agentleFS
takimdigitaltakimdigital/saas-pipeline/skills/02b-redesign/SKILL.md

redesign-visual

Skill1 starsChanged 2 months agoMITView on GitHub

Use when the user wants an app's interface to look better, be restyled, polished, made distinctive, or given a named reference look ("Linear-style"), when it still looks like boilerplate defaults, or when they ask for a design review of an existing UI. Triggers: "make this look better", "redesign the UI", "it looks generic", "polish the design", "the spacing is off". Not for functional, backend, or deployment changes.

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 0 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read PIPELINE_STATUS.md §1-3 first (idea, stack, what was scaffolded).

    line 17
  2. Very specific:

    Commit DESIGN-NOTES.md and REVIEW.md.

    line 99
  3. Very specific:

    Delete the redesign marker.

    line 105

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Phase 2.5 — Redesign

Goal: better-looking, coherent UI with evidence, not vibes. Deterministic scripts do the checking; a design engine does the taste; a vision pass looks at the result. Everything is bounded so it cannot turn into an open-ended polish loop.

Read PIPELINE_STATUS.md §1-3 first (idea, stack, what was scaffolded). Read shared/references/note-style.md — every artifact this phase writes is note-form.

Step 0 — Preflight (three checks, in order)

node shared/scripts/setup_harness.cjs --check          # vision + screenshot plumbing
node 02b-redesign/scripts/design_context.cjs --root .  # routes, tokens, components map
  1. Vision tier (details: references/vision-and-shots.md). In order of preference: A workflow vision child — agent(prompt, { provider: 'deepseek-official', model: 'deepseek-flash' }) — works with zero configuration; B the user switches with /model to DeepSeek-V4.1-Flash (the same model at the same price); C subagent-model-selection in settings.yaml + a new session; D blind mode — gates only, and say so plainly. Record the tier in vision_route.

Marker on (Tier-1 scope guard): create <project>/.dsh/state/redesign.active. Remove it in Step 6, always — a stale marker blocks legitimate Phase 2/4 work and install_scope_hook.cjs --check reports it after 24 h.

Step 1 — Pick the engine (never invent a design process)

TierEngineNotes
1impeccable — Apache-2.0, pbakaus/impeccable, DSH-native, already on this machineimpeccable context once, then a small verb: polish (6.5 KB ref), audit (7.7 KB), layout, typeset, colorize
2anthropics/skills frontend-design — 2 files, ~9 KBVendored once when impeccable is absent; names the concrete "AI tells" to avoid
3none availableGate + design-rules.md checklist only; flag it as lower rigor in the status file

Open impeccable's new-work.md (51 KB) or critique.md (42 KB) only when the user's request names a full redesign — the words "redesign"/"overhaul", or a named new visual world. Say what it costs before loading either. Tier 3 never freehands a substitute design process.

Installing impeccable into a project: it carries a 16.8 MB engine binary — make sure .dsh/ is git-ignored first, or install it at the user root. Refresh with npx impeccable update; record version + source in design_engine_version / design_engine_source.

Removed from this pipeline: ConardLi/garden-skills web-design-engineer — 35 KB always-loaded and three blocking "🛑 actually wait" checkpoints that stall an unattended run.

Step 2 — Run the engine, then the gate

Give the engine the real codebase and docs/design/DESIGN-NOTES.md, not a description. Free-only still applies (Google Fonts / Fontsource / system stacks satisfy any paid-font intent).

node 02b-redesign/scripts/design_gate.cjs --root .      # must be clean before any eye looks

The gate is exit-coded and finds the failures agent redesigns actually have: INLINE_COLOR (a raw hex outside the token block), UNDEFINED_TOKEN (var(--x) never defined), CONTRAST (a -foreground/background pair below 3:1, or below AA 4.5:1 for body text). Known-legitimate contexts (email HTML, OG images, next/font, Radix runtime variables) are ignored by default; extend with docs/design/gate-ignore.json, or --strict to see everything. A finding a script could catch belongs in this script — add it there and it is caught for free next time.

Also: bash 02b-redesign/scripts/check_redesign_scope.sh <branch> before any commit, and the Tier-1 hook (references/enforcement.md) if it is installed.

Step 3 — Look at it (bounded: 2 rounds maximum)

Screenshots at 1440x900 and 390x844, saved under docs/design/shots/ (git-ignored). Then one vision pass writes docs/design/REVIEW.md using the template in references/vision-and-shots.md: one finding per line, always file:line, always a concrete change. Fix everything in one batch, take one confirmation round, stop.

Cost discipline: 1024 tokens per image, always a cache miss ($0.00015) because a re-render is new bytes. Re-shoot only changed routes; never keep a growing gallery in context.

Step 4 — Accessibility and structure (free, deterministic)

lighthouse_audit (browser MCP) or the project's own a11y tooling, compared against the number recorded at the start of this phase — a redesign that lowers accessibility fails, whatever it looks like. The engine's own audit verb covers the same ground when impeccable is present.

Step 5 — Commit and record

  • Work on a branch (redesign/<short>) whenever current_phase: live. Pre-launch, direct commits are fine, same as Phase 2.
  • Separate commits for: structural/token changes, component restyling, content polish.
  • PIPELINE_STATUS.md §3b: redesign_ran, direction, mode, engine + version + source, design_notes_path, review_path, shots_dir, vision_route, last_redesign_commit.
  • Commit DESIGN-NOTES.md and REVIEW.md; never commit docs/design/shots/.
  • Live app: push, CI green, user looks at the preview, then merge (Coolify deploys main).
  • Pre-deploy: hand off to 03-deploy-vps/SKILL.md when the user is happy.

Step 6 — Cleanup, always

Delete the redesign marker. Run node shared/scripts/note_lint.cjs on the artifacts you wrote, then move to Phase 3 (or Phase 4's Steps 5-6 for a live app).

What this phase does NOT do

Doesn't touch backend logic, auth, the database, dependencies or CI — the scope guard and the hook enforce that, they do not merely advise it. Doesn't re-scope features Phase 2 disabled (re-enabling one is a Phase 2/4 decision with the user). Doesn't polish past round 2. Doesn't maintain its own copy of design recipes — that is the engine's job, and a local copy would drift.

Files in this phase

  • references/vision-and-shots.md — vision tiers, screenshot policy, the bounded loop, REVIEW template
  • references/enforcement.md — Tier-1 hook + Tier-2 script, and why prose isn't enough
  • references/design-rules.md — the deterministic MUST/SHOULD/NEVER checklist (tier-3 fallback)
  • scripts/design_context.cjs — routes/tokens/components → DESIGN-NOTES.md
  • scripts/design_gate.cjs — inline colour / undefined token / contrast gate
  • scripts/check_redesign_scope.sh — Tier-2 git-diff scope gate (reads scope-policy.json)
  • scripts/scope-policy.json — the shared denylist both tiers read
  • scripts/redesign_scope_hook.cjs + shared/scripts/install_scope_hook.cjs — Tier-1 hook and its installer

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.