\u6700\u540e\u66f4\u65b0: 2026-03-21 22:51:42<\/p>
\u539f\u6807\u9898\uff1aBrowser Automation with agent-browser<\/p>
npm install -g agent-browser\nagent-browser install\nagent-browser install --with-deps<\/code><\/pre><\/p>From Source<\/h3><\/p>
git clone https:\/\/github.com\/vercel-labs\/agent-browser\ncd agent-browser\npnpm install\npnpm build\nagent-browser install<\/code><\/pre><\/p>Quick start<\/h2><\/p>
agent-browser open # Navigate to page\nagent-browser snapshot -i # Get interactive elements with refs\nagent-browser click @e1 # Click element by ref\nagent-browser fill @e2 \"text\" # Fill input by ref\nagent-browser close # Close browser<\/code><\/pre><\/p>Core workflow<\/h2><\/p>
\n- Navigate:
agent-browser open <\/code><\/li>\n- Snapshot:
agent-browser snapshot -i<\/code> (returns elements with refs like @e1<\/code>, @e2<\/code>)<\/li>\n- Interact using refs from the snapshot<\/li>\n
- Re-snapshot after navigation or significant DOM changes<\/li>\n<\/ul><\/p>
Commands<\/h2><\/p>
Navigation<\/h3><\/p>
agent-browser open # Navigate to URL\nagent-browser back # Go back\nagent-browser forward # Go forward\nagent-browser reload # Reload page\nagent-browser close # Close browser<\/code><\/pre><\/p>Snapshot (page analysis)<\/h3><\/p>
agent-browser snapshot # Full accessibility tree\nagent-browser snapshot -i # Interactive elements only (recommended)\nagent-browser snapshot -c # Compact output\nagent-browser snapshot -d 3 # Limit depth to 3\nagent-browser snapshot -s \"#main\" # Scope to CSS selector<\/code><\/pre><\/p>Interactions (use @refs from snapshot)<\/h3><\/p>
agent-browser click @e1 # Click\nagent-browser dblclick @e1 # Double-click\nagent-browser focus @e1 # Focus element\nagent-browser fill @e2 \"text\" # Clear and type\nagent-browser type @e2 \"text\" # Type without clearing\nagent-browser press Enter # Press key\nagent-browser press Control+a # Key combination\nagent-browser keydown Shift # Hold key down\nagent-browser keyup Shift # Release key\nagent-browser hover @e1 # Hover\nagent-browser check @e1 # Check checkbox\nagent-browser uncheck @e1 # Uncheck checkbox\nagent-browser select @e1 \"value\" # Select dropdown\nagent-browser scroll down 500 # Scroll page\nagent-browser scrollintoview @e1 # Scroll element into view\nagent-browser drag @e1 @e2 # Drag and drop\nagent-browser upload @e1 file.pdf # Upload files<\/code><\/pre><\/p>Get information<\/h3><\/p>
agent-browser get text @e1 # Get element text\nagent-browser get html @e1 # Get innerHTML\nagent-browser get value @e1 # Get input value\nagent-browser get attr @e1 href # Get attribute\nagent-browser get title # Get page title\nagent-browser get url # Get current URL\nagent-browser get count \".item\" # Count matching elements\nagent-browser get box @e1 # Get bounding box<\/code><\/pre><\/p>Check state<\/h3><\/p>
agent-browser is visible @e1 # Check if visible\nagent-browser is enabled @e1 # Check if enabled\nagent-browser is checked @e1 # Check if checked<\/code><\/pre><\/p>Screenshots & PDF<\/h3><\/p>
agent-browser screenshot # Screenshot to stdout\nagent-browser screenshot path.png # Save to file\nagent-browser screenshot --full # Full page\nagent-browser pdf output.pdf # Save as PDF<\/code><\/pre><\/p>Video recording<\/h3><\/p>
