A new repo every day
We take a trending open-source repository, install it in a clean container, and run it against real files. Then we report exactly what came back — including the parts that broke. Every number on this page and in every video comes from the run itself.
Three hundred people starred this skill catalog in four days. It promises to catalog every skill folder on your machine, and flag the risky ones. We pointed it at five real skill folders, and it flagged two of them critical. It quarantined the flagged skill exactly where its own README says it would. Then restoring it with the id it just gave us said the skill wasn't there.
scan |
CLEAN | 4ms | GET /api/skills — the real drawer scan |
quarantine |
CLEAN | 4ms | POST /api/skills/quarantine — moves the flagged skill out of its drawer |
restore-stale |
DIVERGED | 4ms | restore called with the id the quarantine response itself just returned |
restore-fresh |
CLEAN | 3ms | the same skill, restored after a fresh scan reports its post-quarantine id |
delete |
CLEAN | 3ms | POST /api/skills/delete — a fresh throwaway skill, deleted for real, each pass |
origin-block |
BLOCKED | 2ms | a mutating call with no Origin header — no browser sent this request |
$ npm install skill-cabinet
node:22-bookworm-slim · node v22.23.2 · aarch64
installed in 2.57s · each task run 5×, timings are medians
run 2026-09-04
Eight hundred people starred this binary verifier in four days. It promises every claim about a binary gets checked against the real bytes. We gave it five true claims about its own binary, and five false ones. Every true claim verified, and all five false ones were refuted. Then we gave it an ordinary text file, and it crashed outright.
verify-true |
CLEAN | 45.7ms | true claims about a real binary's own header |
verify-false |
CAUGHT | 45.04ms | the same claims, bitwise-complemented — every one should be wrong |
auto-real |
CLEAN | 47.66ms | auto-triage of a real ELF binary already in the image |
auto-crash |
CRASHED | 43.44ms | the same command, on an ordinary text file already in the image |
parse-pure |
CLEAN | 45.02ms | structural parse, pure-Python core (the default `pip install reverify`) |
parse-full |
CLEAN | 345.94ms | structural parse after `pip install "reverify[full]"` — the documented upgrade, same binary |
$ pip install reverify
python:3.11-slim · python 3.11.16 · aarch64
installed in 1.69s · each task run 5×, timings are medians
run 2026-09-04
One thousand people starred this Tailwind class merger in four days. It promises the same answer as the library it replaces. We gave both libraries two hundred thousand class strings. Every single one came back identical. Zero disagreements. It claims thirty times faster. We measured forty-three.
pairs |
CLEAN | 1391ms | same-prefix conflicts |
crossed |
CLEAN | 908ms | different-prefix pairs |
decorated |
CLEAN | 310ms | variants, negatives, important |
surface |
CLEAN | 602ms | documented API surface |
$ npm install cn tailwind-merge clsx
node:22-bookworm-slim · node v22.23.2 · aarch64
installed in 3.43s · each task run 5×, timings are medians
run 2026-09-04
Twenty thousand people starred this document converter. Two commands in a clean container, and it was running. A Word document came back perfect in sixteen milliseconds. A spreadsheet kept its numbers but lost its header row. Then a scanned page stopped it dead, needing OCR.
report.docx |
CLEAN | 16ms | converted whole |
books.xlsx |
DAMAGED | 16ms | header row came back empty; the column names were demoted into the body |
scan.pdf |
REFUSED | 19ms | refused by default — OCR is opt-in and sends the file to a hosted service |
$ npm install -g @firecrawl/anydoc
node:22-bookworm-slim · node v22.23.2 · aarch64
installed in 2.74s · each task run 5×, timings are medians
run 2026-09-02
Nothing here is estimated. Timings are medians of consecutive runs inside a clean container; star counts are read from the GitHub API with the date shown. A failure is reported as loudly as a success — that is the whole point.