sextant-write-tests

---

name: sextant-write-tests description: Use when writing test cases, adding test coverage, or verifying code behavior through tests. Stronger signals: "write tests", "add tests", "unit test", "integration test", "test coverage", "TDD", "test this function". Also triggered automatically after a bug fix to write the reproduction test. Apply this skill before starting any test-writing work.

!awk 'f;/^---$/{c++}c==2{f=1}' ${CLAUDE_SKILL_DIR}/../principles/SKILL.md

![ -d .gitnexus ] && awk 'f;/^---$/{c++}c==2{f=1}' ${CLAUDE_SKILL_DIR}/../tool-gitnexus/SKILL.md || true

---

Test Writing Workflow

Core Principle

The value of tests is not "line coverage," but accurately reporting regressions when code changes. Good tests are a safety net for future refactoring — they go red only when behavior is truly broken, not when unrelated changes are made.

---

Entering from a Bug Fix

If this test session was triggered by a bug fix, apply the following adjustments before entering the standard workflow:

1. Write the reproduction test first — before anything else.

The root cause and trigger conditions are already known from the bug fix. Use that analysis directly:

def test_<bug_description>():
    # This test MUST FAIL before the fix is applied.
    # It documents the exact condition that triggered the bug.
    result = <call using the trigger conditions from the bug fix>
    assert <the correct behavior that was previously broken>

2. Skip Step 1 (code analysis) for the fixed function. Jump directly to Step 2, using the impact assessment from the bug fix as input for "what callers depend on."

3. Prioritize boundary tests adjacent to the bug. Focus on the boundary conditions the bug exposed — null, zero, extreme values, race conditions.

4. Name the reproduction test clearly:

def test_calculate_discount_zero_rate_does_not_divide_by_zero():
    # fix: rate=0 caused ZeroDivisionError (see bug fix root cause)
    assert calculate_discount(price=100, rate=0) == 100

Once the reproduction test is in place and passing, skip to Step 2 for any additional coverage.

If the reproduction test still fails after the fix was applied: stop the test-writing workflow. Report to the user: "The reproduction test is still failing — the fix may be incomplete. Please re-trigger with sextant-fix-bug to re-evaluate the root cause, or describe the remaining failure and I'll investigate with baseline rules." Do not proceed to Step 2.

---

Before You Start

Five Characteristics of Good Tests (F.I.R.S.T)

Characteristic	Description
Fast	Single test < 100ms, full test suite < 30s
Independent	No dependencies between tests, no dependency on execution order
Repeatable	Results consistent across any environment; no dependency on external services, system time, or random numbers
Self-validating	Results are pass/fail; no manual log inspection required
Prioritized by Impact	Write tests for high-impact functions first: core business logic, functions with many callers, code that is hard to debug manually

Choosing the Test Level

Test Pyramid (bottom to top: decreasing quantity, increasing cost)
─────────────────────────────────────────────
        ╱ ╲           E2E Tests — few, verify key user paths
       ╱   ╲
      ╱─────╲         Integration Tests — moderate, verify inter-module collaboration
     ╱       ╲
    ╱─────────╲       Unit Tests — many, verify behavior of individual functions/classes
─────────────────────────────────────────────

Selection criteria:

• Pure functions, computation logic, data transformation → Unit tests
• Inter-module interaction, database reads/writes, API calls → Integration tests
• Complete user operation flows → E2E tests (cover core paths only)

---

Complete Execution Workflow

Step 1: Analyze the Code Under Test

Before writing tests, thoroughly understand the behavioral contract of the code under test.

Questions to answer:

• What is the responsibility of the function/class under test?
• What are its inputs? What is the valid range and boundary values for each input?
• What are its outputs (return values + side effects)?
• What preconditions does it have?
• Which exceptions might it throw? Under what conditions?
• Which external collaborators does it depend on? Do these need to be mocked?

🔗 When GitNexus is available, use context MCP tool to get callees (external dependencies) automatically.

Step 2: Design Test Boundaries

Divide the behavioral space into discrete test regions.

Three required test scenarios:

① Happy Path — typical input → expected output; each meaningful parameter combination

② Boundary Conditions:

• Null/zero/empty collection/empty string
• Maximum/minimum/just beyond range
• Single-element collection (off-by-one happens most here)
• Type boundaries (integer overflow, float precision)

③ Error Path:

• Invalid input → expected exception/error code
• External dependency failure → expected degradation/error handling
• Concurrency/race conditions (if applicable)
• Timeout scenarios (if network/IO is involved)

Test Boundary Matrix
─────────────────────────────────────────────────────
Target under test: <function/class name>

Happy path:
  case_1: <input description> → <expected output>

Boundary conditions:
  boundary_1: <boundary input> → <expected behavior>

Error path:
  error_1: <invalid input> → <expected exception/error>
─────────────────────────────────────────────────────

Step 3: Determine Test Structure

Naming conventions: Test names must describe behavior, not repeat the function name.