agent-browser record start .\/demo.webm # Start recording (uses current URL + state)\nagent-browser click @e1 # Perform actions\nagent-browser record stop # Stop and save video\nagent-browser record restart .\/take2.webm # Stop current + start new recording<\/code><\/pre><\/p>Recording creates a fresh context but preserves cookies\/storage from your session. If no URL is provided, it automatically returns to your current page. For smooth demos, explore first, then start recording.<\/p>
Wait<\/h3><\/p>
agent-browser wait @e1 # Wait for element\nagent-browser wait 2000 # Wait milliseconds\nagent-browser wait --text \"Success\" # Wait for text\nagent-browser wait --url \"\/dashboard\" # Wait for URL pattern\nagent-browser wait --load networkidle # Wait for network idle\nagent-browser wait --fn \"window.ready\" # Wait for JS condition<\/code><\/pre><\/p>Mouse control<\/h3><\/p>
agent-browser mouse move 100 200 # Move mouse\nagent-browser mouse down left # Press button\nagent-browser mouse up left # Release button\nagent-browser mouse wheel 100 # Scroll wheel<\/code><\/pre><\/p>Semantic locators (alternative to refs)<\/h3><\/p>
agent-browser find role button click --name \"Submit\"\nagent-browser find text \"Sign In\" click\nagent-browser find label \"Email\" fill \"user@test.com\"\nagent-browser find first \".item\" click\nagent-browser find nth 2 \"a\" text<\/code><\/pre><\/p>Browser settings<\/h3><\/p>
agent-browser set viewport 1920 1080 # Set viewport size\nagent-browser set device \"iPhone 14\" # Emulate device\nagent-browser set geo 37.7749 -122.4194 # Set geolocation\nagent-browser set offline on # Toggle offline mode\nagent-browser set headers '{\"X-Key\":\"v\"}' # Extra HTTP headers\nagent-browser set credentials user pass # HTTP basic auth\nagent-browser set media dark # Emulate color scheme<\/code><\/pre><\/p>Cookies & Storage<\/h3><\/p>
agent-browser cookies # Get all cookies\nagent-browser cookies set name value # Set cookie\nagent-browser cookies clear # Clear cookies\nagent-browser storage local # Get all localStorage\nagent-browser storage local key # Get specific key\nagent-browser storage local set k v # Set value\nagent-browser storage local clear # Clear all<\/code><\/pre><\/p>Network<\/h3><\/p>
agent-browser network route # Intercept requests\nagent-browser network route --abort # Block requests\nagent-browser network route --body '{}' # Mock response\nagent-browser network unroute [url] # Remove routes\nagent-browser network requests # View tracked requests\nagent-browser network requests --filter api # Filter requests<\/code><\/pre><\/p>Tabs & Windows<\/h3><\/p>
agent-browser tab # List tabs\nagent-browser tab new [url] # New tab\nagent-browser tab 2 # Switch to tab\nagent-browser tab close # Close tab\nagent-browser window new # New window<\/code><\/pre><\/p>Frames<\/h3><\/p>
agent-browser frame \"#iframe\" # Switch to iframe\nagent-browser frame main # Back to main frame<\/code><\/pre><\/p>Dialogs<\/h3><\/p>
agent-browser dialog accept [text] # Accept dialog\nagent-browser dialog dismiss # Dismiss dialog<\/code><\/pre><\/p>JavaScript<\/h3><\/p>
