OpenClaw AI Assistant Hands-On Testing and Ink Plum Blog Updates | Week 10 Strawberry Weekly Report 2026

This article is published and updated on Cao Mei You Ren’s Blog and synchronized across multiple platforms. For any updates, the version on the blog takes precedence. You can also check the latest version via the Original Link at the end of the article.

Foreword

Welcome to Cao Mei Weekly! This is an AI-curated newsletter by Cao Mei You Ren, aiming to provide you with the latest blog updates, GitHub activities, personal updates, and recommendations from other weekly publications.

This week, I’d like to talk about OpenClaw (jokingly referred to as the “crayfish” by netizens), which has been making waves in the community lately. I jumped on the bandwagon myself, tinkering with it from installation to hands-on testing, and I have some very practical insights to share with you all.

OpenClaw is indeed popular, but honestly, its “entry barrier” isn’t low. While the Docker image simplifies the installation process significantly, configuring and successfully running it still took me quite some time.

Additionally, as a programmer, I must warn everyone: this thing poses real security risks.

Every time you integrate a plugin (such as connecting to QQ or Weibo), you’re essentially opening a new backdoor for yourself. If you plan to add it to group chats or allow multiple people to share access, even a slight misstep in permission control could render the entire environment uncontrollable. My suggestion is to set up a clean, isolated server environment dedicated solely to running OpenClaw—don’t experiment recklessly on production machines or local computers storing sensitive data. Even locally, try using a virtual machine or spare device. There’s no need to risk your assets just for novelty’s sake.

Weibo plugin details: https://gitee.com/wecode-ai/openclaw-weibo
QQ plugin details: https://q.qq.com/qqbot/openclaw/login.html

Another practical issue is—it’s incredibly “token-hungry.” Without a programming-related subscription plan, relying solely on pay-as-you-go pricing makes the cost unbearable for average users. Currently, AI assistants are more like high-powered automation tools—great at scripting and optimizing workflows, freeing us from repetitive tasks. But they’re far from invincible; the steering wheel remains firmly in your hands.

If you’re serious about subscribing, check out: MiniMax Coding Plan, GLM Coding Plan

This week, development on Mo Mei Blog has also progressed steadily.

The main focus this week was adding commercialization and ad network integration, now supporting Google AdSense or domestic ad networks like Baidu and Tencent to generate revenue.

Added support for the ActivityPub protocol, enabling social federation.

Additionally, for ASR (Automatic Speech Recognition), frontend direct connection via Volcano Engine was implemented, allowing speech recognition without backend bridging—resulting in faster response times.

After extensively using various AI tools lately, my stance remains unchanged: models determine the ceiling, tools determine the floor.

Whether it’s command-line interfaces or editor integrations, tools define the “lower bound” of collaboration, but the AI model’s capabilities directly dictate the “upper bound” of outcomes.

I’ve experimented with letting AI (e.g., GLM-5) handle coding entirely—while it’s impressively powerful, errors still accumulate over iterations in large, complex projects.

Without rigorous human review, these minor bugs can snowball into major issues. So my conclusion stands: AI is a powerful copilot, but the destination and route must always be human-driven.

In short: models set the ceiling, tools set the floor, and humans guard the baseline.

Finally, some random thoughts.

The explosion of large models in recent months has left many friends inside and outside the tech circle feeling anxious. This sentiment started brewing as early as late 2022.

But lately, I’ve found some peace with it. At its core, AI remains confined to processing tasks within computers—it can tirelessly write code but can’t decide the soul of your creation. In this era where everything can be AI-generated, if you’re not steering your project’s direction yourself, your existence as a developer loses meaning.

Therefore, to avoid being phased out prematurely, we must leverage these tools to focus on more creative tasks. The time saved should be invested in higher-level thinking.

That’s all for now. See you next time, bye~

Latest GitHub Repositories

rss-impact-next - 2026-03-04 23:54:06
An RSS-based Hook system. This project is a full-stack refactored version.

GitHub Releases

rsshub-never-die

v1.3.1 - 2026-03-03 20:39:25

Summary:
Version 1.3.1 primarily includes the following bug fixes:

Error Handling Optimization: Updated the status code types in error handling to ensure correct response statuses are returned.
Configuration Logic Simplification: Removed the priority and backup properties from NodeConfig, simplifying node configuration logic.

v1.3.0 - 2026-03-03 20:02:37

Summary:
Version 1.3.0 Highlights

This release includes the following updates:

New Features

Optimized instance selection logic, supporting mandatory, backup, and weighted configurations.

