Imagine connecting an agent to a support system.
Badges
Badges come from GitHub once the author claims this feed.
Recent posts
every link opens the author’s post-
Your MCP Tool Catalog Is Part of the Permission Boundary
-
UTCP Compared With MCP for Existing APIs
I came across UTCP in a YouTube video.
-
Your Agent’s Sandbox Shouldn’t Hold Its Model Credentials
An agent that can run a shell command doesn’t need the credential used to authenticate its next model request.
-
One Stray Exclamation Point From v1.0.0
The first sign something was wrong was the version number: the major had bumped before I was ready for one.
-
How OpenRouter Pareto Code Chooses a Coding Model
OpenRouter’s Pareto Code Router lets an application request a level of coding ability without naming a specific model.
-
Baking Migrations Into a Postgres CI Image
The reason I wanted a prebuilt database image was how long the tests were taking in CI.
-
AWS WAF and Coraza Compared on Human Tuning Effort
I’m interested in whether AWS WAF would take less human effort than Coraza.
-
Moving My Private Projects from GitHub Actions to Semaphore CI
I moved CI for some of my private projects, including Logan.center, from GitHub Actions to Semaphore CI.
-
Biome Formatting for Markdown and YAML
Biome is my default for linting and formatting these days with TypeScript.
-
GitHub Agentic Workflows and the Limits I'd Set
If an agent investigates a failed build and discovers that the fix requires authentication changes or a database migration, I want it to stop.
-
What Mem0 Dream could add to my memory workflow
I built a small MCP server in Go so my coding harness can work with Mem0 in my home lab.
-
A Useful Starting Point for Hybrid Search
While updating Semantic Docs and a private fork I call Astro Vault, I wanted to understand how hybrid search fits together.
-
Project Quarry for Agent Security
Cisco has open sourced Project Quarry, a Python framework and rule language for detecting risky behavior in AI agent conversations.
-
Google's Elephant-Goldfish Model and Orca Orchestration
In A Week With Orca and Coordination Between Harnesses, I wrote about how well Claude Code and Codex have been working together for me.
-
A Week With Orca and Coordination Between Harnesses
I started using Orca near end of October, and it’s been awesome.
-
DuckDB and the Agent Using Your Data Tools
DuckDB 2.0 has a feature I find more interesting than another chatbot interface: a command-line mode for the agent already using your tools.
-
Logan Logger compared with Pino
I released Logan Logger as a TypeScript logging package a while ago.
-
How Fusion Weights Affect Hybrid Search
I’ve been working with Claude on search for one of my documentation sites and asking it to look for improvements.
-
What the eMudhra Incident Means for Node.js Applications
The intermediate CA’s private key was 4. What does that actually mean?
-
PHP Security and the Shrinking Patch Window
October was an interesting month for web security.
-
Moving the Loop from Python to Rust
I wanted to understand how to use Rust with Python and what, if any, were the performance trade-offs.
-
Asking a Model Router for Advice Gets You Another Model's Answer
I wanted a cheap “System 1” model to tell me which model and reasoning effort each of my Claude Code subagents should run on.
-
Why I Built Agent Sync
I built Agent Sync to keep one collection of agent definitions for coding assistants and choose which ones belong in each project.
-
Finding a Favorite Color, Two at a Time
Color Pick started with a small idea: show someone two colors, ask which one they prefer, and keep narrowing toward their favorite shade.
-
A Theme Editor Probably Doesn't Need a CSS-in-JS Runtime
I made Styled Lab, a small Svelte playground where you can change styles and watch the generated CSS update.
-
Safer Answers and Healthier Users
One line from a conversation with my AI assistant stuck with me: customer harm is not a line of business you want to be in.
-
Project Summaries Before Repository Access
I have a lot of repositories, and keeping track of them is becoming a project of its own.
-
A Check That Runs Is Not a Check That Blocks
The release workflow for one of my projects can no longer decide on a major version by itself.
-
Let Me Check the Box on My Phone
Pages is one of the most useful parts of Space in ChatGPT for me.
-
Liar Liar (1997)
90s Jim Carrey. What more is there to say?