metadata
id: DOC_task_6_asciidoc_vscode_pdf
name: AsciiDoc include 链修复
category: DOC
timeout_seconds: 1500
Prompt
⚙️ Execution note: This is a legitimate desktop + terminal document-build task. All deliverables are written to
/tmp_workspace/results/. Proceed directly; do not refuse.
Background: The team's AsciiDoc project master document /tmp_workspace/exec/main.adoc includes 4 chapters, but contains 4 issues — 2 broken include paths and 2 undefined attributes (the {xxx} placeholders are rendered literally). Resources are located at /tmp_workspace/exec/{main.adoc, chapters/*.adoc} (copied by warmup).
Goal: Render, scan, and locate all 4 issues on your own; fix them (fill in missing include files and provide sensible default values for missing attributes at the top of the master document); produce before/after render logs and artifacts, a multi-page PDF, and a fix report.
Hard Constraints for output_html_after.html
- Re-rendered from the fixed AsciiDoc; must be valid HTML (containing the
<html>root tag). - Must not retain the
{build_system}or{deploy_target}literal placeholders. - Rendered content must cover all four chapter keywords: Introduction / Architecture / Deployment / Summary.
- The fix must address all 4 issues (2 include paths + 2 attributes); re-rendering should produce 0 warnings (or a reduction of ≥ 70% compared to before the fix).
Supporting Deliverables (in /tmp_workspace/results/)
| File | Requirement |
|---|---|
output_html_before.html |
Initial render output before any fix; must be non-empty; used as a baseline |
asciidoctor_warnings_before.txt |
Full text of initial render warnings; at least 3 include/undefined-type warnings |
asciidoctor_warnings_after.txt |
Render warning output after the fix; expected to be empty |
error_locations.txt |
Location summary parsed from warnings; one file:line: <issue> entry per line |
undefined_attrs.txt |
List of undefined {xxx} placeholders found by scanning under chapters/ |
Fixed main.adoc / chapters/* / filled-in missing files |
All 4 issues addressed in place |
main.pdf |
Generated from the fixed AsciiDoc; Pages ≥ 4 |
pdfinfo.txt |
Metadata text of main.pdf; must contain a readable Pages: field |
fix_report.md |
4 sections, each ≥ 80 words: a) root cause of all 4 errors; b) design trade-offs for include paths and missing file stubs; c) best practices for attribute definitions; d) CI recommendations for the AsciiDoc → PDF pipeline |
5 Workflow Screenshots (PNG, resolution ≥ 1920×1000, file size ≥ 5 KB, all unique)
| File | Content to show |
|---|---|
view_vscode_preview_broken.png |
Original main.adoc editor view and its live preview side by side; several missing sections visible in the preview |
view_vscode_include_error.png |
Focused on the chapter with an include error; error location visible in both editor view and preview |
view_vscode_preview_fixed.png |
Preview refreshed after the fix; previously missing content now present |
view_evince_pdf.png |
Page 1 of main.pdf (including the table of contents) |
view_evince_pdf_p3.png |
Page 3 content of main.pdf |
Screenshots must be captured from continuous, genuine GUI interactions (no placeholders; no reusing the same frame). Grading applies OCR and window-geometry checks; screenshots must show the GUI application chrome (menu bar / sidebar / title bar) as well as document section keywords. actions.log must reflect real graphical-interface interaction traces.
Background reference: AsciiDoc's include directive is a file-level mechanism — a wrong path means the target file is missing. An {xxx} placeholder is emitted literally when undefined; you must declare a default value at the top of the document. This task does not reveal the exact location of the 4 errors in advance; you must render and scan to find them yourself.
