refactor(workflow): relax tests/unit/ Hypothesis mandate to guidance

Replace the hard rule that tests/unit/ must use @given with guidance:
use Hypothesis for properties spanning many inputs, plain pytest for
specific behaviors or single edge cases. @pytest.mark.slow remains
mandatory on all @given-decorated tests.

Affected files: AGENTS.md, tdd/SKILL.md, implementation/SKILL.md, verify/SKILL.md

Author: nullhack

.opencode/skills/implementation/SKILL.md

@@ -309,7 +309,7 @@ If during implementation you discover a behavior not covered by existing accepta
 - Note the gap in TODO.md under `## Next`
 - The PO will decide whether to add a new Example to the `.feature` file
 
-Extra tests in `tests/unit/` are allowed freely (coverage, edge cases, etc.) — these do not need `@id` traceability. **Every test in `tests/unit/` must be a Hypothesis property test: `@given` is required, `@pytest.mark.slow` is mandatory, plain `assert` tests without `@given` are forbidden.**
+Extra tests in `tests/unit/` are allowed freely (coverage, edge cases, etc.) — these do not need `@id` traceability. Use Hypothesis (`@given`) for properties that hold across many inputs; use plain pytest for specific behaviors or single edge cases. `@pytest.mark.slow` is mandatory on every `@given`-decorated test.
 
 ## Signature Design
 

.opencode/skills/tdd/SKILL.md

@@ -105,21 +105,21 @@ The correct test asserts on the return value. The wrong test breaks if you renam
 
 Tests in `tests/features/` are generated from `@id` criteria — use plain pytest there.
 
-Tests in `tests/unit/` cover gaps not represented by any acceptance criterion. **Every test in `tests/unit/` must be a Hypothesis property test.** No plain `assert` tests without `@given` are permitted in `tests/unit/`.
+Tests in `tests/unit/` cover gaps not represented by any acceptance criterion. Any test style is valid — plain `assert` or Hypothesis `@given`. Use Hypothesis when the test covers a **property** that holds across many inputs (mathematical invariants, parsing contracts, value object constraints). Use plain pytest for specific behaviors or single edge cases discovered during refactoring.
 
 | Situation | Location | Tool |
 |---|---|---|
 | Deterministic scenario from a `.feature` `@id` | `tests/features/` | Plain pytest (generated) |
-| Pure function needing many input combinations | `tests/unit/` | Hypothesis `@given` |
+| Property holding across many input values | `tests/unit/` | Hypothesis `@given` |
+| Specific behavior or single edge case | `tests/unit/` | Plain pytest |
 | Stateful system with sequences of operations | `tests/unit/` | Hypothesis stateful testing |
 
 **Never use Hypothesis for**: I/O, side effects, network calls, database writes.
 
 ### `tests/unit/` Rules
 
-- `@given(...)` is **required** on every test — no exceptions
-- `@pytest.mark.slow` is **mandatory** on every Hypothesis test
-- `@example(...)` is optional but encouraged for known corner cases
+- `@pytest.mark.slow` is **mandatory** on every `@given`-decorated test (Hypothesis is genuinely slow)
+- `@example(...)` is optional but encouraged when using `@given` to document known corner cases
 - `@pytest.mark.unit` or `@pytest.mark.integration` still required (one each)
 
 ## Markers (4 total)
@@ -147,11 +147,11 @@ When in doubt, start with `unit`. Upgrade to `integration` if the implementation
 
 ## Hypothesis Tests
 
-Required decorator order for every `tests/unit/` test:
+When using `@given` in `tests/unit/`, the required decorator order is:
 
 ```python
 @pytest.mark.unit        # required: exactly one of unit or integration
-@pytest.mark.slow        # required: mandatory on all Hypothesis tests
+@pytest.mark.slow        # required: mandatory on all @given tests
 @given(x=st.floats(min_value=-100, max_value=100, allow_nan=False))
 @example(x=0.0)          # optional: document known corner cases
 @settings(max_examples=200)
@@ -166,7 +166,7 @@ def test_wall_bounce_c4d5e6f7(x: float) -> None:
     assert result >= 0
 ```
 
-A test missing `@given` or `@pytest.mark.slow` in `tests/unit/` is a FAIL at Step 5 review.
+A `@given`-decorated test missing `@pytest.mark.slow` is a FAIL at Step 5 review.
 
 ### Meaningful vs. Tautological Property Tests
 

.opencode/skills/verify/SKILL.md

@@ -129,7 +129,6 @@ Read the source files changed in this feature. **Do this before running lint/sta
 | Every `@id` has a mapped test | Match `@id` tags in `.feature` files to test functions | All mapped | Missing test | Write the missing test |
 | No `@id` used by two functions | Check for duplicate `@id` hex in test function names | None | Duplicate found | Consolidate into Hypothesis `@given` + `@example` or escalate to PO |
 | Function naming | Test names match `test_<rule_slug>_<8char_hex>` | All match | Mismatch | Rename function |
-| All `tests/unit/` tests use `@given` | Read every test in `tests/unit/`; check for `@given` decorator | All have `@given` | Any test lacks `@given` | Rewrite as a Hypothesis property test |
 | All Hypothesis tests have `@pytest.mark.slow` | Read every `@given`-decorated test for the `@slow` marker | All present | Any missing | Add `@pytest.mark.slow` |
 
 #### 4g. Code Quality — any FAIL → REJECTED
@@ -225,5 +224,4 @@ OR
 | Duplicate `@id` in tests | 0 |
 | Empty evidence cells | 0 |
 | Orphaned tests | 0 |
-| Plain tests without `@given` in `tests/unit/` | 0 |
 | Hypothesis tests missing `@pytest.mark.slow` | 0 |

AGENTS.md

@@ -84,13 +84,14 @@ tests/
   features/<feature-name>/
     <rule-slug>_test.py               ← one per Rule: block, stubs from gen-tests
   unit/
-    <anything>_test.py                ← developer-authored extras (Hypothesis only)
+    <anything>_test.py                ← developer-authored extras (no @id traceability)
 ```
 
-Every test in `tests/unit/` **must** be a Hypothesis property test:
-- `@given(...)` is required — no plain `assert` tests without `@given`
-- `@pytest.mark.slow` is mandatory on every Hypothesis test
-- `@example(...)` is optional but encouraged for known corner cases
+Tests in `tests/unit/` are developer-authored extras not covered by any `@id` criterion. Any test style is valid — plain `assert` or Hypothesis `@given`. Use Hypothesis when the test covers a **property** that holds across many inputs (mathematical invariants, parsing contracts, value object constraints). Use plain pytest for specific behaviors or single edge cases discovered during refactoring.
+
+- `@pytest.mark.slow` is mandatory on every `@given`-decorated test (Hypothesis is genuinely slow)
+- `@example(...)` is optional but encouraged when using `@given` to document known corner cases
+- No `@id` tags — tests with `@id` belong in `tests/features/`, generated by `gen-tests`
 
 ## Gherkin Format