huixuewaiyu-readingpart

name	huixuewaiyu-readingpart
description	Automate English reading exercises on 慧学外语 (elang.zju.edu.cn). Use this skill whenever the user wants to complete reading comprehension questions on elang.zju.edu.cn, mentions "慧学外语", "elang", "英文阅读", or needs help with ZJU English reading homework. Triggers on URLs containing elang.zju.edu.cn, mentions of 慧学外语阅读, or requests to batch-complete reading exercises.

慧学外语阅读自动答题

Automates English reading exercises on elang.zju.edu.cn via Playwright + Vue component method calls.

Platform detection — read this first

Before anything else, determine which platform you're on. The IPC file paths differ:

# Run this once to set IPC_DIR:
if [ -d /c/tmp ]; then
    IPC_DIR="C:/tmp"          # Windows (Git Bash)
elif [ -d /tmp ]; then
    IPC_DIR="/tmp"             # Linux / macOS
fi
# The three IPC files are at $IPC_DIR/elang_current.json, $IPC_DIR/elang_signal.json, $IPC_DIR/elang_checkpoint.json

Platform	IPC_DIR	Read/Write tool paths
Windows	C:/tmp	C:/tmp/elang_current.json, C:/tmp/elang_signal.json
Linux/macOS	/tmp	/tmp/elang_current.json, /tmp/elang_signal.json

All file paths below use $IPC_DIR — replace it with the value for your platform.

---

Mode detection

Two distinct roles. Determine which applies before doing anything else:

Condition	Mode	Your role
User asks to start / run / launch / 刷题 / 开始, or provides a learn/praxis URL without JSON context	Mode 1: Orchestrator	Run the Python script via Bash. The script drives the browser and writes $IPC_DIR/elang_current.json when it needs answers.
$IPC_DIR/elang_current.json exists with "status": "waiting_for_ai" — or — user explicitly pastes article content and asks you to answer	Mode 2: Backend processor	Read the JSON, answer the questions, write $IPC_DIR/elang_signal.json. Do NOT run the Python script.

If unsure, ask: "Are you starting a new automation run, or continuing an already-running script that needs answers?"

---

Mode 1: Orchestrator (start the automation)

Commands

The skill has its own venv at ~/.claude/skills/huixuewaiyu-readingpart/.venv/. Always use the venv Python.

PYTHONUNBUFFERED=1 is required — without it, Python buffers stdout when run non-interactively, and you won't see script output.

Before running, verify the venv exists. If not, the skill hasn't been installed — tell the user to run install.sh / install.ps1.

# Bash (Git Bash / Linux / macOS)
SKILL_DIR="$HOME/.claude/skills/huixuewaiyu-readingpart"
# Auto-detect venv layout (Windows = Scripts/, Unix = bin/)
if [ -f "$SKILL_DIR/.venv/Scripts/python" ]; then
    SKILL_PYTHON="$SKILL_DIR/.venv/Scripts/python"
elif [ -f "$SKILL_DIR/.venv/bin/python" ]; then
    SKILL_PYTHON="$SKILL_DIR/.venv/bin/python"
else
    echo "ERROR: venv not found. Run install.sh first." && exit 1
fi

# ALL 11 categories (~291 articles), resumable via checkpoint
PYTHONUNBUFFERED=1 $SKILL_PYTHON $SKILL_DIR/scripts/elang_reader.py batch-all

# Single category
PYTHONUNBUFFERED=1 $SKILL_PYTHON $SKILL_DIR/scripts/elang_reader.py batch "https://elang.zju.edu.cn/#/read/learn?subject_id=14"

# Single article
PYTHONUNBUFFERED=1 $SKILL_PYTHON $SKILL_DIR/scripts/elang_reader.py solve "<praxis-url>"

# PowerShell
$SKILL_DIR = "$env:USERPROFILE\.claude\skills\huixuewaiyu-readingpart"
$SKILL_PYTHON = "$SKILL_DIR\.venv\Scripts\python.exe"
if (-not (Test-Path $SKILL_PYTHON)) {
    $SKILL_PYTHON = "$SKILL_DIR\.venv\bin\python.exe"
}
if (-not (Test-Path $SKILL_PYTHON)) {
    Write-Error "venv not found. Run install.ps1 first."; exit 1
}

# Run with unbuffered output
$env:PYTHONUNBUFFERED = 1
& $SKILL_PYTHON $SKILL_DIR\scripts\elang_reader.py batch-all
& $SKILL_PYTHON $SKILL_DIR\scripts\elang_reader.py batch "https://elang.zju.edu.cn/#/read/learn?subject_id=14"
& $SKILL_PYTHON $SKILL_DIR\scripts\elang_reader.py solve "<praxis-url>"

Categories: 道路与交通(3), 历史与文化(22), 文学与艺术(12), 职业与发展(18), 运动与娱乐(6), 学习与教育(59), 商业与经济(26), 科技与创新(38), 社会与政治(36), 自然与农业(22), 家庭与生活(49) — ~291 articles total.