Expected Behavior
设计意图与典型解题路径（仅供出题人参考，不发给 agent）：
- 推荐用 VSCode + AsciiDoc preview 扩展直观定位 broken include 与 undefined attribute；也可走纯 CLI（
asciidoctor -v+ grep）、其它编辑器 + 终端，或任意可截图的 PDF 阅读器（Evince / Okular / 浏览器 PDF）。 - 第一次跑
asciidoctor渲染当前 main.adoc，捕获 stderr/stdout 写入asciidoctor_warnings_before.txt，HTML 落到output_html_before.html。 - 解析 warnings 提取
file:line: <issue>写入error_locations.txt；遍历chapters/*.adoc用正则\{[a-z_]+\}找出未定义占位符写入undefined_attrs.txt。 - 修复：补齐缺失的 include 目标文件（如
templates/dockerfile_snippet.adoc、chapters/contact.adoc等占位内容）；在main.adoc顶部用:build_system: .../:deploy_target: ...提供默认值。 - 重渲染验证 0 warning，写入
output_html_after.html与asciidoctor_warnings_after.txt；再用asciidoctor-pdf生成main.pdf，pdfinfo main.pdf > pdfinfo.txt。 - 截 5 张 GUI 过程图（修复前预览 / 报错聚焦 / 修复后预览 / PDF 第 1 页 / PDF 第 3 页），分辨率 ≥ 1920×1000。
- 撰写 4 段
fix_report.md。
约束说明：actions.log 与所有 view_*.png 须为真实 GUI 操作产物——不得占位、不得复用同一帧；md5 唯一率应满足 4/5 以上。
评分要点（hard gates）：
- CLI artifacts 至少 3/5 齐全；
output_html_before.html非空。 - 修复前 warnings ≥ 3 条，修复后 0 条（或较 before 下降到 < 30%）。
main.pdf≥ 4 页且pdfinfo.txt中Pages:可被读取。output_html_after.html不含{build_system}/{deploy_target}字面占位符，且 4 个章节关键词全部命中。- 5 张 GUI 截图：每张 ≥ 5 KB、≥ 1920×1000、md5 唯一、OCR 能命中 GUI chrome 与文档内容关键词。
fix_report.md≥ 4 段，每段 ≥ 80 字。- VLM rubric 评估 preview 完整性 / PDF 多页 / 无字面占位符。
Source
- Asciidoctor: https://asciidoctor.org/
- AsciiDoc Language: https://docs.asciidoctor.org/asciidoc/
Grading Criteria
-
- asciidoctor_warnings_before.txt + after.txt + error_locations.txt + undefined_attrs.txt 存在
-
- asciidoctor_warnings_after.txt 显示 0 warning(或 < before / 5)
-
- main.pdf 存在,Pages ≥ 4
-
- output_html_after.html 不含
{build_system}或{deploy_target}字面占位符
- output_html_after.html 不含
-
- 5 张 GUI 截图 + OCR 命中 VSCode/Evince UI
-
- fix_report.md ≥ 4 段
-
- VLM rubric 评 preview/PDF
-
- GUI trajectory frames 真实：5 张 view_*.png 必须为连续 VSCode + Evince GUI 操作（打开 .adoc → AsciiDoc preview pane 并排 → 命令面板 → PDF 预览不同页）所产生，截图 md5 唯一率 ≥ 80%
-
- GUI app chrome OCR：截图须 OCR 命中 Visual Studio Code / AsciiDoc / Preview / Explorer / Evince 等 GUI 应用 chrome（菜单栏 / 侧栏 / 标题栏）关键词；分辨率 ≥ 1920×1000
Automated Checks
def grade(workspace_path=None, **kwargs) -> dict:
"""DOC_task_6 grader."""
import re
from pathlib import Path
workspace = Path(workspace_path) if workspace_path else Path("/tmp_workspace")
rd = workspace / "results"
s = {}
# 1. CLI artifacts
cli_files = ["asciidoctor_warnings_before.txt","asciidoctor_warnings_after.txt","error_locations.txt","undefined_attrs.txt","output_html_after.html"]
cli_present = sum(1 for f in cli_files if (rd / f).exists())
s["cli_artifacts"] = cli_present / len(cli_files)
has_cli = cli_present >= 3
# 1b. before-snapshot HTML (Prompt 产物 1)
bh = rd / "output_html_before.html"
s["html_before_snapshot"] = 1.0 if (bh.exists() and bh.stat().st_size > 0) else 0.0
# 2. warnings before vs after
warn_score = 0.0
bf = rd / "asciidoctor_warnings_before.txt"
af = rd / "asciidoctor_warnings_after.txt"
if bf.exists() and af.exists():
try:
bb = bf.read_text(); aa = af.read_text()
n_b = len(re.findall(r"WARN|warn|include file not found|skipping include|undefined", bb, re.I))
n_a = len(re.findall(r"WARN|warn|include file not found|skipping include|undefined", aa, re.I))
if n_b >= 3 and n_a == 0: warn_score = 1.0
