| name | design-research |
| description | Run the three-window research method on any design question: /last30days (engagement — what's alive now), /last30years (survival — what's load-bearing), /last300years (endurance — what governs taste). Trigger: "research X for this design", "what's the state of the art in Y", "has Z ever worked".
|
| kind | skill |
| license | CC-BY-4.0 |
| runtimes | ["claude-code","codex","hermes"] |
design-research
When to use
Before betting the repo, a project, or a client on a tool, method, or principle.
What it does
- 30 days: rank by observed engagement — releases, stars, pushes, dated
announcements. Freshest verified signal wins; reputation counts for nothing.
- 30 years: rank by survival — still-used + built-upon + hype-survived
(the 3-probe test). Tier 🏛/🌳/🌱; under-10-years can never outrank a survivor.
- 300 years: rank by endurance with a Whig-history guard — keep the
surviving kernel, flag the superseded dogma (Modulor's ruler, not his city).
- Synthesize the Lindy read: bet interfaces on survivors, hot-swap the
young layer, and record the anti-portfolio (who was louder and died).
Example
The repo's own digests are the golden output: research/last30days/DIGEST.md,
research/last30years/DIGEST.md, research/last300years/DIGEST.md.
Verification (eval-with-teeth)
Every entry carries a URL; dated claims carry dates; anything unverifiable is
marked UNVERIFIED; single-source promotional claims are flagged as such.
A digest with unlabeled uncertainty is a failed run — no evidence ⇒ No.
Safety
Research grounds decisions; it doesn't make them. Surface trade-offs, let the
human pick at genuine forks.
Cross-runtime
Needs web search in the harness; degrades to knowledge-cutoff answers with an
explicit staleness warning.
