AGENT SKILL

test-dev

Use when writing xUnit unit tests, integration tests with real databases, E2E Playwright tests, or Rust (cargo test/nextest) and Go (go test) tests. Do not use for application or persistence implementation.

License 许可
MIT
最近更新
2026-11-15
Stars
★ 23
Forks
3

01 前言 Front Matter

name
test-dev
license
MIT
description
Use when writing xUnit unit tests, integration tests with real databases, E2E Playwright tests, or Rust (cargo test/nextest) and Go (go test) tests. Do not use for application or persistence implementation.

02 资源文件 6

03 SKILL.md 正文

Test Developer

Test policy lives in the path declared as <test-coding-standard> in the repository root AGENTS.md. Language-level C# decisions live in the path declared as <csharp-coding-standard>. Fetch those with docs_get_coding_standard_for_technology before the references below, naming each technology and the topic you need; read a file only when the tool is missing — this skill is procedure, not policy.

Identify your sub-task and read ONLY the relevant reference before proceeding.

Sub-Task When to Use Reference
Unit Test Patterns xUnit, NSubstitute, FluentAssertions, pure domain/service/validator logic tests Unit Test Patterns
Integration Test Patterns Real SQL Server, LocalDB/Docker, WebApplicationFactory, repository and API tests Integration Test Patterns
E2E / Browser Tests Playwright MCP tools, user flows, form submissions, multi-step scenarios E2E Test Patterns
Rust and Go Test Patterns #[cfg(test)] modules, tests/ integration crates, cargo nextest, table-driven _test.go tests, t.Run subtests, -race, fakes behind traits and interfaces Rust and Go Test Patterns
Mock Usage Analysis Classify mocks as dead/unreachable/redundant/used; detect over-mocking Mock Usage Analysis
Test Maintainability Magic values, duplicate setup, assertion overload, brittle naming — detect and refactor Test Maintainability

Rule: Read only the reference(s) relevant to your current task. Do not pre-load all references.

Execution & Parallelization Invariants

  • Adversarial RED/GREEN role boundary: Maintain strict adversarial separation between RED phase test authoring (test-dev) and GREEN phase implementation (csharp-dev, etc.). Never collapse RED and GREEN into a single worker dispatch: test authors author tests and verify failures; implementers make them pass, while preserving independent verification at task completion.
  • Same-role task bundling: Tasks belonging to the same specialist role are bundled in same-role micro-task dispatches to minimize context thrash.
  • Build once, test many with direct assembly runner: Do not rebuild or recompile the entire solution on every iteration. The direct assembly runner is the language-specific form for a .NET solution — its exact assembly path and target framework come from the host's test standard, in its ## Completion section — while other stacks use their own runners (such as cargo test or go test). Use --no-build for targeted test runs, or execute the pre-built test assembly directly: dotnet <path-to-test-assembly>.dll -class "<Namespace>.<TestClass>"
  • Zero-stash RED proof rule: Forbid git stash for RED proofs; never run git stash to fake clean working state. Run the RED proof against a clean base checkout, throwaway copy, or isolated worktree before any implementation edits exist.
  • Per-task artifact isolation: Isolate intermediate build and review artifacts using unique per-task paths, passing dotnet build --artifacts-path .agent_scratchpad/<task-id>/artifacts for build output isolation, and kyber-weave review gates . --agent <agent> --out artifacts/task-gate-<task-id>.json (or .agent_scratchpad/<task-id>/task-gate.json) so concurrent task-gate runs never overwrite one another.

Completion Digest

The Discovery: codegraph N, docs N, direct reads N header precedes the 6-field completion digest schema:

STATUS: PASS | FAIL | ESCALATION
FILES: <comma-separated-paths>
TESTS: total=N failed=N passed=N
COMMIT: <commit-sha | NONE>
REMOTE_HEAD: <remote-head-sha | UNPUSHED>
TASK_GATE: PASS | FAIL | NONE_DECLARED | NONE_APPLICABLE | UNAVAILABLE — agent=test-dev gates=<n> failed=<ids|none> skipped=<n> report=<path> runs=<k>

04 同仓库更多 Skills