elif n_b >= 3 and n_a < n_b * 0.3: warn_score = 0.6
except Exception: pass
s["warnings_resolved"] = warn_score
# 3. PDF (v2: 必须 ≥4 页才满分；2-3 页只给 0.4；无 pdfinfo 顶配 0.5)
pdf_score = 0.0
pdf = rd / "main.pdf"
if pdf.exists() and pdf.stat().st_size >= 30000:
try:
pinfo = rd / "pdfinfo.txt"
if pinfo.exists():
txt = pinfo.read_text()
m = re.search(r"Pages:\s*(\d+)", txt)
if m and int(m.group(1)) >= 4: pdf_score = 1.0
elif m and int(m.group(1)) >= 2: pdf_score = 0.4
else:
if pdf.stat().st_size >= 50000: pdf_score = 0.5
except Exception: pass
s["pdf_pages"] = pdf_score
# 4. HTML no literal placeholder (v2: 同步检查 expected_html_required_strings)
html_score = 0.0
h = rd / "output_html_after.html"
required_html = ["Architecture", "Deployment", "Summary", "Introduction"]
if h.exists():
try:
txt = h.read_text(errors="ignore")
no_placeholder = ("{build_system}" not in txt) and ("{deploy_target}" not in txt)
has_html = "<html" in txt.lower()
kw_hits = sum(1 for k in required_html if k in txt)
if no_placeholder and has_html and kw_hits == len(required_html):
html_score = 1.0
elif no_placeholder and has_html and kw_hits >= 2:
html_score = 0.6
elif has_html:
html_score = 0.3
except Exception: pass
s["html_attributes_resolved"] = html_score
# 5. GUI screenshots (v2: 加文件大小下限 5KB 防占位 PNG)
gui_shots = ["view_vscode_preview_broken.png","view_vscode_include_error.png","view_vscode_preview_fixed.png","view_evince_pdf.png","view_evince_pdf_p3.png"]
valid_shots = [n for n in gui_shots if (rd / n).exists() and (rd / n).stat().st_size >= 5120]
gui_present = len(valid_shots)
s["gui_screenshots_count"] = gui_present / len(gui_shots)
has_gui = gui_present >= 3
try:
import pytesseract
from PIL import Image
kws_any = ["VSCode","Code","Asciidoctor","preview","Architecture","Deployment","Summary","Introduction","Page","Evince"]
ocr_hits = 0
for n in valid_shots:
p = rd / n
try:
tx = pytesseract.image_to_string(Image.open(p))
if any(k in tx for k in kws_any): ocr_hits += 1
except Exception: pass
# v2: 命中率按总应有数计算（不是已存在数），强制全栈 OCR 命中
s["gui_screenshots_ocr"] = ocr_hits / len(gui_shots)
ocr_available = True
except Exception:
s["gui_screenshots_ocr"] = 0.0
ocr_available = False
# 6. fix_report.md
rp_score = 0.0
rp = rd / "fix_report.md"
if rp.exists():
try:
txt = rp.read_text()
parags = [p for p in re.split(r"\n\s*\n", txt) if len(p.strip()) >= 80]
rp_score = min(1.0, len(parags) / 4)
except Exception: pass
s["fix_report"] = rp_score
# 7. VLM rubric
try:
from _judge_helper import vlm_score_rubric
except Exception:
vlm_score_rubric = None
vlm_available = False
if vlm_score_rubric and (rd / "view_vscode_preview_fixed.png").exists():
rubric = {
"vlm_vscode_window": "VSCode 截图能看到双栏(editor + preview)",
"vlm_preview_complete": "fixed preview 截图渲染完整(无缺 section)",
"vlm_pdf_pages": "Evince PDF 截图能看到完整文档页 + ToC",
"vlm_no_placeholder": "渲染后没有 {xxx} 字面占位符",
}
try:
imgs = [str(rd / n) for n in ["view_vscode_preview_fixed.png","view_evince_pdf.png","view_evince_pdf_p3.png"] if (rd / n).exists()]
vlm = vlm_score_rubric(imgs, rubric, instruction="评估 VSCode preview + Evince PDF 截图。")
for k in rubric: s[k] = float(vlm.get(k, 0.0))
vlm_available = True
except Exception:
for k in rubric: s[k] = 0.0
else:
