AI4S Agent
Meta-skill řetězení explore, survey, experiment a paper-writer pro AI4S výzkum
Funguje s nastavením
Co umí
Pure-markdown meta-skill, který orchestrates four-stage AI-for-Science research pipeline (research-explorer, literature-survey, experiment-suite, paper-writer) end to end, sharing artifacts by deterministic slug convention. Spouští se, když uživatel chce whole research package na topic; pro single stage it defers na individual skill. Vyžaduje full skill stack, LaTeX a web access k actually run.
Testovací report
Meta-skill explicitně does no work of its own, it only orders four downstream skills, takže single-skill install leaves it broken; honest install runs install.sh k bring in research-explorer, literature-survey, experiment-suite a paper-writer too. Its deterministic slug contract reproduced identical slugs across topics as specified, a downstream literature-survey skill commendably forbids fabricated citations. Scored setup, not pass, protože honest end-to-end run (60-200 real web-fetched citations plus experiment execution a two LaTeX PDFs) je multi-hour task not verifiable v context, takže no baseline delta could be measured.
Testováno: 2026-07-17 · Claude Code 2.x (agent harness)
Instalace
git clone https://github.com/ai4s-research/ai4s-skills mkdir -p ~/.claude/skills cd ai4s-skills && SKILLS_DIR=~/.claude/skills ./install.sh # installs the WHOLE stack; the meta-skill is non-functional without its 4 downstream skills
Příkazy a ukázkové prompty
/ai4s-agentMeta-skill řetězení explore, survey, experiment a paper-writer pro AI4S výzkum
Skilly se spouštějí běžnými požadavky — žádné příkazy k zapamatování. Po instalaci ho aktivují prompty jako tyto (anglicky):
Run the full explore-survey-experiment-paper pipeline on this topicProduce a literature survey and experiment suite for our new studyGenerate a full paper draft starting from just our research topic