maksimzayats/specx/specx-tests

Claude Code and agent skill by maksimzayats/specx

Trust tier 2
  • community
LLM review pending
Source check pending

Checking that the SKILL.md file is still in the repository.

Add or refine tests for specx Python services. Use when creating unit tests for use cases/services, integration tests for FastAPI controllers or infrastructure adapters, e2e smoke tests, architecture import guardrails, DI override tests, pytest fixtures, or coverage and boundary checks.

Install

npx vskill@latest install maksimzayats/specx/specx-tests

SKILL.md

specx Tests

Use this skill when behavior, wiring, or architecture boundaries need tests. Read references/testing.md before creating test files.

Test Layers

  • tests/_support/: generic clients, DB helpers, and shared integration

helpers only. This is not a test suite and does not hold project-specific doubles.

  • tests/unit/: core services, use cases, and capabilities resolved from a

fresh application container returned by the project's get_container().

  • tests/integration/: real internal graph tests. Core use-case integration

tests call resolved use cases against the transactional DB; delivery integration tests exercise HTTP mapping; migrations prove Alembic behavior.

  • tests/e2e/: optional whole-app smoke flows.
  • tests/guardrails/: optional programmatic

specx.testing.architecture.assert_specx_architecture wrappers for genuinely project-specific extra rules. Standard packaged rules run through uv run specx check.

Rules

  • Test behavior and boundaries, not implementation ceremony.
  • Required generated coverage is currently scoped to core services, use cases,

and capabilities.

  • Mirror source module paths directly with flat test files, for example

tests/unit/core/tasks/services/test_title_service.py.

  • Do not create per-target test folders, harness.py, target factories, or

target harnesses.

  • tests/unit/conftest.py owns the fresh real-app Container fixture for unit

tests and any project-wide test overrides. tests/integration/conftest.py owns the transactional DB-backed container fixture for integration tests.

  • Test functions receive container, register any scenario-specific overrides

before resolution, then call container.resolve(Target).

  • If a complete replacement is needed by every test in one module, a

module-local container fixture may register it before returning the container.

  • Keep one-off class-based test doubles in the test_*.py module that uses

them. When the same double is reused by multiple unit modules, put it in a mirrored tests/unit/core/<scope>/{capabilities,gateways,repositories}/fake_<source_module>.py file.

  • Do not create tests/_support/fakes, tests/**/_fakes.py, generic

_scenarios.py, fake modules outside those mirrored unit port/capability packages, or double classes in conftest.py.

  • Use MagicMock or AsyncMock inline in the test function when only one

behavior needs to be changed for that scenario. Prefer autospeccing when call signatures matter and spec_set when unexpected attributes must fail.

  • Unit tests replace external IO, time, randomness, network, Redis, database,

and framework resources with local doubles or inline mocks.

  • Integration tests use the real internal app graph. Do not mock internal use

cases, services, or capabilities; stub only external systems when needed.

  • Add core use-case integration tests under tests/integration/core/... for

use cases that inject a UoW manager; delivery tests should own HTTP mapping, not be the only persistence proof.

  • Persistence integration tests use the production database family when

dialect behavior matters. A rollback harness may not replace isolated commit-visible tests for locking, concurrency, isolation, or after-commit behavior.

  • Core health tests cover required-dependency readiness and any reusable probe

services and use cases. Delivery probe tests cover /healthz and /readyz as operational endpoints, not versioned business API routes. /healthz must prove a lightweight process response only; /readyz must prove required infrastructure readiness, including a real bounded DB check for SQLAlchemy services.

  • Probe route tests assert Cache-Control: no-store, readiness failure returns

503, probe routes are excluded from OpenAPI, and legacy /api/v1/health is absent when replacing old generated health endpoints.

  • Unit-test logging configurators by overriding logging settings,

monkeypatching logging.config.dictConfig, and asserting the generated stdlib config. Use caplog only when a log record is meaningful behavior.

  • Unit-test FastAPI lifecycle managers by overriding closeable infrastructure

resources and asserting shutdown order. Route integration helpers must run ASGI lifespan explicitly.

  • Use httpx2, not legacy httpx, for generated HTTP client and ASGI transport

tests. Enter LifespanManager, then pass the yielded manager's manager.app to ASGITransport so request scopes receive lifespan state.

  • FastAPI route tests compare response status codes with fastapi.status

constants, not raw integer literals.

  • Use container.resolve(...) for normal synchronous graph construction, even

when the resolved use case has an async execute(...); use await container.aresolve(...) only when DI construction itself has async providers.

  • Mock fixtures should register one external collaborator for the behavior

under test. Do not bundle unrelated mocks in a dict or class-keyed fixture.

  • Use native pytest fixtures for test dependencies. Do not enable

diwire.integrations.pytest_plugin, and do not use Injected[...] parameters in tests.

  • AnyIO runs tests on every installed supported backend by default. If the app

graph is asyncio-specific, override the top-level anyio_backend fixture to return "asyncio"; leave it unpinned only when the suite intentionally supports every installed backend.

  • Do not add filler smoke tests that only assert container.resolve(...)

returns an instance.

  • Do not hand-build application graphs in test bodies. Resolve project classes

from the container; local test doubles may be instantiated in the test module before registration.

  • Keep unit tests free from FastAPI request objects and real external IO.
  • Every test directory must include an empty __init__.py file.
  • Use uv run specx check as the default guardrail mechanism for specx

boundaries such as docstrings, use-case inputs, UoW injection, route paths, direct persistence dependency rejection in use cases, container imports, and AGENTS.md command coverage.

  • Disable built-in guardrails only with exact semantic IDs under

[tool.specx].ignore and a project reason recorded beside the configuration.

  • Generated projects use [tool.specx].select = ["ALL"]. Narrower projects

enable technology-specific families explicitly with [tool.specx].extend-select; FastAPI projects select fastapi.

  • Add extra_rules only for project-specific checks that are not covered by a

built-in SpecxRuleId; use the programmatic wrapper for those projects.

  • Existing workflows and projects with custom rules may use

references/render_architecture_guardrails.py to render the tiny wrapper.

Code Style

Use blank lines as logical separators in all code. Keep related statements together, but separate independent setup, action, assertion, response, branch, and transformation groups so long blocks stay readable.

References

  • references/testing.md - folder layout, fixtures, unit/integration examples,

and architecture guardrail snippets.

  • references/render_architecture_guardrails.py - compatibility renderer for

the tiny specx architecture wrapper.

Security

  • Security scanPassed, 100 of 100Pattern checks on the skill's files.
  • Model reviewNot runA model review is scheduled for the next rescan.

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

7-day trend score
0

Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/maksimzayats/specx/specx-tests/badge)](https://verified-skill.com/skills/maksimzayats/specx/specx-tests)

Something wrong with this skill or its page? Tell us.