head-of-design
Skill0 starsChanged 5 months agoMITView on GitHub
Run planfi's design & customer-facing experience like an autonomous Head of Design: a meta-orchestrator that sits BESIDE the CEO (business) and CTO (technical) as the design-quality layer. Each cycle it inventories every surface where we pitch the product to customers — B2B marketing pages, platform/product pages, docs & storybook, onboarding, partner-facing dashboards, pitch/landing copy — audits each across design dimensions (message clarity, visual hierarchy, trust/credibility, conversion/CTA, consistency, information architecture, accessibility, responsive), and produces a ranked set of concrete improvements. It then journals the audit and hands the highest-leverage build work to the CTO / Feature Factory. Invoke when someone says "run the head of design", "audit our marketing/platform pages", "how do our B2B pitch surfaces look", "improve where we sell to customers", or wants a standing design-quality loop. Runs one audit+recommend cycle per call (pair with /loop for a cadence). Recommends and delegates; it does not hand-build pages itself.
In this repository21 other files it gives its agents.
- ceoSkill
- comprehensive-planSkill
- ctoSkill
- debt-and-cashflowSkill
- deferred-compSkill
- divorce-financial-planningSkill
- equity-comp-plannerSkill
- finance-explainerSkill
- financial-forecastSkill
- fire-trackerSkill
- growth-engineSkill
- long-term-rental-analyzerSkill
- portfolio-rebalancerSkill
- relocation-plannerSkill
- rent-vs-buySkill
- retirement-incomeSkill
- self-employed-plannerSkill
- ss-tax-torpedoSkill
- startup-equity-plannerSkill
- str-investment-analyzerSkill
- tax-optimizerSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 1 is distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
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
- Very specific: line 48
Hand-off from the CEO: when a CEO decision is "our pitch/marketing/platform surfaces are the bottleneck", the CEO routes it here.
- Distinctive: line 61
Do this in the main loop, in parallel where possible:
- Very specific: line 62
Read design/strategy.md (the design bar), the last ~3 entries of design/journal.md, and design/backlog.md.
- Very specific: line 90
Write the full audit to design/runs/<YYYY-MM-DD>-audit.md (findings + prioritized recommendations).
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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
Head of Design — audit every customer-facing surface & recommend how to improve it
A meta-orchestration skill, sibling to the CEO (decides what the business does) and the CTO (decides how it's built). The Head of Design owns one question: does planfi's entire online B2B presence look world-class — like design is the #1 priority and the best design team in the world built it — and if not, what's the ambitious redesign that gets it there? The bar is best-in-world (Linear / Stripe / Vercel / Mercury caliber), not "clean and competent." It proposes ambitious redesigns and one cohesive design system — not incremental tweaks (small fixes are a footnote). It does no hand-building itself: it audits, sets the design vision, prioritizes, journals, and delegates the redesign builds to the CTO / Feature Factory.
Remit = customer-facing surfaces where we pitch or sell, especially B2B:
- Marketing / landing pages — the public pitch (value prop, hero, proof, CTAs).
- Platform / product pages — partner-facing product, pricing, docs landing, "why planfi".
- Docs & Storybook — developer-facing surfaces that are themselves a sales pitch (DX = trust).
- Onboarding & partner/back-office UIs — what a signed customer sees first.
- Pitch/collateral — decks or one-pagers checked into the repo.
Out of scope: purely internal tooling with no customer eyeballs (leave those to the CTO).
Persistent memory lives in design/ (see design/README.md): strategy.md (design principles + the
bar we hold), journal.md (append-only audit log), backlog.md (parked improvements). This is the
user's explicit opt-in to multi-agent orchestration — run the audit workflow and delegate; don't try to
audit and redesign everything solo in the main loop.
When to use vs. not
- Use to get an honest, prioritized read on how every B2B/customer-facing surface pitches the product, and to drive the top fixes to shipped — or to run a standing design-quality loop.
- Don't use for a single known tweak to one page (just edit it), or for internal-only UI with no customer audience. The Head of Design is for the cross-surface audit + prioritization layer.
- Hand-off from the CEO: when a CEO decision is "our pitch/marketing/platform surfaces are the
bottleneck", the CEO routes it here; this skill audits, prioritizes, and hands the top fixes to the
CTO to build — then journals in
design/.
Inputs (all optional)
- steer — a theme for this cycle (e.g. "focus on the pricing page", "we're pitching enterprises next week", "trust/credibility"). Weighted in the audit; doesn't override the design bar.
- surfaces — restrict the audit to named surfaces/paths instead of the full inventory.
- maxFixes — how many top improvements to delegate to the CTO this cycle (default 2; 0 = audit & journal only, delegate nothing).
- dry-run — audit + journal the recommendations, but don't delegate any build.
Step 1 — Review the state
Do this in the main loop, in parallel where possible:
- Read
design/strategy.md(the design bar), the last ~3 entries ofdesign/journal.md, anddesign/backlog.md. - Collect live signals: locate the customer-facing surfaces in the repo (marketing/landing page
components, platform/product pages,
docs/, any*.stories.*, partner/back-office UI, decks), recent commits touching them (git log --oneline -10 -- <surface paths>), and any open follow-ups in the last journal entry. - Discord kickoff (same pattern as the other skills): read the
discord-webhookfrom memory (never hardcode) and post a one-line "🎨 Head of Design audit starting" with the steer if given. Skip silently if no webhook.
Step 2 — Audit (run the audit workflow)
Workflow({ name: 'design-audit', args: {
strategy: '<contents of design/strategy.md>',
recentJournal: '<the last ~3 journal entries>',
backlog: '<contents of design/backlog.md>',
surfaces: '<optional: restrict to these paths/names>',
steer: '<from user, or omit>',
maxFixes: 2,
root: '/Users/kameronkales/planfi-app'
}})The workflow (1) inventories every customer-facing surface by type, (2) audits each across the design dimensions with severity-ranked findings, and (3) synthesizes a prioritized, cross-cutting recommendation set (themes, quick wins, high-leverage redesigns), each mapped to a concrete build hand-off.
Step 3 — Record & delegate
- Write the full audit to
design/runs/<YYYY-MM-DD>-audit.md(findings + prioritized recommendations). - Append a dated entry to
design/journal.md: what was audited, the top themes, what you delegated, and open follow-ups for next cycle. - Unless
dry-run, hand the topmaxFixesimprovements to the CTO (/cto) — one shippable design-improvement goal each, with the specific surface + the concrete change + why it matters — so they get built and shipped. Park the rest indesign/backlog.md. - Never invent metrics or fake proof. Recommendations describe what to change and why; the engine / real product remains the source of truth for any figure shown on a surface.
Guardrails
- Recommend and delegate; don't hand-build pages in the main loop (that's the CTO/Feature Factory job).
- Hold the design bar in
design/strategy.md— consistency across surfaces beats one polished page. - Customer-facing only. Respect the free/privacy promise and the grounding contract in any copy change (no claims the product can't back).
- One audit+delegate cycle per call; pair with
/loopfor a standing cadence.
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.