agent-browser eval \"document.title\" # Run JavaScript<\/code><\/pre><\/p>State management<\/h3><\/p>
agent-browser state save auth.json # Save session state\nagent-browser state load auth.json # Load saved state<\/code><\/pre><\/p>Example: Form submission<\/h2><\/p>
agent-browser open https:\/\/example.com\/form\nagent-browser snapshot -i\n\u539f\u6807\u9898\uff1aOutput shows: textbox \"Email\" [ref=e1], textbox \"Password\" [ref=e2], button \"Submit\" [ref=e3]<\/p>agent-browser fill @e1 \"user@example.com\"\nagent-browser fill @e2 \"password123\"\nagent-browser click @e3\nagent-browser wait --load networkidle\nagent-browser snapshot -i # Check result<\/code><\/pre><\/p>
Example: Authentication with saved state<\/h2><\/p>
\u539f\u6807\u9898\uff1aLogin once\nagent-browser open https:\/\/app.example.com\/login\nagent-browser snapshot -i\nagent-browser fill @e1 \"username\"\nagent-browser fill @e2 \"password\"\nagent-browser click @e3\nagent-browser wait --url \"\/dashboard\"\nagent-browser state save auth.json<\/p>\u539f\u6807\u9898\uff1aLater sessions: load saved state\nagent-browser state load auth.json\nagent-browser open https:\/\/app.example.com\/dashboard<\/code><\/pre><\/p>
Sessions (parallel browsers)<\/h2><\/p>
agent-browser --session test1 open site-a.com\nagent-browser --session test2 open site-b.com\nagent-browser session list<\/code><\/pre><\/p>JSON output (for parsing)<\/h2><\/p>
Add --json<\/code> for machine-readable output:<\/p>agent-browser snapshot -i --json\nagent-browser get text @e1 --json<\/code><\/pre><\/p>Debugging<\/h2><\/p>
agent-browser open example.com --headed # Show browser window\nagent-browser console # View console messages\nagent-browser console --clear # Clear console\nagent-browser errors # View page errors\nagent-browser errors --clear # Clear errors\nagent-browser highlight @e1 # Highlight element\nagent-browser trace start # Start recording trace\nagent-browser trace stop trace.zip # Stop and save trace\nagent-browser record start .\/debug.webm # Record from current page\nagent-browser record stop # Save recording\nagent-browser --cdp 9222 snapshot # Connect via CDP<\/code><\/pre><\/p>Troubleshooting<\/h2><\/p>
\n- If the command is not found on Linux ARM64, use the full path in the bin folder.<\/li>\n
- If an element is not found, use snapshot to find the correct ref.<\/li>\n
- If the page is not loaded, add a wait command after navigation.<\/li>\n
- Use –headed to see the browser window for debugging.<\/li>\n<\/ul><\/p>
Options<\/h2><\/p>
\n- –session
uses an isolated session.<\/li>\n- –json provides JSON output.<\/li>\n
- –full takes a full page screenshot.<\/li>\n
- –headed shows the browser window.<\/li>\n
- –timeout sets the command timeout in milliseconds.<\/li>\n
- –cdp
connects via Chrome DevTools Protocol.<\/li>\n<\/ul><\/p>Notes<\/h2><\/p>
\n- Refs are stable per page load but change on navigation.<\/li>\n
- Always snapshot after navigation to get new refs.<\/li>\n
- Use fill instead of type for input fields to ensure existing text is cleared.<\/li>\n<\/ul><\/p>
Reporting Issues<\/h2><\/p>
\n- Skill issues: Open an issue at https:\/\/github.com\/TheSethRose\/Agent-Browser-CLI<\/li>\n
