메뉴
Autonomous research agent: literature, ideas, experiments, writing, review, rebuttal, presentation, theory. Triggers: 'research', 'write paper', 'survey', 'polish', 'translate', 'review', 'rebuttal', 'slides', 'poster', '写论文', '文献调研', '润色', '翻译', '去AI味', '审稿'.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Run GPTZero AI detection on one or more review files via the app.gptzero.me dashboard. Requires Chrome running with --remote-debugging-port=9222 and the user logged into GPTZero. Use when user says "check ai rate", "gptzero", "检测AI率", "run gptzero", or wants to verify that review prose passes AI detection.
Write a venue-form peer review of someone else's paper. Humanization is automatic. Thin shim over the `research-review` CLI. The CLI runs three steps internally for every paper: (1) codex writes a long detailed free-form draft, no template constraint; (2) extract_judgment compresses the draft into a structured judgment dict (numerics + per-field bullets); (3) prose is regenerated under real-human sentence templates, with numerics preserved from step 1. No draft input, no draft option. Every paper goes through the same three-step pipeline. For self-critique with no AI-detection concern, use /self-review.
Critically review your own paper to find real weaknesses before submission. Output is meant to be read by you (the author) and fed into a revision loop, not submitted to a venue. So: no AI-rate constraint, no corpus-template humanization, no GPTZero verification. The skill optimizes for harsh, specific, paper-grounded critique instead of polite peer-review-ese. Sibling skill: - peer-review: write a venue-form review with prose humanized via corpus templates. Use when you are reviewing someone else's paper and need GPTZero <=cap. Also handles "humanize an existing review draft" via the --draft option.
Write publication-ready ML/AI/Systems papers for NeurIPS, ICML, ICLR, ACL, AAAI, COLM, OSDI, NSDI, ASPLOS, SOSP. Use when drafting papers from research repos, structuring arguments, verifying citations, or preparing camera-ready submissions. Includes LaTeX templates, reviewer guidelines, and citation verification workflows.
| name | agentic-research |
| description | Autonomous research agent: literature, ideas, experiments, writing, review, rebuttal, presentation, theory. Triggers: 'research', 'write paper', 'survey', 'polish', 'translate', 'review', 'rebuttal', 'slides', 'poster', '写论文', '文献调研', '润色', '翻译', '去AI味', '审稿'. |
Autonomous research agent built with OpenProgram (Agentic Programming paradigm).
Two-level autonomous loop: Level 1 picks a research stage (literature, writing, review, ...), Level 2 dispatches to functions within that stage. Cross-model review uses GPT (via Codex) as reviewer and Claude as author.
/agentic-research "<your task>"
Single entry point — the agent reads your task and autonomously decides which functions to call.
/agentic-research "Survey recent work on LLM uncertainty"
/agentic-research "Polish this paragraph for NeurIPS: <text>"
/agentic-research "翻译这段中文为英文LaTeX: <中文草稿>"
/agentic-research "Review paper/ as a NeurIPS reviewer"
/agentic-research "Run the full pipeline for topic 'LLM Uncertainty', venue NeurIPS"
# Basic usage (--work-dir is required to run a task)
research-harness --work-dir /abs/path "your task"
research-harness --list # list all functions
# Cross-model review: Claude writes, GPT reviews (ARIS design)
research-harness --work-dir /abs/path "Review the paper at ./project/" \
--provider claude-code \
--review-provider openai-codex
# Custom models
research-harness --work-dir /abs/path "Survey LLM uncertainty" \
--provider openai --model gpt-4o \
--review-provider openai-codex
# Operation log is written automatically to <work-dir>/OPERATION_LOG.md
from research_harness.main import research_agent
from openprogram.providers import create_runtime
# Single model
rt = create_runtime(provider="claude-code")
rt.set_workdir("/abs/path/to/work-dir")
result = research_agent(task="Survey LLM uncertainty", runtime=rt)
# Cross-model review
exec_rt = create_runtime(provider="claude-code")
exec_rt.set_workdir("/abs/path/to/work-dir")
review_rt = create_runtime(provider="openai-codex")
review_rt.set_workdir("/abs/path/to/work-dir")
result = research_agent(
task="Review the paper at ./project/ as EMNLP reviewer",
runtime=exec_rt,
review_runtime=review_rt,
)