AI Debugging in Practice: Solutions for 5 Common Mini Program Errors

70% of Development Time Is Spent Debugging — What Can AI Change?

Here's a painful statistic: new developers may only spend 30% of their time actually writing code, with the remaining 70% consumed by debugging errors. You've surely experienced this scenario — you finish writing code, confidently hit run, and the screen floods with red error messages. Then begins the frantic cycle of searching, copy-pasting, and repeated attempts, only to find the problem still persists two hours later.

This is the final episode (EP4) of the "Mini Program + AI: Ship a Product in One Day" series. The creator demonstrates how to leverage AI to clear the last mile from development to launch, using 5 real high-frequency errors encountered during development. More importantly, he distills a reusable "AI Debugging Trio" methodology that every developer should bookmark.

bilibili source: 小程序+AI一天出产品 EP4

The Core Method of AI Debugging: The Trio Template

Telling AI "I got an error" is like going to a doctor and saying "I don't feel well" — the doctor can't prescribe anything. AI debugging similarly requires precise input, specifically three things:

1. Complete Error Messages (Every Single Character)

Those messages that look like gibberish are actually the most critical clues. A Stack Trace is a complete record of the call chain when an error occurs at runtime. Starting from the exact location where the error happened, it traces back layer by layer to the original function call, containing key data such as file names, line numbers, function names, and error types. After training on massive amounts of code, AI large language models can quickly identify problem patterns in this information through pattern matching — so a complete stack trace is essentially providing AI with a precise "medical record."

Many people habitually "clean up" error messages before sending them to AI, which actually loses critical information. The correct approach is to copy and paste them exactly as they appear.

2. Relevant Code Snippets (Not Everything)

You don't need to paste the entire project, but include roughly ten lines before and after the error line so AI can see the context.

3. Your Expected Behavior

This point is easily overlooked but critically important. Telling AI "I expect clicking the button to navigate to the detail page" is far more effective than simply throwing an error message at it.

Additionally, the creator specifically recommends adding "Please explain why this problem occurred" at the end of your question. This isn't just about solving the current problem — it's about truly understanding the cause so you don't step on the same landmine next time. It's like learning to swim: you don't just want to not drown this time; you want to actually learn proper swimming technique.

Real-world comparison data: Vague questions require an average of 5 back-and-forth exchanges to solve a problem, while precise questions (using the Trio Template) basically get it done in one shot, with answer accuracy jumping from 30% to over 90%.

Five High-Frequency Errors: Practical Breakdown

These five errors are the most frequently occurring issues on the WeChat Mini Program developer forum. According to the creator's statistics, they account for nearly 60% of beginner questions combined. Solving these five gets you past the toughest hurdle for newcomers.

Notably, most of these errors aren't code logic mistakes but platform configuration issues. WeChat Mini Programs use a cloud development (Serverless) architecture that fully delegates infrastructure like servers, databases, and storage to the platform. Developers don't need to worry about operations, but the trade-off is strict adherence to platform-specific rules and limitations. Infrastructure configurations that would be handled manually in traditional development become platform rule constraints in cloud development — which is why beginners often search through their code for ages without finding the cause.

Error 1: Cloud Function Returns Error Code (-1)

This error trapped the creator for a full hour because the code itself was completely fine. After pasting the complete error to AI, it responded instantly: Your cloud function code is written, but it hasn't been deployed to the cloud. Local code and cloud are two different things — it's like writing a letter but never mailing it.

Here you need to understand how cloud functions work: Cloud Functions are a serverless computing model that runs in Tencent Cloud's Node.js environment, called from the client via the wx.cloud.callFunction interface. The key point is that cloud function code in the local development environment is just source files — it must go through an "upload and deploy" operation to be compiled and deployed to cloud containers for execution. This deployment process includes code packaging, dependency installation (npm install), and container startup, typically taking 20-60 seconds.

The fix is extremely simple: in the WeChat Developer Tools, right-click the cloud function folder, click "Upload and Deploy," and wait 30 seconds.

Error 2: Database Permission Denied

AI analysis pointed out: WeChat Cloud Database's default permission is "only creator can read/write," so anonymous users are naturally denied.

WeChat Cloud Database uses a Role-Based Access Control (RBAC) model with four preset permission levels: only creator can read/write, only creator can write but everyone can read, only admin can write but everyone can read, and all users can read/write. The default "only creator can read/write" means each data record is bound to the creator's openid, and only that user can access their own data. This design prioritizes data security, but for content that needs public display (like product listings or announcements), developers must manually adjust to a more permissive level.

Go to the cloud development console, change the permission settings, and it's immediately resolved.

These two errors would take about 2 hours to figure out on your own; using the AI debugging template took only 10 minutes.

This one is particularly tricky — it sometimes works in the development tools but breaks on real devices. AI hit the nail on the head: The parameters contain Chinese characters or special characters, and URL parameters weren't encoded, causing them to be truncated. It's like a delivery label with some characters smudged by water — of course the package can't be delivered.

URL specifications only allow a subset of safe characters from the ASCII character set. When URLs contain Chinese characters, spaces, &, = and other special characters, these may be misinterpreted as URL structural delimiters, causing parameters to be truncated or lost. This issue doesn't always reproduce in development tools because the simulator may be more lenient with URL parsing, while real device WebView kernels enforce URL specifications more strictly — which is why real device testing is so important.

Fix: Use encodeURIComponent to encode parameters when passing them, and decodeURIComponent to decode when receiving — two lines of code and done.

