"
+      PGDATABASE: due_work_harness_psycopg2
+    steps:
+      - uses: actions/checkout@v7
+      - uses: astral-sh/setup-uv@v7
+        with:
+          enable-cache: true
+      - run: uv sync --locked --no-default-groups --extra django --extra celery --group xdist
+      - run: uv pip uninstall psycopg psycopg-binary && uv pip install psycopg2-binary
+      - run: uv run --no-sync pytest tests/django --ds=tests.django.settings -n 4
+
+  # Redis-backed queues, and Celery's real worker as a child process on a Redis broker.
+  redis:
+    runs-on: ubuntu-latest
+    services:
+      redis:
+        image: redis:7
+        ports: ["6379:6379"]
+        options: >-
+          --health-cmd "redis-cli ping" --health-interval 5s --health-timeout 5s --health-retries 10
+    env:
+      REDIS_URL: redis://localhost:6379/14
+    steps:
+      - uses: actions/checkout@v7
+      - uses: astral-sh/setup-uv@v7
+        with:
+          enable-cache: true
+      - run: uv sync --locked --no-default-groups --extra rq --extra celery --extra exploration --group xdist
+      # Each xdist worker takes its own four Redis databases (tests/redis_databases.py), so cases that
+      # empty theirs cannot wipe another worker's; four workers use all 16 of the service's databases.
+      - run: uv run --no-sync pytest tests/rq tests/celery_worker -p no:django -rxX -n 4 --durations=10 --due-work-explore=smoke --due-work-require-assessed --due-work-profile-report=profiles-redis.json
+
+      - name: Require the promised adopters and executed capabilities
+        run: |
+          uv run --no-sync python scripts/check_adoption_report.py profiles-redis.json --contract "catalog on RQ" --contract "catalog through Celery" --verified E G H I
+          uv run --no-sync python scripts/check_adoption_report.py profiles-redis.json --contract "rq reference jobs" --verified G J --exercised H B
+      - name: Keep executed profile evidence
+        if: always()
+        uses: actions/upload-artifact@v4
+        with:
+          name: profiles-redis
+          path: |
+            profiles-*.json
+            demos/**/profiles-*.json
+          if-no-files-found: error
+
+  mongodb:
+    runs-on: ubuntu-latest
+    env:
+      MONGODB_URI: mongodb://localhost:27017/?directConnection=true
+    steps:
+      - uses: actions/checkout@v7
+      - uses: astral-sh/setup-uv@v7
+        with:
+          enable-cache: true
+      - run: uv sync --locked --no-default-groups --extra mongodb
+      # Transactions need a replica set on this disposable runner.
+      - run: |
+          docker run -d --name due-work-mongodb -p 27017:27017 mongo:7 --replSet dwh --bind_ip_all
+          for attempt in $(seq 1 30); do
+            if docker exec due-work-mongodb mongosh --quiet --eval 'rs.initiate({_id:"dwh",members:[{_id:0,host:"localhost:27017"}]})'; then break; fi
+            if [ "$attempt" -eq 30 ]; then exit 1; fi
+            sleep 1
+          done
+      - run: uv run --no-sync pytest tests/mongodb -q
+
+  aiokafka:
+    runs-on: ubuntu-latest
+    services:
+      kafka:
+        image: apache/kafka:3.9.1
+        ports: ["9092:9092"]
+        env:
+          KAFKA_NODE_ID: "1"
+          KAFKA_PROCESS_ROLES: broker,controller
+          KAFKA_LISTENERS: PLAINTEXT://:9092,CONTROLLER://:9093
+          KAFKA_ADVERTISED_LISTENERS: PLAINTEXT://localhost:9092
+          KAFKA_CONTROLLER_LISTENER_NAMES: CONTROLLER
+          KAFKA_CONTROLLER_QUORUM_VOTERS: 1@localhost:9093
+          KAFKA_LISTENER_SECURITY_PROTOCOL_MAP: CONTROLLER:PLAINTEXT,PLAINTEXT:PLAINTEXT
+          KAFKA_OFFSETS_TOPIC_REPLICATION_FACTOR: "1"
+          KAFKA_GROUP_INITIAL_REBALANCE_DELAY_MS: "0"
+        options: >-
+          --health-cmd "/opt/kafka/bin/kafka-topics.sh --bootstrap-server localhost:9092 --list"
+          --health-interval 5s --health-timeout 10s --health-retries 12
+    env:
+      KAFKA_BOOTSTRAP_SERVERS: localhost:9092
+    steps:
+      - uses: actions/checkout@v7
+      - uses: astral-sh/setup-uv@v7
+        with:
+          enable-cache: true
+      - run: uv sync --locked --no-default-groups --extra aiokafka
+      - run: uv run --no-sync pytest tests/aiokafka -q
+
+  prefect:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v7
+      - uses: astral-sh/setup-uv@v7
+        with:
