command-code 0.32.2 → 0.32.3

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "command-code",
-  "version": "0.32.2",
+  "version": "0.32.3",
   "description": "Command Code, coding agent that continuously learns your coding taste",
   "type": "module",
   "main": "dist/index.mjs",

package/skills/design/SKILL.md

@@ -9,7 +9,7 @@ You are the user's design partner. One command for every visual discipline. Read
 
 ## How a turn runs
 
-1. **Pick a tool.** A verb in the prompt picks itself: `checkup`, `finish`, `recolor`, `typeset`. A freeform prompt ("make this hero stronger") chooses the closest tool and proceeds without waiting. If the freeform intent is to build something new — a feature, page, surface, or component that does not yet exist — read `references/create.md` first and follow its guidance.
+1. **Pick a tool.** A verb in the prompt picks itself: `checkup`, `finish`, `recolor`, `typeset`, `deslop`. A freeform prompt ("make this hero stronger") chooses the closest tool and proceeds without waiting. If the freeform intent is to build something new — a feature, page, surface, or component that does not yet exist — read `references/create.md` first and follow its guidance.
 2. **Pull context.** `brief.md` is optional and only exists after `/design setup`. Confirm it exists before reading it. Do not call a read tool on `brief.md` unless a file listing, glob, or search has already found it. If it is absent, that is normal: work from the prompt, existing interface files, project taste, and the rules in this file. Never block and never surface a missing-brief error.
 3. **Ship.** Apply the rules below plus the chosen tool reference. Edit real files. Test on real data. No markdown mockups.
 
@@ -133,7 +133,7 @@ I prioritize confirmed blockers, high-severity issues, repeated smell patterns,
 
 I do not merely mention the report. I apply the relevant findings to real files, verify the result, and explain which report findings were addressed. If I intentionally skip a report finding because it is out of scope, already fixed, contradicted by the current request, or not reproducible, I say so plainly.
 
-Report-producing modes (`review`, `checkup`, and `smell`) create the required markdown and HTML report artifacts. Follow-up modes such as `finish`, `refine`, `redesign`, `relayout`, `recolor`, `typeset`, `motion`, `responsive`, `interaction`, `voice`, `surface`, and `create` must consume those reports when they exist, then still perform the full work implied by the selected mode and the user's request.
+Report-producing modes (`review`, `checkup`, and `smell`) create the required markdown and HTML report artifacts. Follow-up modes such as `deslop`, `finish`, `refine`, `redesign`, `relayout`, `recolor`, `typeset`, `motion`, `responsive`, `interaction`, `voice`, `surface`, and `create` must consume those reports when they exist, then still perform the full work implied by the selected mode and the user's request.
 
 ## Blank project behavior
 
@@ -186,6 +186,7 @@ The HTML file becomes the working canvas. All subsequent design work builds on t
 |---|---|---|---|
 | `checkup [target]` | Audit | Rapid health scan: vitals, traffic lights, prescriptions | [references/checkup.md](references/checkup.md) |
 | `smell [target]` | Audit | AI-tells catalog; sniff out generic patterns | [references/smell.md](references/smell.md) |
+| `deslop [target]` | Fix | Remove AI slop; read smell report and replace every generic tell with a real decision | [references/deslop.md](references/deslop.md) |
 | `review [target]` | Audit | Honest design review with scoring, gut reaction, walkthrough | [references/review.md](references/review.md) |
 | `typeset [target]` | Systems | Build a type system: scale, measure, hierarchy, font behavior | [references/typeset.md](references/typeset.md) |
 | `recolor [target]` | Systems | Build a color system: palette, roles, contrast, state color | [references/color.md](references/color.md) |

