tc9011 Blog

website·Indexed 31 days ago

A personal technical blog featuring study notes and tutorials on web development. Content includes detailed guides on Angular and AngularJS (MVC architecture, routing, and components), Hexo blog configuration and optimization, Echarts integration, Bootstrap components, and JavaScript DOM API notes from "JavaScript Advanced Programming".

Tokens: 203K
Snippets: 848
Records: 1.3K
Agent score: 55%

What's inside tc9011

Overview of skills-manager
skills-manager is a companion tool for vercel-labs/skills. While skills handles the installation and management of AI Agent skills, skills-manager manages the backup, synchronization, and distribution of these skills across different machines using Git and GitHub. It treats the ~/.agents/ directory as a Git repository to keep skill files and the .skill-lock.json file in sync.
https://www.tc9011.com/posts/2026/skills-manager-%E8%B7%A8%E8%AE%BE%E5%A4%87%E5%90%8C%E6%AD%A5%E4%BD%A0%E7%9A%84ai-agent-skills/
Overview of Angular Ivy Rendering Engine
Ivy is the next-generation rendering engine for Angular designed to improve the experience for users, developers, and the community. Its primary goals include:
- For Users: Faster application startup times by reducing the amount of resources the browser needs to download.
- For Developers: Simplified APIs and a more streamlined build system.
- For the Community: A more accessible compilation pipeline that facilitates third-party contributions.
Ivy is designed to be 100% backward compatible, meaning applications can upgrade to Ivy without requiring code changes while gaining performance benefits.
https://www.tc9011.com/posts/2018/ivy-angular%E4%B8%8B%E4%B8%80%E4%BB%A3%E6%B8%B2%E6%9F%93%E5%BC%95%E6%93%8E/
Overview of AI Agent Tools and Tool Calling
Tools are functions or programs that allow LLM-based applications to perform tasks beyond the model's native capabilities, such as interacting with external systems, retrieving real-time data, or executing actions. Tools act as the 'eyes' and 'hands' of an Agent. A tool definition serves as a contract between the model and the tool, requiring a clear name, parameters, and a natural language description of its purpose.
https://www.tc9011.com/posts/2026/%E8%AF%91agent%E5%B7%A5%E5%85%B7%E4%B8%8Emcp%E4%BA%92%E6%93%8D%E4%BD%9C%E6%80%A7/
Overview of the Pi coding agent
Pi is a minimal coding agent designed with a 'code-and-run' philosophy. It is the underlying engine for OpenClaw and other agentic tools. Unlike many agents that rely on pre-built plugins or MCP (Model Context Protocol), Pi encourages agents to 'learn' new capabilities by writing and executing code. Its core is intentionally small, featuring a minimal system prompt and only four built-in tools: Read, Write, Edit, and Bash.
https://www.tc9011.com/posts/2026/%E8%AF%91piopenclaw-%E5%86%85%E9%83%A8%E7%9A%84%E6%9E%81%E7%AE%80-agent/
Overview of Context Engineering for AI Agents
Context Engineering is the process of dynamically assembling and managing information within an LLM's context window to build stateful, intelligent agents. Unlike static Prompt Engineering, Context Engineering handles the entire payload based on the user, conversation history, and external data to maximize relevance and minimize noise. It involves coordinating external systems like RAG databases, session stores, and memory managers to inject the most relevant information for a specific task.
https://www.tc9011.com/posts/2026/%E8%AF%91%E4%B8%8A%E4%B8%8B%E6%96%87%E5%B7%A5%E7%A8%8B-%E4%BC%9A%E8%AF%9D%E4%B8%8E%E8%AE%B0%E5%BF%86/
Overview of Claude Code Skills
Skills are a primary extension point in Claude Code. Rather than being simple markdown files, they are folders containing scripts, assets, data, and a configuration file used to register dynamic Hooks. This structure allows the Agent to discover, explore, and operate on the contents to perform complex tasks.
https://www.tc9011.com/posts/2026/%E8%AF%91%E6%9E%84%E5%BB%BAclaude-code%E7%9A%84%E7%BB%8F%E9%AA%8C%E6%88%91%E4%BB%AC%E5%A6%82%E4%BD%95%E4%BD%BF%E7%94%A8skills/
Overview of Low Quality Image Placeholder (LQIP) implementations
LQIPs are abstract images (dominant colors, simple gradients, or blurred low-res versions) shown while a high-resolution image is loading to improve perceived performance and user experience. Common implementation methods include:
1. Low-res Bitmaps: Loading a separate small image file (increases HTTP requests).
2. Base64 Inlining: Embedding a low-res image as a base64 string in HTML/CSS (increases file size by ~33%).
3. Blurhash: Encoding a tiny version of the image into a base83 string, which is decoded via JavaScript and rendered to a <canvas> (requires JS execution and DOM readiness).
4. SQIP: Using a small SVG as a placeholder (vector-based and responsive, but still larger than a simple CSS variable).
5. Pure CSS: Using CSS variables and gradients to render placeholders (fastest, non-blocking, and minimal data transfer).
