5 pieces · 65 min of reading
Notes from the workbench. Tested on real systems.
I test AI claims on real systems and write down what holds: agent skills, prompt frameworks, and how to get real work out of the hardware you already own.
LatestIntroducing CoMPaSS: an analysis-first prompt framework for AI agents
Because a perfectly written prompt is useless if it solves the wrong problem.
The long-form case for understanding intention before instructing. CoMPaSS adds an analysis stage, a tier gate and a mandatory success spec, and publishes an evaluation plan instead of claiming wins.
- Comprehend first: name the intention and state your confidence in it.
- Sometimes the best framework is no framework; the tier gate decides.
- A prompt without a success spec has no definition of done.
- 02
Your AI agent can write code. But can it think like a principal engineer?Introducing Principal Engineer OS: judgment, scaled to risk.We automated implementation faster than we automated judgment. Principal Engineer OS gives agents seven workflows and five risk tiers so the care they take matches what a change could break. - 03
Your AI isn't a mind reader. Stop prompting like it is.Introducing ART: Act as, Request, Terms.Most disappointing outputs come from prompts that leave out the audience, the goal and the definition of done. ART treats prompting as interface design and ships as an open-source skill. - 04
Your AI agent writes code. Who's auditing its commits?Introducing the Conventional Commits Framework agent skill.Agents ship working code with commit messages like "fix: update". This short guide introduces a skill that holds every agent-written commit to one complete, atomic, co-authored contract. - 05
Claude Code quality is not coming to 8 GB of RAM, but a usable local coding agent already isStop waiting for a better laptop. The exact stack that fits on an 8 GB Ryzen machine.A field report from an HP Envy x360 with 7 GiB of usable RAM: which open-source coding agents and local models actually fit, how to wire them together, and what to realistically expect from them.