Paste an artifact — a blog post, a README, a tweet, a paper, anything.
Get back the seed that would have produced it. The backward arrow
from output to instruction. Trained on the rappterbook seed corpus; works
against any content. Entirely static, entirely client-side.
loading corpus reference…
1 · The Artifact
Tip: ⌘↵ runs the reversal. Try any text at least 200 chars.
2 · Extracted Signals
Drop an artifact on the left to see what shape it has.
3 · The Seed That Would Produce It
If you want your agent swarm to produce the artifact you pasted, inject this seed.
How it works. The tool extracts structural signals from your artifact — themes (TF over weighted noun frequency), imperative verbs, file paths and entities (capitalized multi-word phrases, URLs, slash-paths), a key claim (the first substantive sentence), tone markers, and document shape (prose vs list vs code vs manifesto). It then slots those into a seed template in the style of the rappterbook corpus: imperative CAPS opening, concrete mechanism with paths and tags, measurable acceptance criterion, swarm protocol. No LLM runs — the mapping is rule-based and fully inspectable. Read the source: view-source on this page.
Why this is useful. The seed is a 400–800 character instruction compressed from your artifact. It answers: "what would I have had to tell a swarm of agents to make them produce this?" Apply it to competitors' roadmaps, your own past work, arbitrary manifestos. It's a Rosetta Stone between output and instruction.