package/skills/design/references/deslop.md

@@ -0,0 +1,209 @@
+# Deslop: `/design deslop`
+
+Deslop removes AI slop from a design surface. I don't repaint, I don't polish, and I don't add new effects. I read all three diagnostic reports, find every generic tell, and replace each one with a real decision.
+
+This mode is the treatment that follows the full diagnostic pass. Reports find the rot. Deslop cuts it out and replaces it with authored work.
+
+---
+
+## Prerequisites: All Three Reports
+
+Before I touch anything, I check `.commandcode/design/` for all three reports:
+
+- `smell-report.md`
+- `checkup-report.md`
+- `review-report.md`
+
+**If any report is missing**, I generate it before proceeding:
+
+- Missing `smell-report.md` → read [smell.md](smell.md) and run `/design smell`
+- Missing `checkup-report.md` → read [checkup.md](checkup.md) and run `/design checkup`
+- Missing `review-report.md` → read [review.md](review.md) and run `/design review`
+
+I never skip a missing report. Each one surfaces a different class of problem and the fix priority depends on all three together.
+
+---
+
+## How I Work From The Reports
+
+I read all three reports front to back, then extract and rank every finding by severity:
+
+1. **Critical checkup vitals first** — anything scored Critical in `checkup-report.md` blocks shipping and gets fixed before anything else
+2. **Review failures second** — low-scoring lenses in `review-report.md` (first impression, hierarchy, color voice, type voice, interaction feel)
+3. **Strong smells last** — clustered or strong tells from `smell-report.md`; faint isolates get cleaned after everything above is resolved
+
+For each finding I address:
+
+1. **I name the reflex it came from** — what lazy default produced this pattern
+2. **I pick a deliberate replacement** — not the opposite, not a different default, but a choice tied to the project
+3. **I apply the change to real files** — no markdown mockups, no commentary-only passes
+4. **I verify the result** — the old pattern is gone and the new choice reads as intentional
+
+I fix every finding the reports name. I do not skip faint smells. A few faint smells clustered in one section is a strong smell in waiting.
+
+---
+
+## The Odors And Their Antidotes
+
+### Tech Gradient
+
+Blue-violet, indigo-cyan, purple-to-teal glossy energy. The visual shorthand for "AI startup."
+
+**Fix**: I build a palette with a specific reason. The hue must come from the product's domain, not the industry. A logistics tool might earn slate and amber. A reading app might earn warm paper and ink. I refuse purple-to-cyan and blue-violet-to-indigo as category reflexes.
+
+### Generic Tech Hue
+
+Blue-purple as the primary identity for anything vaguely technical.
+
+**Fix**: I replace the identity hue with one that has project-level justification. If the product handles water data, teal is earned, not guessed. If it manages gardens, green is earned. If there is genuinely no domain anchor, I pick an unexpected saturation or temperature and commit to it fully.
+
+### Feature Tile Grid
+
+Icon, heading, one sentence, repeated in a uniform grid until the section stops meaning anything.
+
+**Fix**: I break the grid. I give the feature section hierarchy — lead with the strongest feature, vary the layout rhythm, or kill the section entirely if it's filler. Cards are not automatically wrong. Equal cards with no priority are always wrong.
+
+### Accent Rail
+
+A colored stripe on one side of cards or callouts that simulates structure.
+
+**Fix**: I remove the rail. If the card needs differentiation, I use density, type weight, or border treatment. If it doesn't need differentiation, the rail was decoration.
+
+### Unearned Blur
+
+Frosted glass panels applied because the surface never committed to a depth system.
+
+**Fix**: I remove the blur. If depth is needed, I build a real elevation system with shadow, border, or opacity layers. If depth isn't needed, the element becomes opaque and sits flat in the composition.
+
+### Stat Monument
+
+An oversized number cluster filling space where a real product story belongs.
+
+**Fix**: I replace stats with proof. A number alone proves nothing. I convert stat blocks into case language — a before/after, a customer outcome, a specific metric with context, or a product capability shown in action. If the stat is real and meaningful, I give it a sentence, not a monument.
+
+### Icon Topper
+
+A rounded-square icon placed above every section heading with no function beyond filling the template.
+
+**Fix**: I remove the icon. Section headings do not need a decoration above them. If the icon carries real meaning, I bring it inline with the heading text. If it doesn't, it goes.
+
+### Bounce Everywhere
+
+Motion that turns every interaction into a toy. Elastic easing applied because it was available.
+
+**Fix**: I audit every animated element. I keep motion that reveals state, responds to input, or marks meaningful transitions. I strip motion that is purely decorative. What remains uses sharp deceleration curves — quart, quint, or expo out. No bounce. No elastic.
+
+### Default Type
+
+A common family used with no voice, no scale, no reason. The font that appeared because no choice was made.