+          enable-cache: true
+      - run: uv sync --locked --no-default-groups --extra prefect --extra exploration
+      - run: uv run --no-sync pytest tests/prefect -q --due-work-explore=smoke --due-work-require-assessed --due-work-profile-report=profiles-prefect.json
+
+      - name: Require the promised adopters and executed capabilities
+        run: |
+          uv run --no-sync python scripts/check_adoption_report.py profiles-prefect.json --contract "catalog through Prefect engine" --verified E G H I
+      - name: Keep executed profile evidence
+        if: always()
+        uses: actions/upload-artifact@v4
+        with:
+          name: profiles-prefect
+          path: |
+            profiles-*.json
+            demos/**/profiles-*.json
+          if-no-files-found: error
+
+  demos:
+    runs-on: ubuntu-latest
+    services:
+      postgres:
+        image: postgres:16
+        env:
+          POSTGRES_PASSWORD: postgres
+        ports: ["5432:5432"]
+        options: >-
+          --health-cmd pg_isready --health-interval 5s --health-timeout 5s --health-retries 10
+    env:
+      PGHOST: localhost
+      PGUSER: postgres
+      PGPASSWORD: postgres
+      PGPORT: "5432"
+    steps:
+      - uses: actions/checkout@v7
+        with:
+          # The check action builds the harness from this checkout; its version comes from git tags.
+          fetch-depth: 0
+      - uses: astral-sh/setup-uv@v7
+        with:
+          enable-cache: true
+      # Saleor too: the README's demo GIF shows the coverage scan of its checkout. Wagtail's source is
+      # what the Wagtail demo's coverage check reads; the demo itself runs the wheel from PyPI.
+      - run: uv run --no-project python demos/fetch_upstream.py procrastinate dbos-demo-apps saleor wagtail
+      - run: uv sync --locked --no-default-groups --extra django --group demos
+      # procrastinate's demos ship only in its source tree, pinned by fetch_upstream.py.
+      - run: uv pip install -e demos/.upstream/procrastinate
+      - name: Postgres without durability
+        run: demos/throwaway_postgres.sh
+      # The durations are the evidence for the next change to this job's speed.
+      - name: procrastinate demo_django
+        env:
+          PGDATABASE: procrastinate_demo
+        run: >-
+          uv run --no-sync pytest demos/procrastinate_demo_django --ds=demos.procrastinate_demo_django.settings
+          -rxX --durations=15 --due-work-require-assessed --due-work-profile-report=profiles-procrastinate.json
+      - name: DBOS transactional-outbox
+        run: uv run --no-sync pytest demos/dbos_transactional_outbox -p no:django -rxX --durations=15 --due-work-require-assessed --due-work-profile-report=profiles-dbos.json
+      # Migrating Wagtail's test database is most of its run, and depends only on the pinned packages
+      # (uv.lock) and the demo's settings: restored from the cached dump when the key matches.
+      - uses: actions/cache@v4
+        with:
+          path: demos/.ci-cache/wagtail_tasks.dump
+          key: wagtail-test-db-pg16-${{ hashFiles('uv.lock', 'demos/wagtail_tasks/settings.py', 'demos/cached_database.py') }}
+      - name: Wagtail's test database, migrated once
+        # As pytest-django creates a test database: every app's migrations, syncdb for the rest, no checks.
+        run: >-
+          uv run --no-sync python demos/cached_database.py test_wagtail_tasks --dump demos/.ci-cache/wagtail_tasks.dump
+          -- uv run --no-sync python -m django migrate --settings=demos.wagtail_tasks.settings
+          --run-syncdb --skip-checks --no-input --verbosity 0
+      # Wagtail 8.0 on django-tasks-db, both from PyPI; the demo's own directory is pytest's rootdir.
+      - name: Wagtail on django-tasks-db
+        working-directory: demos/wagtail_tasks
+        env:
+          PGDATABASE: wagtail_tasks
+        run: uv run --no-sync --project ../.. pytest --reuse-db --due-work-verify --durations=15 --due-work-require-assessed --due-work-profile-report=profiles-wagtail.json
+      - name: the README's demo GIF prints what the harness prints
+        run: uv run --no-sync pytest demos/test_readme_gif.py -p no:django --durations=15
+      # Each demo is an adopter: its handoffs are checked the way an adopter's CI checks them.
