3be9b17, the version that was reviewed and scanned here. Newer commits reach this catalog only after a new review. License: MIT, by Off Grid AI.
Test like a user: no mocks of your own code
This standard works for web apps, services, mobile apps and desktop apps. "The project's own code" means everything in the repo you are testing: its source folders, services, stores, hooks, components, screens, navigators, parsers and registries. Find those folders first (for example src/, app/, lib/, packages/*).
The contract
- Aim for the strongest faithful proof of product behaviour, not the fastest green test or the shortest path to an answer. Exercise the broadest user journey needed to expose real failures. Never trade coverage quality, maintainability or UI/UX confidence for speed.
- Do not bless a half-done migration, duplicated navigation, leftover compatibility clutter, or a journey that technically works but confuses people. Test the intended end state.
- Do not mock the project's own code. Keep its services, stores, hooks, components, screens, navigators, parsers and registries real.
- Prove product behaviour with an integration test: use the real router or navigation stack, reach the state through real user actions, and run the real production path. On web, assert the URL, history and deep links. On mobile, mount the real screen and navigation stack and use real gestures.
- Assert what the user can observe. Internal calls, mock calls and implementation details are not product outcomes.
- For a replacement or a change to how things are organised, assert both sides: every needed item and action can still be found through the new structure, and the old or duplicate surface is gone. Redirects and source-text assertions alone do not prove a clear user experience.
- Check changed UI by eye at the relevant wide and narrow sizes, including expanded and collapsed, loading, empty and error, keyboard and focus, and the absence of duplicate navigation. Screenshot checks add to behaviour assertions; they do not replace them. Capture only after loading and every animation have settled, then look at the actual pixels for blank or see-through content, clipping, overflow, cut-off text, unclear hierarchy and confusing layout. A saved screenshot, a DOM visibility assertion or a green test is not by itself a visual pass.
- Fake only a boundary the test environment cannot control or reproduce faithfully, such as a native device API, a model runtime, a remote server, a full disk or running out of memory. Put the fake at that boundary and keep all of the project's logic above it real.
- Do not use direct store
setStateto set up a precondition for an integration test. - Use unit tests for pure functions and narrow contracts. They do not replace the integration test for changed product behaviour.
- When native code (for example Swift or Kotlin) owns behaviour that can be tested natively, add native tests (XCTest, JUnit or the platform's equivalent). Use a shared contract test when more than one platform builds the same capability.
Workflow
- State the behaviour the user can see and the failure being guarded against.
- Define the intended end state and name the old paths or surfaces that must disappear.
- Find the real screen, navigation path, production stack and outside boundary.
- Reuse an existing boundary fake, or add the smallest faithful fake at that boundary.
- Write the integration test through real actions, URLs and history where they apply, and visible assertions.
- For a bug, run the test against the broken code and confirm it fails for the reason you expect.
- Fix the production seam and remove the old path. Do not patch the test around a poor abstraction.
- Run the visual check, the focused test, related suites, and the project's lint, format and type checks. Run native tests when the change crosses a native boundary.
- Remove leaked timers, subscriptions, processes and pending promises. A suite that needs a force-exit flag to hide unfinished work is not clean.
Review verdict
Fail the review when a test mocks the project's own code, makes up state instead of using the product, asserts only calls or source text, skips a known bug, keeps a duplicate or replaced journey, leaves out visual or navigation proof for changed UI, or passes while async work outlives teardown. Name the exact file and the real integration test that should replace it.
For the deeper method (fakes vs stubs vs mocks, observing the transition, debugging a test), see reference/test-doctrine.md in this skill. The companion skill bug-trap is the short checklist.
Test like a user: no mocks of your own code
This standard works for web apps, services, mobile apps and desktop apps. "The project's own code" means everything in the repo you are testing: its source folders, services, stores, hooks, components, screens, navigators, parsers and registries. Find those folders first (for example src/, app/, lib/, packages/*).
The contract
- Aim for the strongest faithful proof of product behaviour, not the fastest green test or the shortest path to an answer. Exercise the broadest user journey needed to expose real failures. Never trade coverage quality, maintainability or UI/UX confidence for speed.
- Do not bless a half-done migration, duplicated navigation, leftover compatibility clutter, or a journey that technically works but confuses people. Test the intended end state.
- Do not mock the project's own code. Keep its services, stores, hooks, components, screens, navigators, parsers and registries real.
- Prove product behaviour with an integration test: use the real router or navigation stack, reach the state through real user actions, and run the real production path. On web, assert the URL, history and deep links. On mobile, mount the real screen and navigation stack and use real gestures.