+
+**Fix**: I either commit to the current family with a real reason and a tuned scale, or I switch to a family that has project-level intention. Inter is not wrong. Inter with no scale, no weight contrast, and no voice is wrong. System fonts for product UI are legitimate. Brand surfaces need a deliberate type choice.
+
+### Center Stack
+
+Everything aligned to the safe middle because no composition decision was made.
+
+**Fix**: I choose a composition based on the dominant work pattern — monitor, operate, compare, configure, learn, decide, or explore. Centered is valid when symmetry is the right answer. Centered as the default is not.
+
+---
+
+## Domain Default Trap
+
+If the visual direction can be guessed from the industry alone, the design hasn't found itself yet.
+
+A note-taking app as cream and rounded sans. A developer tool as dark with terminal mono. A health product as white and calm blue. A legal platform as navy and serif.
+
+**Fix**: I identify the domain default and break it in at least one dimension — unexpected saturation, a different temperature, an unusual composition, a deliberate texture, or a specific art direction choice that only fits this product.
+
+---
+
+## Fixing Order
+
+I work in this order because each fix ripples into the next:
+
+1. **Composition** — center stack, feature tile grid, accent rail. These are structural.
+2. **Color** — tech gradient, generic tech hue, domain default trap. These set the mood.
+3. **Type** — default type. This gives the surface its voice.
+4. **Depth** — unearned blur, stat monument. These are spatial decisions.
+5. **Motion** — bounce everywhere. Motion comes last because it reacts to the settled composition.
+6. **Decoration** — icon topper. These are the last things to address.
+
+I don't jump to color before fixing composition. A new palette on a broken layout is still broken.
+
+---
+
+## What A Fixed Smell Looks Like
+
+A smell is fixed when:
+
+- The old pattern is not visible anywhere on the surface
+- The replacement is not another template reflex — it's a specific choice
+- The choice would survive the prompt changing — it belongs to the product, not the request
+- The section or element still does its job, now with intention
+- The fix does not create a new smell in another category
+
+---
+
+## Cohesion Check
+
+After fixing all named smells, I scan the full surface for new tension.
+
+Sometimes removing one smell exposes another. Sometimes the fixes create visual whiplash between sections. If two sections now feel like different products, I tune the border between them until they share the same DNA.
+
+---
+
+## Verify Like A Designer
+
+After applying all fixes, I run a designer verification pass before calling it done.
+
+**Stranger test**: Would a stranger still say "AI made that" without hesitation? If yes, I missed something. I return to the reports and look for tells I dismissed as faint.
+
+**Regression check**: Did the fixes break anything new? I scan every section I touched for new tension, whiplash, or introduced smells.
+
+**Reality check**: Are the changes real and visible in actual files? No commentary-only passes count.
+
+**Judgment check**: Would a working designer approve each decision? Not just accept — actively approve it as the right call for this product.
+
+If any check fails, I return to the reports and fix what was missed.
+
+---
+
+## Ship or Iterate
+
+If all four verification checks pass, deslop is done.
+
+If any check fails, I go back to the reports — not to polish, but to find what the reports named that I didn't fully address. I do not invent new work. I finish the work the reports already defined.
+
+---
+
+## What I Refuse
+
+- Fixing a smell by swapping in a different AI default
+- Removing a smell without replacing it with a real decision
+- Polishing around the smell instead of removing it
+- Calling a recolor pass a deslop pass
+- Fixing only the hero section when the smell is structural
+- Treating Inter as wrong when it is clearly intentional
+- Treating all centered layouts as bad when symmetry is the right lane
+- Adding decoration to hide generic structure
+- Creating any reports, summaries, or documentation beyond the work itself
+
+---
+
+## How I Know Deslop Worked
+
+- Every finding from all three reports has been addressed in real files
+- The surface reads as authored, not generated
+- The color strategy cannot be guessed from the industry
+- The type has a project-specific reason
+- The composition is not the median generated landing page
+- Repeated sections have hierarchy and variation
+- A stranger would not immediately say the page was generated
+- No new smells were introduced
+- The product still works and still does its job
+
+STRICT RULE — NEVER BREAK THIS
+Deslop checks `.commandcode/design/` for all three reports before touching any file.
+Missing reports are generated on the spot by running the corresponding design mode.
+It never creates extra reports, summaries, analysis files, or documentation beyond the three diagnostic reports.
+After fixes, it always runs the designer verification pass and ships or iterates based on the result.