Plugin Research Powerpack — Development skill for Claude Code
claude code plugin: evidence-grounded web research on the research powerpack mcp — cited answers, deep reports and comparison corpora on disk, fact-checks, and hooks that tie every claim to a verbatim.
How to install Plugin Research Powerpack
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open yigitkonur/plugin-research-powerpack and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Plugin Research Powerpack does
claude code plugin: evidence-grounded web research on the research powerpack mcp — cited answers, deep reports and comparison corpora on disk, fact-checks, and hooks that tie every claim to a verbatim quote
Alternatives in Development
- Notebooklm Skill — Use this skill to enable Claude Code to communicate directly with your Google NotebookLM notebooks 5.9k ★
- GDScript VS C# — Skill Instruction Comparison — GDScript vs C# — Skill Instruction Comparison skill 2.8k ★
- Eval Compare — /eval-compare — Version Comparison 216 ★
README
plugin-research-powerpack
a claude code plugin that makes your agent do web research like it has to defend every sentence. answers come with verbatim quotes, every number points at the exact line that says it, and the plugin itself refuses to let a made-up citation through.
it bundles the **research powerpack** mcp server (search, plan, and quote-verified extraction) and wraps it in skills, agents and hooks that do the boring discipline for you.
why it exists
agents are bad at web research in a very specific way. they skim a search snippet, paraphrase it with confidence, and drop a link that doesn't actually say the thing. or they paste five pages of raw html into context and torch 40k tokens to answer one question.
this plugin fixes both:
- only quotes count. search results are treated as leads, never evidence. a claim needs a quote the server pulled from the page itself, with a line number.
- every quote gets an id.
E-a1b2c3, stable across agents and sessions, kept in a ledger on disk instead of in your context window. - the hooks check the work. a cited number has to actually appear in the cited quote. a quote that's just a "Latest" button or an avatar link gets flagged as thin. unknown ids, uncited numbers, blank table cells,
TBDs and dead links block a research run from finishing. - tool output gets compacted. the model still sees every quote, but across 130 recorded extraction calls it read 62% fewer characters than the raw output, and 74% fewer on searches (top 12 leads, the rest stay in the ledger).
install
claude plugin marketplace add yigitkonur/plugin-research-powerpack
claude plugin install research@plugin-research-powerpack
or from inside a claude code session:
/plugin marketplace add yigitkonur/plugin-research-powerpack
/plugin install research@plugin-research-powerpack
or in one line:
claude plugin install research --marketplace yigitkonur/plugin-research-powerpack
the ins
Related Skills
Refcheck
Reference integrity for the scientific literature: checks whether a paper's citations exist, match what was ci
Build Dossier
Deep source-cited dossier on a subject (person, business, family/principals, panel, entity warranting multi-se
Truth Shield
Stop Claude from lying to you. A Claude Code skill that fact-checks every claim against real sources — code fi
Ars Dr Fact Check
ARS deep-research fact-check mode — claim-by-claim verification report
Callsheet
Turn any recording of a call into one self-contained HTML readout: local Whisper, multi-agent analysis, hand-a
Fact Drift
Check cited facts against upstream sources and local citation files.
Related Agents
Manual Verifier
Independent fact-checker for answers built from the Ford service manual. Give it the vehicle, the question and
Reinforcement Learning Secretary
Grounded secretary for Reinforcement Learning Brain. Use for source-cited answers about reinforcement learning
Gauntlet Loop Secretary
Grounded secretary for Gauntlet Loop Brain. Use for source-cited questions about designing, running, evaluatin