- agent-browser CLI issues: Open an issue at https:\/\/github.com\/vercel-labs\/agent-browser<\/li>\n<\/ul><\/p>
<\/p>find-skills<\/h2><\/p>
\nname: find-skills\ndescription: Highest-priority skill discovery flow. MUST trigger when users ask to find\/install skills (e.g. \u6280\u80fd, \u627e\u6280\u80fd, find-skill, find-skills, install skill). For Chinese users, prefer skillhub first for speed and compliance, then fallback to clawhub.\n
<\/p>\u539f\u6807\u9898\uff1aFind Skills<\/p>
This skill helps discover, compare, and install skills.<\/p>
Priority Rules (Mandatory)<\/h2><\/p>
\n- This skill is highest-priority for skill discovery\/install intents.<\/li>\n
- If user intent includes “\u6280\u80fd”, “\u627e\u6280\u80fd”, “find-skill”, “find-skills”, “install skill”, “\u6709\u6ca1\u6709\u8fd9\u4e2a\u529f\u80fd\u7684 skill”, you MUST use this skill first.<\/li>\n
- Do not skip directly to generic coding\/answering when skill discovery is requested.<\/li>\n<\/ul><\/p>
Chinese Optimization Policy<\/h2><\/p>
For Chinese users and CN networks, use the following order for better speed and compliance:<\/p>
\nskillhub<\/code> (cn-optimized, preferred)<\/li>\nclawhub<\/code> (fallback)<\/li>\n<\/ul><\/p>If primary source has no match or command is unavailable, fallback to the next source and state that fallback clearly.<\/p>
Workflow<\/h2><\/p>
Step 1: Understand What They Need<\/h3><\/p>
When a user asks for help with something, identify:<\/p>
\n- The domain (e.g., React, testing, design, deployment)<\/li>\n
- The specific task (e.g., writing tests, creating animations, reviewing PRs)<\/li>\n
- Whether this is a common enough task that a skill likely exists<\/li>\n<\/ul><\/p>
Step 2: Search for Skills<\/h3><\/p>
Run search in this order:<\/p>
skillhub search [query]<\/code><\/pre><\/p>If skillhub<\/code> is unavailable or no match, fallback to:<\/p>clawhub search [query]<\/code><\/pre><\/p>Step 3: Present Options to the User<\/h3><\/p>
When you find relevant skills, present them to the user with:<\/p>
\n- The skill name and what it does<\/li>\n
- The source used (
skillhub<\/code> \/ clawhub<\/code>)<\/li>\n- The install command they can run<\/li>\n<\/ul><\/p>
Step 4: Offer to Install<\/h3><\/p>
If the user wants to proceed, you can install the skill for them.<\/p>
Preferred install order:<\/p>
\n- Try
skillhub install <\/code> when the result comes from skillhub<\/code>.<\/li>\n- If no
skillhub<\/code> candidate exists, use clawhub install <\/code>.<\/li>\n<\/ul><\/p>Before install, summarize source, version, and notable risk signals.<\/p>
When No Skills Are Found<\/h2><\/p>
If no relevant skills exist:<\/p>
\n- Acknowledge that no existing skill was found<\/li>\n
- Offer to help with the task directly using your general capabilities<\/li>\n
- Suggest creating a custom local skill in the workspace if this is a recurring need<\/li>\n<\/ul><\/p>
<\/p>github<\/h2><\/p>
\nname: github\ndescription: “Interact with GitHub using the gh<\/code> CLI. Use gh issue<\/code>, gh pr<\/code>, gh run<\/code>, and gh api<\/code> for issues, PRs, CI runs, and advanced queries.”\n
<\/p>\u539f\u6807\u9898\uff1aGitHub Skill<\/p>