- Assert what the user can observe. Internal calls, mock calls and implementation details are not product outcomes.
- For a replacement or a change to how things are organised, assert both sides: every needed item and action can still be found through the new structure, and the old or duplicate surface is gone. Redirects and source-text assertions alone do not prove a clear user experience.
- Check changed UI by eye at the relevant wide and narrow sizes, including expanded and collapsed, loading, empty and error, keyboard and focus, and the absence of duplicate navigation. Screenshot checks add to behaviour assertions; they do not replace them. Capture only after loading and every animation have settled, then look at the actual pixels for blank or see-through content, clipping, overflow, cut-off text, unclear hierarchy and confusing layout. A saved screenshot, a DOM visibility assertion or a green test is not by itself a visual pass.
- Fake only a boundary the test environment cannot control or reproduce faithfully, such as a native device API, a model runtime, a remote server, a full disk or running out of memory. Put the fake at that boundary and keep all of the project's logic above it real.
- Do not use direct store
setStateto set up a precondition for an integration test. - Use unit tests for pure functions and narrow contracts. They do not replace the integration test for changed product behaviour.
- When native code (for example Swift or Kotlin) owns behaviour that can be tested natively, add native tests (XCTest, JUnit or the platform's equivalent). Use a shared contract test when more than one platform builds the same capability.
Workflow
- State the behaviour the user can see and the failure being guarded against.
- Define the intended end state and name the old paths or surfaces that must disappear.
- Find the real screen, navigation path, production stack and outside boundary.
- Reuse an existing boundary fake, or add the smallest faithful fake at that boundary.
- Write the integration test through real actions, URLs and history where they apply, and visible assertions.
- For a bug, run the test against the broken code and confirm it fails for the reason you expect.
- Fix the production seam and remove the old path. Do not patch the test around a poor abstraction.
- Run the visual check, the focused test, related suites, and the project's lint, format and type checks. Run native tests when the change crosses a native boundary.
- Remove leaked timers, subscriptions, processes and pending promises. A suite that needs a force-exit flag to hide unfinished work is not clean.
Review verdict
Fail the review when a test mocks the project's own code, makes up state instead of using the product, asserts only calls or source text, skips a known bug, keeps a duplicate or replaced journey, leaves out visual or navigation proof for changed UI, or passes while async work outlives teardown. Name the exact file and the real integration test that should replace it.
For the deeper method (fakes vs stubs vs mocks, observing the transition, debugging a test), see reference/test-doctrine.md in this skill. The companion skill bug-trap is the short checklist.
Scanned by
NVIDIA SkillSpector
Automated static analysis (no LLM) · v2.12.0 · 2026-11-20
2 observations in 2 categories
- MCP Rug Pull 1
- Prompt Injection 1
SkillSpector looks for patterns such as shell commands, credential access, network calls or unpinned dependencies. An observation is something worth reading in the files before you install, not a confirmed vulnerability. Components that run code locally, like hooks, MCP servers and security tooling, naturally show more of them. Open one to see the lines it points at.
Exfiltration CommandsPrompt Injection · reference/test-doctrine.md:162
| 160 | Seeding the tested state with `store.setState(...)` or a direct `createThing(...)` call is the same |
| 161 | shortcut as `toHaveBeenCalled`. It makes up a state the user reached by tapping, so the test never |
| 162 | proves the tapping works. Get there the way the user does: send messages to make the conversation, |
| 163 | tap through the form to make the project, drag the slider to change the setting, pick the item in |
| 164 | the picker to load it. The ONLY things you may pre-place are leaves truly outside the system: a file |
npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.MCP Rug Pull · reference/test-doctrine.md:248
| 246 | Debugging, decide whether the fault is the harness or the product, and REPORT product faults rather |
| 247 | than quietly reshaping the test until it passes. |
| 248 | - Run the related tests for the source files it touches (in Jest: `npx jest --findRelatedTests |
| 249 | <files>`) to catch knock-on breakage now, not later. |
| 250 | - Run the type check. Commit. One test per commit with a real message. Next. |
Learn more about NVIDIA SkillSpector or how components are reviewed.
Agent compatibility
Checked by NVIDIA SkillEvaluator Harness Check · v1.0.1 · 2026-11-20
How each agent reads this skill once it is installed, from the agents' documented loading rules. No agent was run. A note means the agent loads the skill but changes part of its text, or drops it.
Claude Code
loadsCodex
loadsCopilot CLI
loadsOpenCode
loadsHermes
loadsOpenClaw
loads
Every agent loads this skill as written. No notes.
Learn more about NVIDIA SkillEvaluator.