upgrade-site
Skill5 starsChanged 5 months agoMITView on GitHub
Orchestrate a full website design upgrade — snapshot the current site, research references via design-dna, synthesize a direction, score the gaps, write a prioritized plan, dispatch implementation to motion-craft and build skills, and verify. Use when asked to "upgrade the site", "make it award-winning", "run the upgrade pipeline", or "audit the site design". Vague requests get a plan, not code changes.
In this repository28 other files it gives its agents.
- AGENTS.mdAGENTS.md
- analyze-variationSkill
- arxiv-searchSkill
- blakean-sceneSkill
- codex-reviewSkill
- create-explanationSkill
- critic-loopSkill
- deepen-taskSkill
- deliver-changeSkill
- design-dnaSkill
- gloss-drift-auditSkill
- image-to-sceneSkill
- marketing-claritySkill
- market-pulseSkill
- media-kitSkill
- motion-craftSkill
- nextjs-i18n-seoSkill
- oss-licence-auditSkill
- plan-review-loopSkill
- shaughnessy-ichingSkill
- swapping-pptx-templatesSkill
- ui-ux-architectSkill
- unslopSkill
- vercel-build-speedSkill
- verify-changeSkill
- verse-to-promptSkill
- video-to-shortsSkill
- voice-calibrationSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 2 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 18
All handoffs are Markdown artifacts beneath the target repository, and all implementation is constrained by the target project's own …
- Distinctive: line 59
If the target's HEAD materially differs from the deployed site, snapshot a local build.
≈ “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.
Files
Upgrade Site (Orchestrator)
You own scope, synthesis, sequencing, and verification.
design-dna observes and abstracts; it never implements.
motion-craft implements against blueprints; it never researches references.
All handoffs are Markdown artifacts beneath the target repository, and all
implementation is constrained by the target project's own architecture and
route policies.
Authority modes
audit Snapshot, research, report only. (explicit "audit")
plan Audit + written upgrade plan. (DEFAULT — vague requests
like "make it look better")
implement Plan + modify code + verify. (explicit only: "implement",
"run the full pipeline")What each mode may write:
audit: evidence + audit report only. May NOT modify application source,
dependencies, configuration, or design docs.
plan: + blueprints, synthesis, plan documents, MOTION.md policy seed.
May NOT modify application source, dependencies, runtime config.
implement: may modify approved source and configuration.The audit/plan restriction is on source mutation, not filesystem writes —
artifacts beneath the contract paths are always permitted for the mode that
produces them. Never auto-edit .gitignore; recommend entries instead.
Target resolution
target_repo = explicit path → current git root → cwdCanonicalize; refuse / or $HOME; all writes beneath the target; log the
resolved target and chosen mode before doing anything else.
Pipeline
- Snapshot key pages (home + 2–4 representative templates) × viewports
390×844 and 1440×900 →
.artifacts/site-upgrade/<run-id>/baseline/. If the target's HEAD materially differs from the deployed site, snapshot a local build; otherwise the production site is an acceptable baseline. Record which was used. - References — invoke
design-dna(research mode unless URLs given). - Synthesize →
<target_repo>/design-dna/synthesis.mdcovering: shared patterns; meaningful disagreements; which reference leads each dimension; combinations that must not be mixed; recommended direction for the target; patterns rejected for brand/content fit. Gap analysis — score the target vs the synthesis, 0–10 per dimension: typography, color, layout, imagery, motion, brand fit — each with "what would make it a 10".
0–2 absent/harmful · 3–4 inconsistent/unfinished · 5–6 competent baseline 7–8 distinctive and coherent · 9 exceptional · 10 reference-quality (rare)Accessibility and performance are gates, not tradable dimensions — no visual ambition may lower either below the current baseline. Record the scores and each "what would make it a 10" in the plan document (or the audit report in audit mode).
- Plan →
<target_repo>/docs/plans/following the target's existing naming conventions. Order: quick wins → signature moments. Include the project-policy seed fordocs/MOTION.md: excluded high-volume templates, animation-eligible surfaces, locale list. Write the seed as a PLAIN section at the top ofdocs/MOTION.md, above any motion-craft managed sections; once written it belongs to the project (motion-craft never edits content outside its managed markers). - Dispatch (implement mode only) — motion items →
motion-craft(inherits implement authority); static design items → the project's build skills (e.g. editorial-designer / frontend-design) following the plan. - Close — re-snapshot changed pages, run a design-review pass if that skill is available, and complete motion-craft's verification gates.
Git safety (implement mode)
Same rules as motion-craft: target must be a git repo; stop on a dirty tree with unrelated changes; prefer a dedicated branch; commit only when the worktree was clean and gates passed — otherwise leave changes uncommitted with a proposed commit breakdown.
Failure behavior
A blocked reference, missing browser tooling, or an unavailable sub-skill
degrades the run (partial blueprint, N/A evidence, deferred items with
rationale) — it never halts the pipeline waiting for input in a
non-interactive run.
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
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.