https://www.tc9011.com/posts/2026/%E8%AF%91%E7%BA%AF-css-%E5%9B%BE%E7%89%87%E5%8D%A0%E4%BD%8D%E7%AC%A6%E7%94%A8%E6%95%B0%E5%AD%A6%E5%92%8C%E9%A2%9C%E8%89%B2%E6%8C%91%E6%88%98%E8%AF%B8%E7%A5%9E/
Overview of Scrum Agile development process
Agile development typically operates in two-week cycles called 'iterations'. A standard Scrum process involves:
1. Requirement Clarification: Understanding user needs.
2. Iteration Planning: Refining core requirements, breaking work into smaller tasks, and estimating effort (e.g., person-days).
3. Daily Stand-up: A short daily meeting to discuss what was done yesterday, what is planned for today, and any blockers.
4. Review/Retrospective: Evaluating the product delivered during the iteration and identifying areas for process improvement.
https://www.tc9011.com/posts/2017/%E5%9F%BA%E4%BA%8Egithub%E7%9A%84%E6%95%8F%E6%8D%B7%E5%BC%80%E5%8F%91/
Understand the core architecture of an AI Agent
An AI Agent is a goal-oriented application that combines a language model (LM) with tools and an orchestration layer to complete tasks autonomously. Its architecture consists of four primary components:
1. Model (The Brain): The central reasoning engine (General, Fine-tuned, or Multimodal LM) that processes information, evaluates options, and makes decisions.
2. Tools (The Hands): Mechanisms that connect reasoning to the external world, such as API extensions, code functions, and data stores (databases or vector stores), allowing the agent to perform actions beyond text generation.
3. Orchestration Layer (The Nervous System): The governance process that manages the operational loop. It handles planning, memory (state), and reasoning strategies (e.g., Chain-of-Thought or ReAct) to decompose complex goals into actionable steps.
4. Deployment (The Body and Legs): The production infrastructure that makes the agent a reliable service, including secure hosting, monitoring, logging, and interfaces for user or Agent-to-Agent (A2A) API access.
https://www.tc9011.com/posts/2026/introduction-to-agents/
Choose a memory storage architecture for AI Agents
The choice of storage architecture determines how an agent retrieves memories and the type of reasoning it can perform:
- Vector Databases: Best for semantic similarity and unstructured natural language ('atomic facts'). They convert memories into embedding vectors to find conceptually close matches regardless of exact keywords.
- Knowledge Graphs: Best for structured relationship queries and complex connections ('knowledge triplets'). Memories are stored as entities (nodes) and relationships (edges), allowing the agent to traverse the graph for reasoning.
- Hybrid Approach: Combines both by enriching structured entities in a knowledge graph with vector embeddings, enabling both relational and semantic search.
https://www.tc9011.com/posts/2026/%E8%AF%91%E4%B8%8A%E4%B8%8B%E6%96%87%E5%B7%A5%E7%A8%8B-%E4%BC%9A%E8%AF%9D%E4%B8%8E%E8%AE%B0%E5%BF%86/
JavaScript maintainability and coding conventions
Maintainable code is defined by five key traits: understandability (intent is clear to others), intuitiveness (logic is obvious), adaptability (handles data changes without full rewrites), extensibility (architecture allows for future growth), and debuggability (provides sufficient info to locate errors).
General coding conventions for maintainability include:
- Readability: Use consistent indentation (typically 4 spaces instead of tabs) and comprehensive comments for functions/methods, large blocks of logic, complex algorithms, and browser-specific hacks.
- Naming: Variables should be nouns (e.g., car, person). Functions should start with verbs (e.g., getName()), and boolean-returning functions should start with is (e.g., isEnabled()).
- Type Transparency: Initialize variables upon definition to imply their data type.
```
// Specify variable type through initialization
var found = false; // Boolean
var count = 0;     // Number
```
https://www.tc9011.com/posts/2018/javascript%E9%AB%98%E7%BA%A7%E7%A8%8B%E5%BA%8F%E8%AE%BE%E8%AE%A1%E5%AD%A6%E4%B9%A0%E7%AC%94%E8%AE%B0%E4%B9%9D/
Operational challenges when moving AI Agents to production
Transitioning AI Agents from prototype to production involves a 'last mile' gap where approximately 80% of the effort is spent on infrastructure and verification rather than core intelligence. Key operational challenges include:
- Dynamic Tool Orchestration: Agents assemble trajectories on-the-fly, requiring robust versioning, access control, and observability for every unique action.
- Scalable State Management: Managing session memory and consistency across large-scale interactions is a complex system design problem.
- Unpredictable Costs and Latency: Variable execution paths make it difficult to predict response times and costs without implementing intelligent budgeting and caching.
- Security Risks: Lack of proper guardrails can lead to prompt injection (e.g., agents giving away products for free) or unauthorized access to internal databases due to misconfigured authentication.
https://www.tc9011.com/posts/2026/%E8%AF%91%E4%BB%8E%E5%8E%9F%E5%9E%8B%E5%88%B0%E7%94%9F%E4%BA%A7/