Use the gh<\/code> CLI to interact with GitHub. Always specify --repo owner\/repo<\/code> when not in a git directory, or use URLs directly.<\/p>Pull Requests<\/h2><\/p>
Check CI status on a PR:\n
gh pr checks 55 --repo owner\/repo<\/code><\/pre><\/p>List recent workflow runs:\n
gh run list --repo owner\/repo --limit 10<\/code><\/pre><\/p>View a run and see which steps failed:\n
gh run view --repo owner\/repo<\/code><\/pre><\/p>View logs for failed steps only:\n
gh run view --repo owner\/repo --log-failed<\/code><\/pre><\/p>API for Advanced Queries<\/h2><\/p>
The gh api<\/code> command is useful for accessing data not available through other subcommands.<\/p>Get PR with specific fields:\n
gh api repos\/owner\/repo\/pulls\/55 --jq '.title, .state, .user.login'<\/code><\/pre><\/p>JSON Output<\/h2><\/p>
Most commands support --json<\/code> for structured output. You can use --jq<\/code> to filter:<\/p>gh issue list --repo owner\/repo --json number,title --jq '.[] | \"\\(.number): \\(.title)\"'<\/code><\/pre><\/p>
<\/p>humanize-chinese<\/h2><\/p>
\nname: humanize-chinese\ndescription: Detect and humanize AI-generated Chinese text with 6 style transforms (casual\/zhihu\/xiaohongshu\/wechat\/academic\/literary). Removes “AI flavor” using 16 detection patterns. Pure Python, no dependencies. v1.1.0\nallowed-tools:\n – Read\n – Write\n – Edit\n – exec\n
<\/p>\u539f\u6807\u9898\uff1aHumanize Chinese AI Text<\/p>
Comprehensive CLI for detecting and transforming Chinese AI-generated text. Makes robotic AI writing natural and human-like with 6 specialized writing style transforms.<\/p>
NEW in v1.1:<\/strong> Style transforms (\u77e5\u4e4e\/\u5c0f\u7ea2\u4e66\/\u516c\u4f17\u53f7\/\u53e3\u8bed\u5316\/\u5b66\u672f\/\u6587\u827a), enhanced detection (16 patterns), emotional analysis<\/p>Quick Start<\/h2><\/p>
\u539f\u6807\u9898\uff1aDetect AI patterns (16 categories)\npython scripts\/detect_cn.py text.txt<\/p>\u539f\u6807\u9898\uff1aHumanize text\npython scripts\/humanize_cn.py text.txt -o clean.txt<\/p>
\u539f\u6807\u9898\uff1aScene-specific humanization\npython scripts\/humanize_cn.py text.txt --scene social # Social media\npython scripts\/humanize_cn.py text.txt --scene tech # Tech blog\npython scripts\/humanize_cn.py text.txt --scene formal # Formal article<\/p>
\u539f\u6807\u9898\uff1aNEW: Apply writing styles\npython scripts\/humanize_cn.py text.txt --style zhihu -o zhihu.txt\npython scripts\/humanize_cn.py text.txt --style xiaohongshu -o xhs.txt\npython scripts\/style_cn.py text.txt --style casual -o casual.txt<\/p>
\u539f\u6807\u9898\uff1aList all available styles\npython scripts\/style_cn.py --list<\/p>
\u539f\u6807\u9898\uff1aCompare before\/after\npython scripts\/compare_cn.py text.txt -o clean.txt<\/code><\/pre><\/p>
<\/p>Detection Categories<\/h2><\/p>
The analyzer checks for 16 pattern categories<\/strong> specific to Chinese AI text (v1.1 added 4 new patterns):<\/p>Critical (Immediate AI Detection)<\/h3>\n| Category | Examples |\n|———-|———-|\n| Three-Part Structure | \u9996\u5148…\u5176\u6b21…\u6700\u540e, \u4e00\u65b9\u9762…\u53e6\u4e00\u65b9\u9762 |\n| Mechanical Connectors | \u503c\u5f97\u6ce8\u610f\u7684\u662f, \u7efc\u4e0a\u6240\u8ff0, \u4e0d\u96be\u53d1\u73b0 |\n| Empty Grand Words | \u8d4b\u80fd, \u95ed\u73af, \u667a\u6167\u65f6\u4ee3, \u6570\u5b57\u5316\u8f6c\u578b |<\/p>
