ULW-RESEARCH — Saturation Research Orchestrator

What it does

A heavyweight research orchestration skill that only activates on an explicit request (ulw-research, ultradebate, hyperdebate, "deep research", "ultra-precise investigation").

• Phase 0 scoping: decomposes the question into 3+ orthogonal axes, writes brief.md, and creates a timestamped session directory under .omo/ulw-research/.
• Format gate: before any team exists, it proposes the final deliverable (PDF + DOCX by default; slides, HTML, Markdown, LaTeX as alternatives) plus a template, then waits for your answer.
• Max-size team: one member per axis, mandatory skeptic/red-team members, plus read-only explore / librarian / browsing lanes with explicit worker floors per query scope.
• EXPAND loop: every lead a member raises is routed immediately and chased until leads run dry (minimum two waves), with bounded "excursions" that must be entered on a stated trigger and exited with a fold-back line.
• Debate + execution proof: contested claims go through attack/defense rounds; performance-, compatibility- or undocumented-behavior claims are settled by running actual scripts and recording CONFIRMED / REFUTED / PARTIAL.
• Audit trail: claim-graph.md, observation-manifest.md, intent-diff.md, debate-log.md, excursion-log.md and a sources ledger are written in real time so the run survives context loss.

Who it's for

• Anyone doing research where a single wrong claim is expensive: legal, medical, financial, procurement, or public-facing work.
• Researchers, analysts and PMs who need a citation-per-claim report with an explicit unresolved/refuted annex.
• Power users on a multi-agent runtime (omo-senpi style) with a generous token budget.

Examples

1. /ulw-research compare three vector databases on latency, licensing and ops cost → axis owners sweep docs, GitHub and benchmarks; performance numbers are re-verified by executed scripts; output ships as PDF + DOCX.
2. ultradebate: do we actually need this architecture migration? → a skeptic attacks the evidence while a contrarian attacks the framing, and every verdict lands in debate-log.md with the residual risk.
3. deep research: how does the amended data-protection rule affect our product? → primary sources only, dated claims locked through the claim-graph gate, anything unverifiable pushed to the annex.

Install the skill I found on Claude Skill Mart.
Copy the packages/omo-senpi/skills/ulw-research folder from the GitHub repo code-yeongyu/oh-my-openagent into my ~/.claude/skills/ulw-research/.
When it's done, tell me in one line what this skill can do.