diff --git a/AGENTS.md b/AGENTS.md
index dcdcf1f1..18861c1b 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -166,7 +166,7 @@ Per-code severity and the structured-routing fields (`target_file`/`field_path`/
 
 - Python `>=3.11`, managed with `uv` (no `pip`). Deps: typer, pydantic v2, httpx, pyyaml, openapi-spec-validator, jsonschema, pymupdf.
 - Prefer immutable patterns (return new values; pure functions outside the I/O modules above).
-- The skill file `skills/loop-apidoc/SKILL.md` is written in **English** (token economy); generated *product* output remains `zh-TW`.
+- The skill file `skills/loop-apidoc/SKILL.md` is written in **English** (token economy); generated *product* output remains `zh-TW`. Validation issue text (`evidence`, `suggested_fix`, `fix_once`) is product output, so it is `zh-TW` too — one report must not address the operator in two languages, which is what happened while `validate/coverage.py` and `validate/response_contract.py` each chose their own.
 - Docsentry governs the selected Markdown documents in `.docsentry.json`: it checks local links and documented package scripts against checked-in evidence. It does not validate HTML manuals, external URLs, prose style, translations, or generated run output.
 - **Documentation language policy (for wider adoption/promotion):** teaching, promotion, and reference docs are **English-primary, Traditional-Chinese-secondary** — write the canonical copy in English so the project reaches the broadest audience, and provide zh-TW as the supporting/localized layer (e.g. `README.md` zh-TW ↔ `README.en.md` English). This applies to the human-facing docs listed under "Release: keep teaching & promotion docs in sync". The only content that stays `zh-TW`-first is *generated product output* (the `api-guide.zh-TW.md` guide and other run artifacts).
 
diff --git a/loop_apidoc/validate/response_contract.py b/loop_apidoc/validate/response_contract.py
index aef7595f..179da88e 100644
--- a/loop_apidoc/validate/response_contract.py
+++ b/loop_apidoc/validate/response_contract.py
@@ -142,10 +142,10 @@ def analyze_response_contracts(openapi: dict) -> ResponseContractAnalysis:
                     code=IssueCode.REQUIRED_INFO_MISSING,
                     severity=Severity.WARNING,
                     location=location,
-                    evidence="successful response has no usable schema contract",
+                    evidence="成功 response 沒有可用的 schema 契約",
                     suggested_fix=(
-                        "Re-read the endpoint response envelope and record its schema; "
-                        "when the source is silent, keep this delivery gap visible."
+                        "重讀這個端點的回應封包並記下它的 schema；"
+                        "若來源確實未載明，就讓這個交付缺口留在報告裡。"
                     ),
                     target_file="endpoints/",
                     field_path="responses",
diff --git a/tests/validate/test_validator.py b/tests/validate/test_validator.py
index 1eaa81bf..0a326905 100644
--- a/tests/validate/test_validator.py
+++ b/tests/validate/test_validator.py
@@ -74,9 +74,7 @@ def test_hollow_success_response_warns_at_operation_boundary():
         if issue.location == "paths./users.get" and issue.field_path == "responses"
     ]
     assert len(response_warnings) == 1
-    assert response_warnings[0].evidence == (
-        "successful response has no usable schema contract"
-    )
+    assert response_warnings[0].evidence == "成功 response 沒有可用的 schema 契約"
 
 
 def test_missing_method_makes_report_not_ok():