Your Coding Agents Just Broke – Yet Nothing in Your Repo Changed?!
This has just made another compelling case for a test-driven approach to your CLAUDE/SKILL.md changes when using coding agents at scale...
Read on → https://open.substack.com/pub/antonymarcano/p/your-coding-agents-just-broke-yet
Here’s the latest #video on the evolution of Stagentic-AI’s genesis project…
Enabling Test-Driven Development of Claude Skills…
https://open.substack.com/pub/antonymarcano/p/test-driven-claude-skill-development
Now, every test-step is carried out by a separate sub-agent.
Given – Haiku
When – Opus (agent under test)
Then – Sonnet
Finally – Haiku
Enjoy…
Just for fun, I pointed NotebookLM's AI Podcast Hosts at my latest article "Taming Claude Code, One Agentic-Test at a Time"...
It's metaphors, and insights – especially at the end – were a genuine surprise!
Listen here (article linked in shownotes):
Another NotebookLM powered podcast debate, this time about my article “Taming Claude Code, One Agentic Test at a Time”.
I was stuck in a “prompt and pray” loop, changing CLAUDE-md and SKILL-md files. Fix one behaviour, break another without even knowing. Until I realised this new problem… has an old solution.
"Taming Claude Code,—One Agentic Test at a Time"
https://open.substack.com/pub/antonymarcano/p/taming-claude-code-one-agentic-test
What we've all got wrong about guardrails for AI coding agents...
I realise now that when people say "guardrails", they usually mean "guidance". But there are true guardrails available. Combine both with gateways and not only do you have greater control over your coding agent, you improve token efficiency...
→ https://ideas.riverglide.com/what-people-get-wrong-about-agent-guardrails-80a2f7f98d73
What if you could develop your SKILL . md files in a test-driven way, just like your code? That was the question I asked myself, and the result was Test-Driven Agentic Behaviours (TDAB)...
Let me show you → https://youtu.be/i3g9ZXiaqh0
If the collective comprehension and articulation of a concept is flawed, and our individual comprehension is flawed, then we cannot expect AI to be any less flawed than the material it has to work with.
This seems to be at the heart of much of the debate in AI augmented coding.
Mutation testing is a method that's becoming increasingly important with AI-augmented coding. Without it, we could risk future bugs in our agent-written code, even if what we release today seems to work.
Read on: https://ideas.riverglide.com/mutants-in-the-machine-the-ai-illusion-of-test-driven-code-37d623642486