+      - name: procrastinate demo_django, every handoff accounted for
+        uses: ./check
+        with:
+          working-directory: demos/procrastinate_demo_django
+          package: ${{ github.workspace }}
+      - name: DBOS transactional-outbox, every handoff accounted for
+        uses: ./check
+        with:
+          working-directory: demos/dbos_transactional_outbox
+          package: ${{ github.workspace }}
+      - name: Wagtail on django-tasks-db, every handoff accounted for
+        uses: ./check
+        with:
+          working-directory: demos/wagtail_tasks
+          package: ${{ github.workspace }}
+
+      - name: Require the promised adopters and executed capabilities
+        run: |
+          uv run --no-sync python scripts/check_adoption_report.py profiles-procrastinate.json --contract "procrastinate demo_django: create book" --contract "procrastinate demo_django: create book, with its fixes" --exercised I
+          uv run --no-sync python scripts/check_adoption_report.py profiles-dbos.json --contract "DBOS transactional-outbox: place order" --verified I --exercised H J
+          uv run --no-sync python scripts/check_adoption_report.py demos/wagtail_tasks/profiles-wagtail.json --contract "wagtail: image and document lifecycle" --contract "wagtail: publishing a page behind a CDN" --exercised I --verified H J
+          uv run --no-sync python scripts/check_adoption_report.py demos/wagtail_tasks/profiles-wagtail.json --contract "wagtail: image and document lifecycle" --exercised E
+      - name: Keep executed profile evidence
+        if: always()
+        uses: actions/upload-artifact@v4
+        with:
+          name: profiles-demos
+          path: |
+            profiles-*.json
+            demos/**/profiles-*.json
+          if-no-files-found: error
+
+  saleor:
+    # Saleor's checkout, unmodified, in Saleor's own environment built from its uv.lock.
+    runs-on: ubuntu-latest
+    services:
+      postgres:
+        image: postgres:16
+        env:
+          POSTGRES_PASSWORD: postgres
+        ports: ["5432:5432"]
+        options: >-
+          --health-cmd pg_isready --health-interval 5s --health-timeout 5s --health-retries 10
+    env:
+      DATABASE_URL: postgres://postgres:postgres@localhost:5432/saleor
+      # Absolute: run.sh works in the demo directory.
+      SALEOR_DATABASE_DUMP: ${{ github.workspace }}/demos/.ci-cache/saleor.dump
+    steps:
+      - uses: actions/checkout@v7
+        with:
+          # The check action builds the harness from this checkout; its version comes from git tags.
+          fetch-depth: 0
+      - uses: astral-sh/setup-uv@v7
+        with:
+          enable-cache: true
+          # Saleor's pin lives in fetch_upstream.py; its lock is fetched after this step.
+          cache-dependency-glob: |
+            uv.lock
+            demos/fetch_upstream.py
+      - run: uv run --no-project python demos/fetch_upstream.py saleor
+      - name: Type-check the demo against Saleor
+        run: demos/saleor_checkout/run.sh --typecheck
+      - name: Saleor, every handoff accounted for
+        uses: ./check
+        with:
+          working-directory: demos/saleor_checkout
+          package: ${{ github.workspace }}
+      - name: Postgres without durability
+        run: demos/throwaway_postgres.sh
+      # Saleor's migrations depend only on its pin (in fetch_upstream.py): their result is cached as a
+      # dump, restored in seconds instead of migrating for two minutes.
+      - uses: actions/cache@v4
+        with:
+          path: ${{ env.SALEOR_DATABASE_DUMP }}
+          key: saleor-test-db-pg16-${{ hashFiles('demos/fetch_upstream.py', 'demos/saleor_checkout/prepare_db.py', 'demos/cached_database.py') }}
+      # One xdist worker per runner CPU. Four workers each migrating their own database took
+      # 193 s apiece; migrating once and cloning a copy per worker costs one migration.
+      - name: Saleor's test databases, migrated once
+        run: demos/saleor_checkout/run.sh --prepare-db 4
+      # The durations are the evidence for the next change to this job's speed. --dist worksteal was
+      # tried: pytest took 75 s and 80 s, inside the 66 to 91 s this step takes with the default on the
+      # same code, as runners vary. No gain was measurable, so the default stays.
