Hi Paul! First off, I love impeccable! I use it all the time for frontend development and it has made me far more confident in developing front end (so thank you!)
One thing I noticed while auditing the always-loaded skill descriptions in my setup: Impeccable's description in skill/SKILL.src.md is around 119 words, which makes it the largest single always-loaded item in my harness by roughly 3x. Since harnesses inject every skill description into every context window, that cost is paid in every session, including ones that never touch UI work.
My read is that the enumeration style (the "design, redesign, shape, critique, audit, polish..." verb list, plus the surface and topic lists) dates from when Impeccable shipped ~20 separate skills that each needed their own trigger terms to compete for selection (the #55 era, it seems). Now that everything is consolidated into one skill, category words should cover the same trigger space, and the sub-command vocabulary already lives in argument-hint and the command table inside the skill body.
Here's a ~45-word sketch of what I mean:
Design, build, or refine any frontend interface: layout, typography, color, motion, accessibility, responsive behavior, UX copy, design systems. Use when the user wants UI designed, critiqued, or improved, asks to make a design bolder or quieter, or invokes an impeccable sub-command. Not for backend-only tasks.
I might be missing selection-accuracy data that justifies the longer form, so totally open to this being wrong. Happy to send a PR if you think it's worth pursuing.
Hi Paul! First off, I love impeccable! I use it all the time for frontend development and it has made me far more confident in developing front end (so thank you!)
One thing I noticed while auditing the always-loaded skill descriptions in my setup: Impeccable's
descriptionin skill/SKILL.src.md is around 119 words, which makes it the largest single always-loaded item in my harness by roughly 3x. Since harnesses inject every skill description into every context window, that cost is paid in every session, including ones that never touch UI work.My read is that the enumeration style (the "design, redesign, shape, critique, audit, polish..." verb list, plus the surface and topic lists) dates from when Impeccable shipped ~20 separate skills that each needed their own trigger terms to compete for selection (the #55 era, it seems). Now that everything is consolidated into one skill, category words should cover the same trigger space, and the sub-command vocabulary already lives in
argument-hintand the command table inside the skill body.Here's a ~45-word sketch of what I mean:
I might be missing selection-accuracy data that justifies the longer form, so totally open to this being wrong. Happy to send a PR if you think it's worth pursuing.