@yuhan1124/draw-prompt 0.4.10 → 0.4.11

package/README.md

@@ -135,7 +135,7 @@ overlay --image out.png --spec spec.json --out final.png      # 按 overlay spec
 visual-check --image final.png --spec spec.json               # 单图质量门：尺寸/画幅/亮度/对比度
 edit-check --reference ref.png --output edited.png            # 改图质量门：主体保留 + 有效变化
 intent-check --request "原始需求" --prompt "生成prompt"       # 检查是否新增/偏离用户未要求内容
-visual-regress references/visual-cases.jsonl                  # 开发仓库本地：多场景 prompt/成品图回归
+visual-regress references/visual-cases.jsonl                  # 开发仓库本地：多场景 prompt/成品图回归，支持 expect_/forbid_ 断言
 lint --prompt "…" [--asset-type poster] [--text 必显文字]    # 生图转化硬约束检查
 benchmark references/golden-cases.jsonl --runs 3             # 开发仓库本地：golden cases 稳定性基准
 revise --sample-id last --reason text_error                  # 按失败分类修订 Prompt
@@ -174,7 +174,7 @@ status                                            # 数据 + 下游通道健康
 4. 只有当用户明确要求“文字必须绝对准确/可后处理”或出图反馈属于 `text_error` 时，才切到 `--strict-text` + `overlay` 两段式兜底。
 5. 用 `visual-check` 验证成品图尺寸、画幅、亮度、对比度和基础细节。
 6. 参考图改图用 `edit-check` 验证“主体保留 + 背景/目标确实变化”。
-7. 模板或策略变动后，在开发仓库本地跑 `visual-regress references/visual-cases.jsonl`，确认多场景回归通过。
+7. 模板或策略变动后，在开发仓库本地跑 `visual-regress references/visual-cases.jsonl`，确认多场景回归通过；用 `expect_asset_type`、`expect_aspect`、`expect_required_text`、`forbid_required_text`、`expect_prompt_contains`、`forbid_prompt_contains` 把真实场景的产品意图固化成门禁。
 
 这条链路的默认目标不是替 gpt-image-2 重做排版引擎，而是减少跑偏、遗漏和廉价风格；
 两段式 overlay 只是文字极端稳定性兜底，不作为普通用户的默认体验。

package/SKILL.md

@@ -8,7 +8,7 @@ description: >-
   画图的指令"、"优化我的出图 prompt"、"按我的风格生成 prompt"，或在用 GPT Image 2 /
   gpt-image-2 出图前需要一段精准提示词时，使用本 skill。
 metadata:
-  version: 0.4.10
+  version: 0.4.11
   openclaw:
     anyBins: ["uv", "python3"]
 ---
@@ -168,7 +168,7 @@ overlay --image out.png --spec spec.json --out final.png      # 精确中文字/
 visual-check --image final.png --spec spec.json               # 单图质量门
 edit-check --reference ref.png --output edited.png            # 参考图改图质量门
 intent-check --request "原始需求" --prompt "生成prompt"       # 意图保真检查
-visual-regress references/visual-cases.jsonl                  # 开发仓库本地：多场景回归
+visual-regress references/visual-cases.jsonl                  # 开发仓库本地：多场景回归，支持 expect_/forbid_ 断言
 lint --prompt "…" [--asset-type poster] [--text 必显文字]    # 生图转化硬约束检查
 benchmark references/golden-cases.jsonl --runs 3             # 开发仓库本地：golden cases 转化稳定性基准
 revise --sample-id last --reason text_error                  # 按失败分类修订 Prompt

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@yuhan1124/draw-prompt",
-  "version": "0.4.10",
+  "version": "0.4.11",
   "description": "Convert natural-language image requests into high-quality gpt-image-2 prompts and Codex handoff blocks.",
   "type": "commonjs",
   "bin": {

package/scripts/prompt_cli.py

@@ -138,7 +138,7 @@ def ensure_home() -> None:
 
 
 SCHEMA_VERSION = 1
-COMPILER_VERSION = "0.4.10"
+COMPILER_VERSION = "0.4.11"
 