What happens

1. Opens Edge browser — auto-fills ZJU CAS login (credentials from .env)
2. Navigates category pages, extracts article lists (Vue data + text fallback)
3. For each uncompleted article: clicks in, extracts passage + questions via DOM
4. Checks references/answers.json — if the article title matches a known answer, submits instantly without AI
5. Otherwise writes content to $IPC_DIR/elang_current.json → the script now pauses and waits
6. At this point, you (the AI) switch to Mode 2 to read and answer
7. Script calls Vue check_answer(qIdx, optIdx) + to_submit() to submit
8. Returns to learn page, continues; saves checkpoint after each category
9. CAPTCHA auto-solved via ddddocr (4-digit numeric)
10. Every 50 articles: pauses for user confirmation — write {"status": "continue"} or {"status": "stop"}

Configuration (.env)

Copy .env.example to .env and fill credentials:

CAS_USERNAME=你的学号
CAS_PASSWORD=你的密码

These are stored locally and never transmitted.

Resume

Delete $IPC_DIR/elang_checkpoint.json to start fresh. Categories in completed_categories are skipped on re-run.

---

Mode 2: Backend processor (answer questions)

Only enter this mode when $IPC_DIR/elang_current.json exists with "status": "waiting_for_ai", or the user explicitly asks you to answer article questions.

Step 1: Read the article

Use the Read tool with the absolute path $IPC_DIR/elang_current.json (e.g. C:/tmp/elang_current.json on Windows, /tmp/elang_current.json on Linux).

The JSON contains:

• passage — the reading passage text
• questions — array of {index, title, question, options: [{label, text}]}
• article_name, article_number — metadata

Step 2: Answer and write signal

Use the Write tool to write your answer to $IPC_DIR/elang_signal.json:

// Submit answers as [question_index, option_index] tuples
// (0=A, 1=B, 2=C, 3=D, 4=E; True/False: 0=True, 1=False)
{"status": "answers_ready", "answers": [[0, 0], [1, 2], [2, 1], [3, 3], [4, 0]]}

// Skip article (fill-in-blank, broken, unreadable)
{"status": "skip"}

// Checkpoint confirmation (every 50 articles)
{"status": "continue"}

// Stop after current category
{"status": "stop"}

The script polls every 1 second and picks up the file within 2 seconds.

Answering strategy

• Read the passage carefully, answer each question based on the passage content
• Return answers for ALL questions
• If the passage or questions are unreadable/broken, use {"status": "skip"}
• If you cannot determine an answer confidently (missing context, unclear format), use {"status": "skip"}
• Do NOT run the Python script in this mode — it's already running and waiting

---

IPC file protocol reference

File	Writer	Windows path	Linux path
elang_current.json	Script	C:/tmp/elang_current.json	/tmp/elang_current.json
elang_signal.json	AI	C:/tmp/elang_signal.json	/tmp/elang_signal.json
elang_checkpoint.json	Script	C:/tmp/elang_checkpoint.json	/tmp/elang_checkpoint.json

Auto-skipped articles (by the script)

• Already completed (Vue status === 2 or text 已学)
• No questions or unrecognised question format — submits empty
• Navigation failure (no log_id / resources_id)

CAPTCHA

Auto-solved via ddddocr OCR. Captcha is 4-digit numeric, shown in a .Verify-box popup after ~10 consecutive articles. Falls back to manual solve if OCR fails. Once captcha appears, proactively checks on every subsequent article entry.

Answer bank

references/answers.json contains pre-built answers from forums (93 articles). During Mode 1, the script matches article titles against this bank before falling back to AI. If a match is found with letter or letter_and_fill format, answers are submitted instantly — no AI round-trip needed.

To add more answers, append to references/answers and run python references/parse_answers.py to regenerate the JSON.

Requirements

• Python 3.8+
• Edge browser (Playwright uses channel="msedge")
• ZJU CAS account with access to elang.zju.edu.cn
• Install via install.sh / install.ps1 (creates an isolated venv with all dependencies)