paper-workshoplisted

CRUCIBLE — a Claude-only adversarial expert workshop that stress-tests a scientific paper and then, opt-in, rebuilds it. ACT I (TRIBUNAL): from the paper PDF (plus a few public cited works it fetches), a topic-adapted fleet of expert referee subagents — every contested claim argued from at least two COMPETING intellectual traditions, plus generalist seats for relevance and understandability — reviews the paper to the sentence, every comment grounded in an exact quote (no fabrication, no confidence scores) and independently re-checked from many angles before delivery. ACT II (ATELIER, opt-in): with the author's source + data + code, it implements the agreed changes as a tracked-changes redline plus a clean revised version, RE-RUNS the analysis under an execution-provenance wall (no number enters the paper unless a logged re-run produced it), and assembles a replication package — every change multi-angle-verified, nothing applied without the author's sign-off on anything that touches the record. Triggers: - "wo
tjhavranek/paper-workshop · ★ 0 · Data & Documents · score 62

Install: claude install-skill tjhavranek/paper-workshop

# paper-workshop (CRUCIBLE) > *Every paper leaves changed.* A two-act simulation of the workshop a paper would get if a specialist were assigned to **every** contested choice, each one argued by **both** rival schools, a generalist panel asked whether it **matters** and is **intelligible**, every comment were **independently re-checked from many angles**, and then the best of the room **rebuilt the paper and its replication package**. The deeper modes can run for a while by design (you choose the depth). The whole tool is Claude-only; it uses Claude 4.8's **Workflow** engine when available and plain subagents otherwise, with live work (fetching cited sources, running the author's code) via the **Agent/Bash** tools. **Honest limits.** See [`LIMITATIONS.md`](LIMITATIONS.md): effectiveness is not measured yet (no recall or false-positive numbers), same-model decorrelation is a design bet, and Act II's rebuild is built and unit-tested but not yet field-proven end to end. ## The two acts and the gate between them - **Act I — TRIBUNAL.** Input: **the paper PDF** — no data or code, and it touches none of your files (it does fetch a few *public* cited works to check the paper's claims). Output: a verified, prioritized finding ledger + a referee-report bundle + a debate transcript + an importance memo + a completeness certificate. Read-only; nothing of the author's is touched. (`workflow/phase1_tribunal.js`) - **[HARD GATE].** The skill **stops** and offers to implement.