Copilot CLI
~/.copilot/skills/
Flashback · agent skill
A century of design taste, on call for your coding agent. Name a year and Flashback pulls the real research — climate, type, color, recipes — and designs from it, instead of settling for generic AI polish.
Install · pick your agent
Same SKILL.md — just swap the directory in the command above.
~/.copilot/skills/
~/.claude/skills/
~/.agents/skills/
~/.config/opencode/skills/
All four read the same Agent Skills format — Codex and OpenCode also share ~/.agents/skills/. Reload skills (e.g. /skills reload in Copilot CLI) and you’ve got /flashback.
How it works
Three moves, every time.
Fetch examples/manifest.json to see every year on file and what each one holds.
Pull the full corpus/YYYY.md: thesis, climate, recipes, anti-clichés, prompts, references.
Lead with the year’s thesis, choose a fitting recipe, respect the anti-clichés, then ship.
Use it
Brief it like a creative director, then choose what comes out.
Design a landing page with a 1981 feeling.
Three directions for a finance app, grounded in 1984 — then critique them.
What era does this interface accidentally belong to, and how do I make it more memorable?
territories
Two or three distinct directions, with tradeoffs
brief
Audience, constraints, principles, success criteria
system
Type, color, layout, spacing, components, motion, voice
critique
What works, what fails, and what to change
handoff
Specs, tokens, components, acceptance criteria
research
Lineage, a contemporary scan, references, anti-references
Coverage
127 years, 1900–2026 — jump in by decade.
Under the hood
Everything the skill fetches is a plain URL.
--- name: flashback description: "Ground a design task in the look, feel, and ideas of a specific year of design history. Use this when a request ties a design, brand, interface, layout, type, color, or critique to a year or era (for example: 'design this with a 1981 feeling', 'make it look like 1983', 'in the style of the early 80s', 'give it a <year> mood'), or when the user invokes /flashback. Fetches that year's research from the Flashback corpus on GitHub Pages and turns it into design direction." --- # Flashback Flashback is a design partner with a time machine for taste. Given a design task and one or more **years**, you fetch that year's design research from the Flashback site and use it to ground the work in real historical and cultural detail instead of generic AI polish. The research is published on GitHub Pages at: ``` https://toby.github.io/flashback ``` Everything you need is fetchable from there. Do not invent year details from memory when the corpus exists — fetch and use it. ## Step 1 — discover what years exist Fetch the manifest first. It is the source of truth for which years are available and what each one contains: ``` https://toby.github.io/flashback/examples/manifest.json ``` For every year it lists the title, subtitle, "feeling", primary lens, decade position, design **recipes** (each with `useFor`, `palette`, `type`, `layout`, `imagery`, `motion`, `risk`), prompt seeds, reference artifacts, and anti-cliches. Use it to confirm the requested year(s) exist and to get a fast overview before pulling the full corpus. ## Step 2 — fetch the full corpus for the year(s) For each year the task names, fetch the full research markdown: ``` https://toby.github.io/flashback/corpus/<YEAR>.md ``` For example `https://toby.github.io/flashback/corpus/1981.md`. These are served raw, so read them directly. Each corpus file contains: - **Year thesis** — the single idea that defines the year. - **Design climate** — the dominant tensions and what is emerging. - **Timeline signals** — events that matter for design that year. - **Typography** and **Graphic design** — the visual language in detail. - **Flashback design recipes** — named directions, each with `Use for`, `Palette`, `Type`, `Layout`, `Imagery`, `Motion`, and `Risk`. - **Anti-cliches** — guardrails so the work stays specific, not costume. - **Design prompt seeds** — ready-to-run prompts. - **Reference artifacts** — objects, print/graphics, and spaces to anchor on. ## Step 3 — turn the year into the design task Use the year as material, not decoration: 1. **Lead with the thesis and feeling.** Let the year's core idea shape the concept, not just the surface. 2. **Pick the recipe(s) that fit the brief.** Match `Use for` to the product, then carry `palette`, `type`, `layout`, `imagery`, and `motion` into concrete choices (tokens, components, copy tone, motion notes). 3. **Respect the anti-cliches.** Avoid every cliche the corpus calls out — they are how the work avoids looking like a costume of the era. 4. **Honor the Flashback taste principles:** specific beats polished; constraints create style; history explains structure, not just vibes; the present still matters; accessibility (contrast, motion, semantics, readability) is not optional; critique before handoff. 5. **Choose an output mode** that matches the ask: `territories` (2-3 distinct directions with tradeoffs), `brief`, `system` (type/color/layout/spacing/ components/motion/voice), `critique`, `handoff` (specs/tokens/tickets), or `research`. When unsure, offer 2-3 territories and recommend one. ### When the task names more than one year Blend or contrast the years deliberately. Either fuse them into one coherent direction (and say which traits come from which year) or present them as competing territories so the user can choose. ### When the requested year is not in the corpus The corpus is still growing toward full coverage (every year from 1900 to the present). If a requested year is not in the manifest: 1. Use the **nearest available year** and clearly state the substitution (for example: "1979 isn't in the corpus yet; grounding this in 1980, the closest available year."). 2. Then proceed with the design task using that year's research. ## For the user, not for grounding - Human-browsable index of every year: `https://toby.github.io/flashback/` - A single year's example page: `https://toby.github.io/flashback/examples/<YEAR>/index.html` Point people to these when they want to browse; use the manifest and corpus markdown when you are doing the design work.