# ✅ Good test naming
def test_discount_returns_zero_when_rate_exceeds_price():
def test_login_fails_with_expired_token():
def test_parse_handles_empty_input_gracefully():

# ❌ Poor test naming
def test_discount():
def test_login():

Naming pattern: test_<behavior>_<condition> or test_<behavior>_when_<condition>_then_<expectation>

AAA Structure (Arrange → Act → Assert):

def test_order_total_applies_discount_for_vip_user():
    # Arrange
    user = create_vip_user()
    items = [Item("book", price=100), Item("pen", price=20)]
    order = Order(user=user, items=items)

    # Act
    total = order.calculate_total()

    # Assert
    assert total == 96.0  # VIP 20% discount: (100 + 20) * 0.8

Each test verifies only one behavior. Multiple asserts checking different aspects of the same behavior are acceptable.

Step 4: Handle External Dependencies

Dependency Type	Isolation Method	Use Case
Pure interface dependency (Repository, Client)	Mock objects	Unit tests
Database	In-memory database / test containers	Integration tests
External API	HTTP mock (e.g., responses, wiremock)	Integration tests
File system	Temp directory + teardown cleanup	File read/write logic
Time/random	Inject controllable clock / fixed seed	Deterministic results needed

Mock discipline:

• Mocked behavior must be consistent with the real implementation's contract
• Don't mock internal implementations of code you don't own — mock interfaces
• If mock setup is more complex than the code under test, the code has too heavy dependencies

🔗 When GitNexus is available, use context callees to distinguish internal vs external dependencies.

Step 5: Implement Tests

Recommended implementation order:

1. Write the most typical happy path case first — verify AAA structure and mock setup are correct
2. Add boundary conditions — cover boundary items in the Step 2 matrix one by one
3. Write error paths — verify error handling meets expectations

Reference existing test style: New tests' file organization, naming style, assertion library, and fixture patterns must be consistent with existing tests in the project.

🔗 When GitNexus is available, use query "<function name> test" to find reference tests.

Step 6: Validate Test Quality

Test Quality Checklist
─────────────────────────────────────────────────────
[ ] Independence: No dependencies between tests? Can run in any order? Can run individually?
[ ] Determinism: No dependency on system time, random numbers, external services?
[ ] Readability: Does reading the test name tell you what's being tested? Is AAA structure clear?
[ ] Validity: If a line of logic in the code under test is deleted, will the test fail?
[ ] Non-brittleness: Does the test depend on implementation details rather than behavior?
[ ] Boundary coverage: Does each case in the Step 2 matrix have a corresponding test?
[ ] Error messages: Is the error message on assertion failure sufficient to locate the problem?
[ ] Speed: Is a single test < 100ms?
─────────────────────────────────────────────────────

"Validity" check technique (mutation testing mindset): Make these changes to the code under test; check if tests fail:

• Change > to >=
• Change return value to None / null
• Comment out a key conditional branch

If tests still pass, they cover lines but don't verify behavior.

---

Forbidden Actions

• Write business logic in tests: Directly assert expected concrete values; don't replicate computation logic
• Assert assertTrue(result is not None): Doesn't verify any behavior
• Depend on execution order: Each test must be self-contained
• Test private methods: Test behavior of public interfaces, not implementation details
• Ignore flaky tests: Fix or delete them — flaky tests cause teams to ignore all failures

---

Sprint State Integration

If .sextant/state.json exists in the project root and the current task matches a sprint task:

• On start: offer to update the task's status from pending → in_progress. Ask: "Update sprint state to mark Task N as in_progress?"
• On completion (acceptance condition met): offer to update status to done. Ask: "Update sprint state to mark Task N as done?"
• On blocker (test failure, missing dependency, unresolvable ambiguity that halts progress): surface the issue, then ask: "Mark Task N as blocked and record the reason in flags?" If confirmed, set status: "blocked" and append {"task": N, "reason": "<one-sentence blocker description>"} to the top-level flags array. Do not proceed to the next task while a task is blocked.

Do not write the file without explicit user confirmation. If the user declines, continue without state updates.

---

Reply Format

Lightweight task (1–3 tests for a single function): one sentence only.

✅ Added <N> tests for `<function>` covering <scenarios> (<test_file>:<line>).

Medium/large task (new test module, full coverage pass, or post-bug-fix reproduction tests): full block.

─── Test Summary ────────────────────────────────────────
① Conclusion:         <one sentence: N tests written for [target], covering [happy/boundary/error paths]>
② Changes:            <test files / functions added or modified, with scenario count per category>
③ Risks / Assumptions: <mock contracts assumed; known coverage gaps; scenarios deferred>
④ Verification:       <Step 6 quality checklist result; all tests passing / not yet run>
⑤ Needs your input:   <behavioral contracts that are ambiguous; tests the user should execute to confirm>
────────────────────────────────────────────────────────