Error 4: wx.request Domain Not Configured

This is WeChat's security mechanism — all network request domains must be registered in the backend whitelist in advance. The key point: This configuration is in the WeChat Official Account Platform backend, not in the code. Many people search through their code for ages without finding the cause.

WeChat Mini Program's domain whitelist mechanism is a critical component of its security sandbox architecture. All network requests made through APIs like wx.request, wx.uploadFile, and wx.downloadFile must have their target domains pre-configured in "Development Management - Development Settings - Server Domains" on the WeChat Official Account Platform. Requirements include: domains must have ICP registration, must support HTTPS (TLS 1.2 and above), cannot use IP addresses or localhost, and can only be modified up to 50 times per month. During development, you can check "Do not verify valid domains" in the developer tools to bypass this restriction, but production environments require proper configuration — which is why many developers don't discover the issue during development but get errors at launch.

Error 5: Cloud Storage Path Images Not Displaying

Images uploaded successfully, the cloud:// path is obtained, but putting it in an Image tag's src just won't display the image. AI's answer was an eye-opener: The cloud:// path is a cloud internal path — like a company intranet address that can't be accessed from the external network. You need to call the getTempFileURL interface to convert it to a temporary HTTPS link for frontend display.

WeChat Cloud Storage uses the cloud:// protocol as an internal resource identifier — a proprietary protocol that can only be directly resolved in server-side environments like cloud functions. The frontend rendering engine (WebView) can only recognize standard HTTP/HTTPS protocol links. The getTempFileURL interface converts cloud:// internal paths into temporary HTTPS links with signatures and expiration periods (typically valid for several hours). This design borrows from AWS S3's Pre-signed URL mechanism, ensuring both storage resource security (unauthorized users cannot directly access resources) and flexible temporary access capabilities.

Time Review: How Much Time Does AI Debugging Actually Save?

Real records from the entire development process: originally estimated at 8 hours, actually completed in 5.5 hours, saving nearly 3 hours.

Debugging: Without AI, estimated 4-5 hours; actual was under 2 hours — this is where efficiency improvement was most dramatic
UI Development: After AI generated base code, only fine-tuning was needed, significantly shortened
Requirements Analysis: AI didn't help much here

This leads to a counterintuitive but important finding: Thinking clearly about "what to build" and "why to build it" is human work that AI cannot replace.

Hidden Checkpoint Before Launch: Privacy Policy & Review

A pitfall many people don't know about: WeChat's review team has increasingly strict requirements for privacy policies. If your mini program collects any user data without a privacy policy, it gets rejected outright.

The creator's approach was clever: describe the mini program's features to AI, have it generate a privacy policy template that meets WeChat's review requirements, make minor edits to company names, and use it directly — review passed on the first try.

Another important reminder: Never skip real device testing with the trial version. The creator discovered a layout issue during real device testing that was completely invisible in the development tools. If it had gone live without testing, that would have been a production bug.

Three Advanced Debugging Techniques

Iterative Debugging

Don't panic when new errors appear after a fix. Send the new error along with AI's previous fix to it, letting AI see the complete repair history. It will provide a more accurate next step.

Screenshot Assistance

For UI display issues, text descriptions are never as good as screenshots. Modern AI supports image understanding (multimodal capabilities) — just send a screenshot directly for more precise problem identification.

Minimal Reproduction

Strip the problematic code down to its simplest form, removing all unrelated code. Like a surgeon clearing surrounding tissue to see exactly where the lesion is.

Minimal Reproducible Example (MRE) is a classic debugging methodology in software engineering and the gold standard for asking questions on technical communities like Stack Overflow. Its core idea is to isolate the problem from complex business environments and construct the simplest code snippet that can reliably trigger the bug. This process itself has diagnostic value — while progressively removing code, developers often discover the issue on their own. For AI-assisted debugging, minimal reproduction code reduces interference from irrelevant information, allowing AI to focus more precisely on the problem's essence, dramatically improving diagnostic accuracy.

The Correct Boundaries of Human-AI Collaboration

The creator mapped AI-assisted development capabilities into a four-quadrant diagram:

AI Fully Automated Zone: Code generation, error fixing, writing documentation — extremely efficient when handed to AI
Human-AI Collaboration Zone: Architecture design, feature decomposition — you make decisions, AI executes. This is the most efficient working mode
Human-Led Zone: Requirements definition, user insights, business judgment — AI can't help here because it doesn't understand your users
Low-Efficiency Zone: Having AI conduct user interviews or market research — answers sound reasonable but often disconnect from reality

As the creator summarized: AI doesn't replace development; it lets you focus on logic and decisions. Repetitive, pattern-based work goes to AI; judgments that require truly understanding users and business must be made by you. Understanding this boundary is more important than any specific prompting technique.

Key Takeaways

The AI Debugging Trio: complete error messages + relevant code snippets + expected behavior description can boost answer accuracy from 30% to over 90%
The five most common WeChat Mini Program errors (cloud function deployment, database permissions, route encoding, domain whitelist, cloud storage paths) account for 60% of beginner questions and can be resolved in under 2 hours using AI templates instead of 4-5 hours
Total mini program development time was reduced from an estimated 8 hours to 5.5 hours, with debugging showing the most significant efficiency gains
AI-assisted development has clear boundaries: code generation and error fixing suit full automation, while requirements definition and business judgment still require human leadership
Advanced debugging techniques including iterative debugging, screenshot assistance, and minimal reproduction can further improve AI debugging efficiency