The Expensive Truth About AgentLock: A Write Gate Can Break Benign Workflows
Tang Weigang
Command Palette
Search for a command to run...
Archive
37 posts published so far
Letta's Architecture Has a Product Boundary Before It Has a Memory Boundary
Tang Weigang
From Agent Loop to CDP Daemon: How browser-use Actually Moves a Page
Tang Weigang
MCP Posture Starts With the Host Config, Not the Server Demo
Tang Weigang
MCP Servers: A First-Run Acceptance Test for Hosts, Capabilities, and Rollback
Tang Weigang
coding-ethos First Run: Policy Seams, Agent Hooks, and Remediation Evidence
Tang Weigang
Before an AI Agent Gets a Browser, Verify agent-workspace-linux's Isolation Contract
Tang Weigang
Sverklo Is Not Just Grep with Embeddings: Verify the Code-Memory Evidence Chain
Tang Weigang
OpenViking Is Not Just a Memory Folder: Verify Hierarchical Context Delivery First
Tang Weigang
Notion MCP: Verify Page Access Before You Trust the Tool List
Tang Weigang
Graphiti First Run: Temporal Facts, Provenance, and Hybrid Search
Tang Weigang
FastMCP: tool discovery is not authorization
Tang Weigang
Before an AI agent touches a repo, give it a verifiable context pack
Tang Weigang
Before Adopting OpenAI Agents SDK, Write the Runtime Contract First
Tang Weigang
Before Adopting DSPy, Prove the LM Program Contract
Tang Weigang
Before Adopting LlamaIndex, Split Ingestion, Retrieval, Agents, and Audit
Tang Weigang
GPT Researcher Is a Source Pipeline. First Prove the Citation Contract.
Tang Weigang
Gemini CLI Brings Gemini Into The Terminal. First Prove The Tool Boundary.
Tang Weigang
OpenHands Can Run Coding Agents. First Decide What They Are Allowed to Touch.
Tang Weigang
Serena Gives Agents Semantic Code Tools. First, Prove the Project Boundary.
Tang Weigang
FastMCP Makes MCP Easy. The First Risk Is Exposing Too Much.
Tang Weigang
smolagents Is Powerful Because It Runs Code. That Is Also the Boundary.
Tang Weigang
Browser Agents Need Receipts, Not Just Clicks: A Practical Read of browser-use
Tang Weigang
Do Not Treat Pydantic AI as an Agent Magic Layer
Tang Weigang
Before You Ship an Agent, Make DeepEval Test the Failure Path
Tang Weigang