+      - name: Saleor checkout
+        run: demos/saleor_checkout/run.sh -rxX --due-work-explore=smoke --due-work-verify -n 4 --reuse-db --durations=15 --due-work-require-assessed --due-work-profile-report=profiles-saleor.json
+
+      - name: Require the promised adopters and executed capabilities
+        run: |
+          demos/.upstream/saleor/.venv/bin/python scripts/check_adoption_report.py demos/saleor_checkout/profiles-saleor.json --contract "saleor checkout" --contract "saleor checkout, Transactions API with automatic completion" --exercised H I
+          demos/.upstream/saleor/.venv/bin/python scripts/check_adoption_report.py demos/saleor_checkout/profiles-saleor.json --contract "saleor payment event reports" --verified E --exercised I
+      - name: Keep executed profile evidence
+        if: always()
+        uses: actions/upload-artifact@v4
+        with:
+          name: profiles-saleor
+          path: |
+            profiles-*.json
+            demos/**/profiles-*.json
+          if-no-files-found: error
+
+  build:
+    # The published wheel alone: it installs without any framework, imports, and ships py.typed.
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v7
+        with:
+          # The version is derived from tags, so build with the history that has them.
+          fetch-depth: 0
+      - uses: astral-sh/setup-uv@v7
+        with:
+          enable-cache: true
+      - run: uv build
+      - run: unzip -l dist/*.whl | grep -q "pytest_obligation/py.typed"
+      - run: unzip -l dist/*.whl | grep -q "due_work_harness/py.typed"
+      - run: uv venv /tmp/wheel && uv pip install --python /tmp/wheel dist/*.whl
+      - run: /tmp/wheel/bin/python scripts/check_core_is_framework_free.py
+      - name: Verify old and new adopters against the installed wheel
+        run: /tmp/wheel/bin/python -m pytest tests/core/test_compatibility.py tests/core/test_package_identity.py -q
+
+  actions:
+    # The published GitHub Actions, run against the example adopter the way a project would use them.
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v7
+        with:
+          fetch-depth: 0
+      - name: check passes on a project whose every handoff is accounted for
+        uses: ./check
+        with:
+          working-directory: examples/adopter
+          package: ${{ github.workspace }}
+      - name: test runs the project's generated suites without Hypothesis
+        uses: ./test
+        with:
+          working-directory: examples/adopter
+          pytest-args: --due-work-require-assessed
+      - name: Generated histories also support optional search
+        uses: ./test
+        with:
+          working-directory: examples/adopter
+          sync-args: --extra exploration
+          pytest-args: --due-work-explore=smoke --due-work-require-assessed --due-work-profile-report=profiles-example.json
+      - name: Replay and sensitivity controls
+        working-directory: examples/adopter
+        run: uv run --no-sync pytest tests/test_catalog_failures.py
+      - name: Require example capabilities
+        working-directory: examples/adopter
+        run: uv run --no-sync python ../../scripts/check_adoption_report.py profiles-example.json --contract "example catalog" --verified E G H I
+      - name: Keep example profile evidence
+        if: always()
+        uses: actions/upload-artifact@v4
+        with:
+          name: profiles-example
+          path: examples/adopter/profiles-example.json
+          if-no-files-found: error
+      - name: check fails once a handoff loses its disposition
+        run: |
+          rm examples/adopter/tests/test_summaries_exemption.py
+          if uvx --from "$GITHUB_WORKSPACE" due-work-harness check --root examples/adopter; then
+            echo "::error::the check passed a project with an unaccounted handoff" >&2
+            exit 1
+          fi

pytest_obligation-0.8.0/.github/workflows/pr-title.yml

@@ -0,0 +1,34 @@
+name: PR title
+
+# Pull requests are squash-merged with their title as the commit message, and
+# release-please reads those commits to choose the next version. The title must
+# therefore be a Conventional Commit: see RELEASING.md.
+
+on:
+  pull_request_target:
+    types: [opened, edited, synchronize, reopened]
+
+permissions:
+  pull-requests: read
+
+jobs:
+  conventional-commit:
+    runs-on: ubuntu-latest
+    steps:
+      # Reads only the title through the API; it never checks out the pull request's code.
+      - uses: amannn/action-semantic-pull-request@v5
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        with:
+          types: |
+            feat
+            fix
+            perf
+            revert
+            docs
+            refactor
+            test
+            build
+            ci
+            chore
+          requireScope: false