 
 PACKAGED_SKILL_FILES = [
@@ -2793,7 +2793,7 @@ def extract_structural_labels(request: str, asset_type: str) -> list[str]:
             value = re.split(stop_words, match.group(1), maxsplit=1, flags=re.IGNORECASE)[0]
             value = re.sub(r"[()（）]", "、", value)
             value = re.sub(r"(?:进入|再到|到|输出|->|→)", "、", value)
-            for part_match in re.finditer(r"[^、,，;；/|]+", value):
+            for part_match in re.finditer(r"[^、,，;；|]+", value):
                 part = part_match.group(0).strip(" \t\n\r，,、：:")
                 part = re.sub(r"^(?:[^：:]{0,12}(?:指标卡|卡片|列表|标题|模块|节点|步骤|栏目|分支))[：:]", "", part)
                 part = re.sub(r"^(?:和|与|及|以及|and|再|则)\s*", "", part, flags=re.IGNORECASE).strip()
@@ -4882,6 +4882,7 @@ def cmd_visual_regress(args: argparse.Namespace) -> int:
     results = []
     lint_errors = 0
     intent_errors = 0
+    expectation_errors = 0
     visual_errors = 0
     missing_images = 0
     for idx, case in enumerate(cases, start=1):
@@ -4907,8 +4908,10 @@ def cmd_visual_regress(args: argparse.Namespace) -> int:
             compiled = visual_case_compile(case)
             lint = compiled["lint"]
             intent = compiled.get("intent_check", [])
+            expectation_findings = visual_case_expectation_findings(case, compiled)
             lint_errors += sum(1 for item in lint if item.get("severity") == "error")
             intent_errors += sum(1 for item in intent if item.get("severity") == "error")
+            expectation_errors += sum(1 for item in expectation_findings if item.get("severity") == "error")
             item = {
                 "id": case_id,
                 "scenario": case.get("scenario") or case.get("tool") or "convert",
@@ -4917,6 +4920,7 @@ def cmd_visual_regress(args: argparse.Namespace) -> int:
                 "aspect": compiled["spec"]["aspect"],
                 "lint": lint,
                 "intent_check": intent,
+                "expectation_findings": expectation_findings,
                 "status": "compiled",
             }
             image = case.get("image") or case.get("output")
@@ -4951,15 +4955,20 @@ def cmd_visual_regress(args: argparse.Namespace) -> int:
         "cases": len(cases),
         "lint_error_count": lint_errors,
         "intent_error_count": intent_errors,
+        "expectation_error_count": expectation_errors,
         "visual_error_count": visual_errors,
         "missing_image_count": missing_images,
-        "pass": lint_errors == 0 and intent_errors == 0 and visual_errors == 0 and missing_images == 0,
+        "pass": lint_errors == 0 and intent_errors == 0 and expectation_errors == 0 and visual_errors == 0 and missing_images == 0,
     }
     output = {"summary": summary, "results": results}
     if args.json:
         print(json.dumps(output, ensure_ascii=False, indent=2))
     else:
-        print(f"visual-regress: cases={summary['cases']} pass={summary['pass']} lint_errors={lint_errors} intent_errors={intent_errors} visual_errors={visual_errors} missing_images={missing_images}")
+        print(
+            f"visual-regress: cases={summary['cases']} pass={summary['pass']} "
+            f"lint_errors={lint_errors} intent_errors={intent_errors} expectation_errors={expectation_errors} "
+            f"visual_errors={visual_errors} missing_images={missing_images}"
+        )
         for item in results:
             print(f"- {item['id']}: {item['status']} asset={item.get('asset_type', '?')} digest={item.get('prompt_digest', '?')}")
     return 0 if summary["pass"] else 1
@@ -5025,6 +5034,71 @@ def compile_visual_case(
     }
 
 
+def case_list(value: object) -> list[str]:
+    if value is None:
+        return []
+    if isinstance(value, list):
+        return [str(item) for item in value if str(item).strip()]
+    return [str(value)] if str(value).strip() else []
+
+
+def visual_case_expectation_findings(case: dict, compiled: dict) -> list[dict]:
+    spec = compiled.get("spec") or {}
+    prompt = str(compiled.get("prompt") or "")
+    findings: list[dict] = []
+
+    expected_asset = case.get("expect_asset_type")
+    if expected_asset and spec.get("asset_type") != expected_asset:
+        findings.append(
+            {
+                "severity": "error",
+                "rule": "case.asset_type_mismatch",
+                "message": f"期望 asset_type={expected_asset}，实际={spec.get('asset_type')}",
+            }
+        )
+
+    expected_aspect = case.get("expect_aspect")
+    if expected_aspect and spec.get("aspect") != expected_aspect:
+        findings.append(
+            {
+                "severity": "error",
+                "rule": "case.aspect_mismatch",
+                "message": f"期望 aspect={expected_aspect}，实际={spec.get('aspect')}",
+            }
+        )
+
+    expected_template = case.get("expect_template_id")
+    if expected_template and spec.get("template_id") != expected_template:
+        findings.append(
+            {
+                "severity": "error",
+                "rule": "case.template_mismatch",
+                "message": f"期望 template_id={expected_template}，实际={spec.get('template_id')}",
+            }
+        )
+
+    labels = set(spec.get("required_text") or [])
+    for text in case_list(case.get("expect_required_text")):
+        if text not in labels:
+            findings.append({"severity": "error", "rule": "case.required_text_missing", "message": f"缺少必显文字：{text}"})
+        elif f'"{text}"' not in prompt and not spec.get("strict_text"):
+            findings.append({"severity": "error", "rule": "case.required_text_not_quoted", "message": f"Prompt 未逐字引用：{text}"})
+
+    for text in case_list(case.get("forbid_required_text")):
+        if text in labels:
+            findings.append({"severity": "error", "rule": "case.required_text_forbidden", "message": f"不应作为必显文字：{text}"})
+
+    for text in case_list(case.get("expect_prompt_contains")):
+        if text not in prompt:
+            findings.append({"severity": "error", "rule": "case.prompt_missing", "message": f"Prompt 缺少片段：{text}"})
+
+    for text in case_list(case.get("forbid_prompt_contains")):
+        if text in prompt:
+            findings.append({"severity": "error", "rule": "case.prompt_forbidden", "message": f"Prompt 不应包含片段：{text}"})
+
+    return findings
+
+
 def has_lint_error(items: list[dict]) -> bool:
     return any(item.get("severity") == "error" for item in items)