research-teamlisted

General-purpose research team (an orchestrator over a group of agents). For any subject (technology, market, academic, product, current affairs), it runs five stages — planning (key-question decomposition, competing hypotheses, disconfirmation plan), parallel collection (collection floors, source grades, search log), independent verification (citation existence, verbatim-quote matching, disconfirmation search, independent-source corroboration), synthesis (evidence ledger with every claim linked to an ID, calibrated confidence vocabulary), and evaluation (deterministic check scripts plus an auditor agent gate) — and produces a corroborated research report. Use it for requests such as "research this", "look into it", "dig deeper", "verify this", "compare these and help me choose", "summarise it with evidence", "market research", "help me pick a technology", in any language. Do not use it for requests that a codebase search alone answers, for reviewing implementation plans and code, or for a pure literature revi
tomkd555/research-skill · ★ 0 · AI & Automation · score 73

Install: claude install-skill tomkd555/research-skill

# General-purpose research team (research-team) This skill orchestrates research on any subject, producing findings in a structure an LLM can interpret and evaluate correctly. It distributes collection, verification, and auditing across subagents, and owns four rule files that are the source of record for the rules themselves. **Output language.** The section headings, the table headers, and the confidence labels are English and only English — in every deliverable, and in every message to the user that carries one, because such a message transcribes the report rather than restating it. A label is written with its band as `roughly even chance (45-65%)`, never translated and never reworded. The prose around them is written in the language of the request — the body of `research_brief.md` and `report.md`, the evidence `claim` text, and every message to the user. `verbatim_quote` stays in the language of the source. Settle the language at Step 0, record it in the ledger's top-level `deliverable_language`, and pass it to `merge_fragments.py` as `--deliverable-language`; nothing else takes a language flag, because `scripts/labels.py` holds one English set of names and the check scripts all read it. A collection agent picks the language of each query for itself while collecting; that judgement is neither planned nor recorded. ## How to start | Situation | What to do | |---|---| | A research request arrives in the main session | Launch the `research-team-lead` agent with the Agent