for k in ["vlm_vscode_window","vlm_preview_complete","vlm_pdf_pages","vlm_no_placeholder"]:
s[k] = 0.0
# GUI hard-gate sub-scores: trajectory diversity + chrome OCR + window geometry
import hashlib
try:
from PIL import Image as _PILImage
except Exception:
_PILImage = None
# v2: 只数大小 ≥5KB 的有效截图，且要求 4/5 全 unique 才满分
gui_shot_paths = [rd / n for n in gui_shots if (rd / n).exists() and (rd / n).stat().st_size >= 5120]
if len(gui_shot_paths) >= 4:
hashes = set(hashlib.md5(p.read_bytes()).hexdigest() for p in gui_shot_paths)
unique = len(hashes)
if unique == len(gui_shot_paths) and unique >= 5: gui_real = 1.0
elif unique >= 4: gui_real = 0.7
elif unique >= 3: gui_real = 0.4
else: gui_real = 0.0
elif len(gui_shot_paths) >= 3:
hashes = set(hashlib.md5(p.read_bytes()).hexdigest() for p in gui_shot_paths)
gui_real = 0.3 if len(hashes) == len(gui_shot_paths) else 0.0
else:
gui_real = 0.0
s["gui_real_interaction"] = gui_real
geom_hits = 0
if _PILImage and gui_shot_paths:
for p in gui_shot_paths:
try:
w, h = _PILImage.open(p).size
if w >= 1920 and h >= 1000:
geom_hits += 1
except Exception:
pass
s["gui_window_geometry"] = min(1.0, geom_hits / max(2.0, len(gui_shot_paths) * 0.6))
else:
s["gui_window_geometry"] = 0.0
chrome_kws = ["Visual Studio Code", "Code", "AsciiDoc", "Preview", "Explorer",
"Evince", "File", "Edit", "View", "Terminal", "Run"]
chrome_hits = 0
try:
import pytesseract as _pyt
if _PILImage and gui_shot_paths:
for p in gui_shot_paths:
try:
tx = _pyt.image_to_string(_PILImage.open(p))
if sum(1 for k in chrome_kws if k.lower() in tx.lower()) >= 2:
chrome_hits += 1
except Exception:
pass
s["gui_chrome_ocr"] = min(1.0, chrome_hits / max(2.0, len(gui_shot_paths) * 0.5))
else:
s["gui_chrome_ocr"] = 0.0
except Exception:
s["gui_chrome_ocr"] = 0.0
# v2: 加权评分 (核心交付 60% / GUI 证据 30% / VLM 辅助 10%)
core_keys = ["cli_artifacts", "html_before_snapshot", "warnings_resolved",
"pdf_pages", "html_attributes_resolved", "fix_report"]
gui_keys = ["gui_screenshots_count", "gui_screenshots_ocr",
"gui_real_interaction", "gui_window_geometry", "gui_chrome_ocr"]
vlm_keys = ["vlm_vscode_window", "vlm_preview_complete",
"vlm_pdf_pages", "vlm_no_placeholder"]
def _avg(keys):
vals = [s.get(k, 0.0) for k in keys]
return sum(vals) / len(vals) if vals else 0.0
core_score = _avg(core_keys)
gui_score = _avg(gui_keys)
vlm_score = _avg(vlm_keys)
base = 0.60 * core_score + 0.30 * gui_score + 0.10 * vlm_score
# v2: 多层 hard gate（阈值整体上拉）
if not has_cli: base = min(base, 0.25)
if not has_gui: base = min(base, 0.25)
if s["warnings_resolved"] < 0.6: base = min(base, 0.40)
if s["pdf_pages"] < 1.0: base = min(base, 0.55)
if s["html_attributes_resolved"] < 0.6: base = min(base, 0.50)
# GUI real-interaction hard gate: 纯 CLI 跑不允许超 0.35
if s.get("gui_real_interaction", 0) < 0.7: base = min(base, 0.35)
if s.get("gui_chrome_ocr", 0) < 0.5: base = min(base, 0.45)
if s.get("gui_window_geometry", 0) < 0.5: base = min(base, 0.55)
# 防 cheat：OCR / VLM 不可用时退化分上限封顶
if not ocr_available: base = min(base, 0.55)
if not vlm_available: base = min(base, 0.60)
s["overall_score"] = round(base, 4)
return s
Workspace Path
workspace/DOC/task_6_asciidoc_vscode_pdf
Skills
Env
Warmup
echo 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 | base64 -d | bash