Bug Fixes

Fixed node pool capacity limits and added safeguards against empty pools during rapid responses.
Updated the src/routes/index.ts file.

caomei-auth

v1.13.0 - 2026-03-07 20:16:36

Summary:
Version 1.13.0 includes the following updates:

New Features

Frontend: Migrated frontend form validation to Zod and completed the migration of OAuth and SSO dialogs.
Validation: Fully migrated the project to Zod validation, fixed related tests, added Zod validation infrastructure, and migrated path parameter handler validation logic.

momei

v1.8.0 - 2026-03-07 20:13:17

Summary:
Version 1.8.0 includes the following updates:

Performance Optimization: Adjusted serverless scheduled task intervals from 5 minutes to 15 minutes.

New Features:

AI Image Generation: Added image generation options, multi-image preview support, and resolutions up to 1K, 2K, and 4K, along with Gemini 3.1 model integration.
ASR (Speech Recognition): Completed async task support, performance optimization, and direct integration into the voice editor.
Ad Network: Implemented domestic ad network adapters, content injection, admin UI, and foundational integration.
Other Features: Added ActivityPub protocol support, deep serverless ecosystem adaptation, external link management, Google Adsense integration, AI task timeout configuration, and admin permission verification.

Bug Fixes:

ASR: Fixed security and linting issues identified during code audits and optimized audio stream state management.
Ads & External Links: Fixed styling issues on ad management pages and improved dialog usability.
Security & Stability: Patched potential XSS vulnerabilities in shortcodes, added API rate limiting, and optimized authentication settings.
Other Fixes: Updated multilingual files, fixed federation code issues, and enhanced image provider support.

Code Refactoring: Improved code structure and query logic, removed deprecated fields and steps, and strengthened HTML processing security and timeout configuration readability.

Latest GitHub Starred Repositories

CaoMeiYouRen starred mini-swe-agent - 2026-03-07 12:48:21
This is an AI tool called SWE-agent designed to automatically resolve GitHub issues or assist via the command line. Its philosophy is extreme simplicity, requiring no complex configuration or massive codebase. The tool achieved a verified score of over 74% on the SWE-bench benchmark. Primarily developed in Python, it has garnered 3,198 stars on GitHub.
CaoMeiYouRen starred SWE-agent - 2026-03-07 12:47:46
SWE-agent is a language model-based automation tool capable of handling GitHub issues and attempting automatic fixes. It can also be applied to offensive cybersecurity or programming competition challenges. Presented at NeurIPS 2024, the project is mainly written in Python and has earned over 18,600 stars on GitHub.
CaoMeiYouRen starred ghfs - 2026-03-06 00:31:37
This project transforms GitHub issues and pull requests into a filesystem format, facilitating offline viewing and batch operations. It caters to both human users and automated agents. Developed primarily in TypeScript, it currently has 145 stars on GitHub.
CaoMeiYouRen starred 91Writing - 2026-03-04 16:07:08
This is an intelligent AI novel-writing tool built with Vue 3 and Element Plus. It integrates multiple AI models to help fiction writers improve their productivity. The main programming language is Vue, and it has received 1,305 favorites on the code hosting platform.
CaoMeiYouRen starred mcp-chrome - 2026-03-04 16:05:00
Chrome MCP Server is a Chrome extension-based Model Context Protocol server. It exposes Chrome browser functionality to AI assistants like Claude, enabling complex browser automation, content analysis, and semantic search. The project is written mainly in TypeScript and has surpassed 10,000 stars on GitHub.

Other Blogs or Weekly Recommendations

Summary

Here’s the translated content in en-US while maintaining all markdown formatting, links, and code blocks:

This week’s updates and highlights are as shown above. Thank you for reading!
You can subscribe to the Caomei Weekly updates through the following channels:

Blog: Caomei Youren’s Blog
Momei Blog: Momei Blog
RSS: Caomei Weekly
Official Account: Caomei Youren’s Backyard
Email Subscription: Caomei Youren’s Blog Subscription

Past Issues

Momei Blog 1.7.0 Release & AI Development Practice | Caomei Weekly Issue 9, 2026 - 2026-03-01 23:22:02
Momei Blog 1.4.0 Release & Open-Source Updates | Caomei Weekly Issue 6, 2026 - 2026-02-08 21:17:40
Momei Blog 1.3.0 Release & Server Data Backup Lesson | Caomei Weekly Issue 5, 2026 - 2026-02-01 23:40:42