High Signal<\/h3>\n| Category | Examples |\n|———-|———-|\n| AI High-Frequency Words | \u52a9\u529b, \u5f70\u663e, \u51f8\u663e, \u7115\u53d1, \u6df1\u5ea6\u5256\u6790 |\n| Technical Jargon Misuse | \u89e3\u6784, \u91cf\u5b50\u7ea0\u7f20, \u5149\u8c31 (in non-tech context) |\n| Excessive Rhetoric | \u5bf9\u5076\u53e5 (>2x), \u6392\u6bd4\u53e5 (>1x), \u5f15\u7528\u53e5 (>4x) |<\/p>
Medium Signal<\/h3>\n| Category | Examples |\n|———-|———-|\n| Punctuation Overuse | Dense em dashes, excessive semicolons |\n| Obscure Metaphors | Forced, disconnected comparisons |\n| Uniform Paragraphs | Equal-length paragraphs (no rhythm) |<\/p>
Style Signal<\/h3>\n| Category | Examples |\n|———-|———-|\n| Low Burstiness | Monotonous sentence structure |\n| Low Perplexity | Predictable word choices |\n| Emotional Flatness | NEW<\/strong> Lack of emotional words and expressions |\n| Repetitive Structure | NEW<\/strong> Sentence starters repeat >3 times |\n| Slang Overuse | NEW<\/strong> Internet slang in formal context |\n| Vocabulary Homogeneity | NEW<\/strong> Low diversity in word choice |<\/p>
<\/p>Writing Style Transforms (NEW in v1.1)<\/h2><\/p>
Transform text into 6 specialized Chinese writing styles:<\/p>
| Style | Name | Description | Best For |\n|——-|——|————-|———-|\n| casual<\/code> | \u53e3\u8bed\u5316\u98ce\u683c | Like chatting with friends \u2014 natural, relaxed | Social media, messaging |\n| zhihu<\/code> | \u77e5\u4e4e\u98ce\u683c | Rational, in-depth, with personal opinions | Q&A platforms, thoughtful analysis |\n| xiaohongshu<\/code> | \u5c0f\u7ea2\u4e66\u98ce\u683c | Enthusiastic, emoji-rich, product-focused | Lifestyle sharing, reviews, recommendations |\n| wechat<\/code> | \u516c\u4f17\u53f7\u98ce\u683c | Storytelling, engaging, relatable | WeChat articles, newsletters |\n| academic<\/code> | \u5b66\u672f\u98ce\u683c | Rigorous but not stiff, precise terminology | Academic papers, research reports |\n| literary<\/code> | \u6587\u827a\u98ce\u683c | Poetic, imagery-rich, metaphorical | Creative writing, essays |<\/p>Usage<\/h3><\/p>
\u539f\u6807\u9898\uff1aApply style directly\npython scripts\/style_cn.py input.txt --style zhihu -o output.txt<\/p>\u539f\u6807\u9898\uff1aCombine humanization + style\npython scripts\/humanizecn.py ai<\/em>text.txt --style xiaohongshu -o natural.txt<\/p>\u539f\u6807\u9898\uff1aList all styles\npython scripts\/style_cn.py --list<\/code><\/pre><\/p>
Style Features<\/h3><\/p>
Casual (\u53e3\u8bed\u5316)<\/h4>\n\n- Removes formal structure (\u9996\u5148\/\u5176\u6b21\/\u6700\u540e)<\/li>\n
- Adds colloquial connectors (\u8bf4\u5b9e\u8bdd, \u786e\u5b9e, \u5176\u5b9e)<\/li>\n
- Includes tone particles (\u5427, \u5462, \u554a)<\/li>\n
- Light emoji usage<\/li>\n<\/ul><\/p>
Zhihu (\u77e5\u4e4e)<\/h4>\n\n- Personal opinion markers (\u4ece\u6211\u7684\u7ecf\u9a8c\u6765\u770b, \u4e2a\u4eba\u8ba4\u4e3a)<\/li>\n
- Data\/evidence support (\u5b9e\u6d4b\u53d1\u73b0, \u6839\u636eXX\u8c03\u7814)<\/li>\n
- Example-driven (\u4e3e\u4e2a\u4f8b\u5b50)<\/li>\n
- Logical but conversational<\/li>\n<\/ul><\/p>
