Greg Wilson retweeted
agent-skills 2026.10.19 is out.
I merged claude-code-workflows and my old agent-skills repo into one library: dgalarza/agent-skills.
Every skill lives once. Claude Code plugins sit on top, and any agent can install a single skill with npx skills.
damiangalarza.com/posts/2026…
Greg Wilson retweeted
I found myself constantly redirecting my agents on how to use the new stacked PRs feature in Github, so I wrote a skill around it.
Install it via npx skills add dgalarza/agent-skills --skill stacked-prs
Greg Wilson retweeted
agent-ready v1.6.1 is out.
No more `CLAUDE.md` symlink. Claude Code 2.1.277+ reads `AGENTS.md` directly; existing `CLAUDE.md` stays.
github.com/dgalarza/claude-c…
Greg Wilson retweeted
A @xai Grok Bot–style teammate needs app connectors and a cloud desktop that keeps working when your laptop closes — then the ability to drive that desktop when there’s no API.
I rebuilt that with @mastra, @composio, and @E2bDev computer use: live Notion OAuth + the agent browsing a real site in a sandbox.
youtu.be/5auulroARF0
Greg Wilson retweeted
I built an AI agent on @DocusignDevs' IAM platform that catches signed contracts before they quietly auto-renew.
It found a live example: a $125K agreement with 15 days left to cancel.
Two Mastra agents read the agreement data over Docusign MCP, a deterministic policy engine classifies the risk (not a prompt), and nothing hits Docusign Workflow Builder without a human approving it first.
Uncle Bob wrote assembly for 15 years before he moved to C. Then he read what his compiler emitted and was horrified by it.
He stayed on C and used those 15 years to write programs that corrected the output.
I don't think your experience is about to stop counting.
He's running agents the same way now, and he'll tell you he can't imagine how wrong he was about where this was going. Wrote up what he does instead, and the parts he says it doesn't fix.
imawilson.com/posts/you-do-n…
Greg Wilson retweeted
Awesome to see the Rails team performing benchmarks of different models.
rubyonrails.org/2026/8/13/ag…
Greg Wilson retweeted
New in Claude Code: your sessions can now message each other.
Instead of having to re-explain yourself in another session, you can now tell Claude to do it. It sends a summary (not your history or files), and the other session picks it up mid-task.
Evals are now wired into observability!
Scores now flow through the same event bus as your logs and metrics, instead of getting bolted onto a trace separately:
→ mastra.ai/reference/observab…
IBM just released a 1-hour course on building agentic knowledge graphs from scratch:
• 00:00 - Introduction to knowledge graphs
• 05:35 - Building your first agentic graph
• 19:59 - Agentic memory powered by graphs
• 30:39 - Graphs for multi-agent orchestration
This 1-hour watch will replace 10 paid courses on agentic engineering.
Watch it today, then learn how to become a knowledge graph engineer in the article below.
Where the series starts, code included. imawilson.com/posts/route-a-…
Greg Wilson retweeted
Coding agents have made me more bullish on Rails for web applications, not less.
I felt it while building a reliable PDF-generation and email flow in a serverless TypeScript application. The business requirement was straightforward: generate an agreement without tying up a web request, then email it reliably.
But there was no background-job path already in the system.
So the feature grew into infrastructure: a job table, a workflow runner, a cron process to find pending work, handlers, and failure semantics. An agent can help write all of that. Someone still has to review it, understand the edge cases, and own it when the next feature depends on it.
In Rails, the smallest credible version has a familiar path: controller → Active Job → PDF generation → Action Mailer.
That is not magic an agent cannot understand. It is a shared contract that gives the agent a narrower problem to solve and gives reviewers a known place to look. Rails also gives you models, migrations, MVC, a secure production baseline, and a quality gate you can make executable with `bin/ci`.
Agents make code cheap to generate. They do not make bespoke systems cheap to own.
For a new web app, I would give Rails far more weight than I did a year ago. For an existing system, audit the custom infrastructure your agents are producing and make its conventions and quality gates explicit.
The video is the fuller argument:
youtu.be/x3nPZP0HzhU
I write weekly about practical production AI engineering:
damiangalarza.com/newsletter…
How I set it up, with the dashboard JSON: imawilson.com/posts/correlat…
Greg Wilson retweeted
A "dark factory" is a production line that runs with the lights off. No humans on the floor. People are starting to describe software the same way: agents that build and ship with you out of the loop.
I don't think that's the goal. But I wanted to find where the actual line is, so this week I started building a software factory with Mastra. The idea is to add autonomy one task at a time, and only as far as each task earns it.
The first station is a Dependabot triage workflow I already run as a Codex automation, rebuilt on Mastra. Right now it's read-only: it reads the changelogs and gives me a merge-or-hold call. It doesn't merge anything yet. Under the hood that's:
1. A GitHub app that listens for webhooks on opened PRs
2. A Slack app wired into Mastra Channels, so the agent reports in and I can reply to it
3. The RubyGems API for pulling dependency changelogs to analyze
Every episode, it moves a little closer to running on its own. Where would you draw that line first?
Built on @mastra. Full walkthrough and the repo:
youtu.be/SQV6b_VvCUo
Greg Wilson retweeted
Anthropic engineer:
"You can build 5 assistants in one afternoon. Each one handles a task you've been doing manually every single day."
In 45 minutes he shows exactly how to do it from scratch, step by step.
Most people are still doing all of this by hand.
Watch the session, then save the guide below.