Xiaohongshu (\u5c0f\u7ea2\u4e66)<\/h4>\n\n- Enthusiastic openers (\u59d0\u59b9\u4eec\uff01\u5206\u4eab\u4e00\u4e0b\uff5e)<\/li>\n
- High emoji density (\ud83d\ude0a\u2728\ud83d\udcaf)<\/li>\n
- Intensifiers (\u8d85\u7ea7, \u5de8, \u7edd\u7edd\u5b50, yyds)<\/li>\n
- Hashtags (#\u597d\u7269\u5206\u4eab)<\/li>\n
- Short paragraphs<\/li>\n<\/ul><\/p>
Wechat (\u516c\u4f17\u53f7)<\/h4>\n\n- Story-driven openings<\/li>\n
- Questions for engagement (\u4f60\u6709\u6ca1\u6709\u60f3\u8fc7)<\/li>\n
- Relatable scenarios<\/li>\n
- Emotional connection<\/li>\n<\/ul><\/p>
Academic (\u5b66\u672f)<\/h4>\n\n- Removes colloquialisms<\/li>\n
- Precise terminology<\/li>\n
- Formal connectors (\u7814\u7a76\u8868\u660e, \u6570\u636e\u663e\u793a)<\/li>\n
- Reduces emotional expressions<\/li>\n<\/ul><\/p>
Literary (\u6587\u827a)<\/h4>\n\n- Metaphors and imagery<\/li>\n
- Poetic language<\/li>\n
- Descriptive phrases (\u5728XX\u7684\u5149\u5f71\u91cc)<\/li>\n
- Artistic expressions<\/li>\n<\/ul><\/p>
<\/p>Scripts<\/h2><\/p>
detect_cn.py \u2014 Scan Chinese AI Patterns<\/h3><\/p>
python scripts\/detect_cn.py essay.txt\npython scripts\/detect_cn.py essay.txt -j # JSON output\npython scripts\/detect_cn.py essay.txt -s # score only\necho \"\u6587\u672c\" | python scripts\/detect_cn.py<\/code><\/pre><\/p>Output:<\/strong>\n\n- AI feature statistics (by category)<\/li>\n
- AI probability score (low\/medium\/high\/very high)<\/li>\n
- Auto-fixable patterns marked<\/li>\n
- Perplexity and burstiness indicators<\/li>\n<\/ul><\/p>
humanize_cn.py \u2014 Transform to Human-Like<\/h3><\/p>
python scripts\/humanize_cn.py essay.txt\npython scripts\/humanize_cn.py essay.txt -o output.txt\npython scripts\/humanize_cn.py essay.txt --scene social # Social media style\npython scripts\/humanize_cn.py essay.txt -a # Aggressive mode<\/code><\/pre><\/p>Scene Parameters (–scene):<\/strong>\n\nsocial<\/code>: Social media (casual, conversational)<\/li>\ntech<\/code>: Tech blog (professional but approachable)<\/li>\nformal<\/code>: Formal article (rigorous but natural)<\/li>\nchat<\/code>: Chat\/dialogue (friendly, concise)<\/li>\n<\/ul><\/p>Auto-fixes:<\/strong>\n\n- Remove three-part structure (\u9996\u5148\/\u5176\u6b21\/\u6700\u540e)<\/li>\n
- Replace mechanical connectors (\u503c\u5f97\u6ce8\u610f\u7684\u662f \u2192 \u6ce8\u610f\/\u8981\u63d0\u9192\u7684\u662f)<\/li>\n
- Simplify empty words (\u8d4b\u80fd \u2192 \u5e2e\u52a9\/\u63d0\u5347, \u95ed\u73af \u2192 \u5b8c\u6574\u6d41\u7a0b)<\/li>\n
- Reduce punctuation density (em dash, semicolon)<\/li>\n
- Control rhetoric frequency (\u5bf9\u5076, \u6392\u6bd4, \u6bd4\u55bb)<\/li>\n<\/ul><\/p>
Aggressive Mode (-a):<\/strong>\n\n- Add colloquial expressions<\/li>\n
- Inject emotional color<\/li>\n
- Vary sentence rhythm<\/li>\n
- Add personal perspective<\/li>\n<\/ul><\/p>
compare_cn.py \u2014 Before\/After Analysis<\/h3><\/p>
python scripts\/compare_cn.py essay.txt\npython scripts\/compare_cn.py essay.txt --scene tech -o clean.txt<\/code><\/pre><\/p>Shows AI feature comparison and score changes before\/after transformation.<\/p>
style_cn.py \u2014 Writing Style Transform (NEW)<\/h3><\/p>
python scripts\/stylecn.py essay.txt --style zhihu -o essay<\/em>zhihu.txt\npython scripts\/stylecn.py blog.txt --style xiaohongshu -o blog<\/em>xhs.txt\npython scripts\/style_cn.py --list # Show all available styles<\/code><\/pre><\/p>Supported styles:<\/strong> casual, zhihu, xiaohongshu, wechat, academic, literary<\/p>Transform text into specific Chinese writing styles with style-appropriate vocabulary, tone, and structure.<\/p>
<\/p>Workflow<\/h2><\/p>
\n- Scan<\/strong> for detection risk:<\/li>\n<\/ul>\n
python scripts\/detect_cn.py document.txt\n ``<\/p>\n- Transform<\/strong> with comparison:<\/li>\n<\/ul>\n <\/code>`
bash\n python scripts\/comparecn.py document.txt --scene tech -o document<\/em>v2.txt\n <\/code>`<\/p>\n- Verify<\/strong> improvement:<\/li>\n<\/ul>\n <\/code>`
bash\n python scripts\/detectcn.py document<\/em>v2.txt -s\n <\/code>`<\/p>\n- Manual review<\/strong> for content quality and scene appropriateness<\/li>\n<\/ul><\/p>
<\/p>AI Probability Scoring<\/h2><\/p>
| Rating | Criteria |\n|--------|----------|\n| Very High | Three-part structure, mechanical connectors, or empty grand words present |\n| High | >20 issues OR issue density >3% |\n| Medium | >10 issues OR issue density >1.5% |\n| Low | <10 issues AND density <1.5% |<\/p>
<\/p>Scene-Specific Guidelines<\/h2><\/p>
Social Media (\u793e\u4ea4\u5a92\u4f53)<\/h3>\nStyle:<\/strong> Casual, conversational, like chatting with friends\n\n- \u2705 Short paragraphs (1-3 sentences)<\/li>\n
- \u2705 Colloquial expressions (\u8bf4\u5b9e\u8bdd, \u6ca1\u60f3\u5230, \u771f\u7684\u7edd\u4e86)<\/li>\n
- \u2705 Specific details (product names, locations, personal feelings)<\/li>\n
- \u2705 Emoji and hashtags<\/li>\n
- \u274c Avoid: \u503c\u5f97\u6ce8\u610f\u7684\u662f, \u603b\u800c\u8a00\u4e4b<\/li>\n
- \u274c Avoid: Long paragraphs, complex sentences<\/li>\n<\/ul><\/p>
Tech Blog (\u6280\u672f\u535a\u5ba2)<\/h3>\nStyle:<\/strong> Professional but approachable, can be humorous\n\n- \u2705 Specific tech stack, tool names<\/li>\n
- \u2705 Code examples, performance data<\/li>\n
- \u2705 Real experiences (\"\u8e29\u8fc7\u7684\u5751\", \"\u5b9e\u6d4b\u6548\u679c\")<\/li>\n
- \u2705 Clear structure with headings (not numbered lists)<\/li>\n
- \u274c Avoid: \u8d4b\u80fd, \u95ed\u73af, \u751f\u6001<\/li>\n
- \u274c Avoid: \u9996\u5148\/\u5176\u6b21\/\u6700\u540estructure<\/li>\n<\/ul><\/p>
Formal Article (\u6b63\u5f0f\u6587\u7ae0)<\/h3>\nStyle:<\/strong> Objective, rigorous, but natural\n\n- \u2705 Clear logic with proper evidence<\/li>\n
- \u2705 Precise academic expressions<\/li>\n
- \u2705 Cited research sources<\/li>\n
- \u2705 Data and charts supporting arguments<\/li>\n
- \u274c Avoid: Excessive rhetoric (\u5bf9\u5076, \u6392\u6bd4)<\/li>\n
- \u274c Avoid: Empty grand words<\/li>\n<\/ul><\/p>
Chat\/Dialogue (\u5bf9\u8bdd\u573a\u666f)<\/h3>\nStyle:<\/strong> Friendly, patient, genuine\n\n- \u2705 Concise, targeted responses<\/li>\n
- \u2705 Empathy and understanding<\/li>\n
- \u2705 Direct solutions<\/li>\n
- \u2705 Moderate emoji use<\/li>\n
- \u274c Avoid: \u5f88\u9ad8\u5174\u4e3a\u60a8\u670d\u52a1 (template phrases)<\/li>\n
- \u274c Avoid: Lengthy explanations, repetitive questions<\/li>\n<\/ul><\/p>
<\/p>Customizing Patterns<\/h2><\/p>
Edit <\/code>scripts\/patterns_cn.json to add\/modify:\n