원클릭으로 Manus에서 모든 스킬 실행

paper-search

스타25

포크3

업데이트2026년 6월 16일 08:16

Find a small list of academic papers on demand — "给我 N 篇目标检测领域顶刊 / 顶会 / SOTA 论文 / 综述" — using the multi-source chain arxiv API → Semantic Scholar → web_search. Distinct from `daily-paper-digest` (scheduled daily delivery) and `arxiv` (single-day category dump): this one is the on-demand "find me N papers about X" surface that the user usually triggers in chat.

설치

Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.

Manus에서 실행

출처

Kkkirito-123

Kkkirito-123/ZLAgent

GitHub 저장소 열기 Creator 저장소 보기

다운로드

Manus에서 실행

paper-search

Trigger

Use this skill when the user asks for a small ad-hoc list of academic papers, typically:

"给我目标检测领域的 5 篇顶刊论文，要最新的"
"AI agent 方向最近半年的 SOTA 论文有哪些"
"推荐 3 篇关于 RAG 评估的综述"
"Find me 5 recent papers on diffusion video generation"

Do NOT use this skill for:

Scheduled daily / weekly digests → daily-paper-digest.
"Today's cs.AI listing on arxiv" (category dump, not topic-targeted) → arxiv.
Reading / summarising a single paper the user already linked → web-summary + read_url.

Inputs

Resolve from the user message before fetching:

topic — the research area in plain words ("目标检测 / object detection", "RAG evaluation"). Required; ask one short follow-up if truly ambiguous.
count — how many papers, default 5, hard cap 8.
recency — "最新 / latest" → last 12 months; "近年 / recent" → last 3 years; explicit window like "2024 以来" → respect verbatim.
kind — empirical / SOTA / survey / benchmark / methods. Default: whichever the user phrased; if silent, prefer SOTA + 1 survey when both are obviously useful.

Steps

Source priority (cheapest + highest signal first):
- arxiv API — https://export.arxiv.org/api/query?search_query=ti:<topic>+OR+abs:<topic>&start=0&max_results=20&sortBy=submittedDate&sortOrder=descending via read_url. Returns Atom XML; parse for title / authors / summary / arxiv id. Best for ML / CS / EE.
- Semantic Scholar Graph API — https://api.semanticscholar.org/graph/v1/paper/search?query=<topic>&limit=20&fields=title,authors,year,venue,citationCount,abstract,url via read_url. Better venue + citation metadata than arxiv. If it returns 429, wait and retry once with a 60s gap; then move on without it.
- web_search — only as a last fallback when arxiv + S2 both yielded fewer than count candidates. Query template: "<topic>" survey 2024 site:arxiv.org then site:openreview.net.
Filter + rank:
- Drop preprints whose abstract is empty / "Withdrawn".
- Prefer top-tier venues when the user asked for 顶刊 / 顶会 (CV: CVPR / ICCV / ECCV / TPAMI; ML: NeurIPS / ICML / ICLR; NLP: ACL / EMNLP / NAACL; speech: INTERSPEECH; systems: SOSP / OSDI; security: S&P / USENIX). Surface arxiv-only entries only when no venued match exists.
- Sort: recency desc → citation desc → venue tier.
- Dedupe by arxiv id / DOI / canonical URL.
Render (one bullet per paper, newest first):
- Title (in original language).
- Primary authors (max 3 then et al.).
- Year + venue if known.
- One-line plain-language takeaway (≤ 80 chars).
- Source URL (arxiv abs preferred over PDF).
Tool budget: aim for ≤ 3 tool calls total. List-task adaptive budget already grants up to 10 — reserve the headroom for retries, not for browsing every candidate.

Verification

Every paper has a real source URL.
Title / authors / year are never invented; if unknown, omit the field and say so plainly.
If the result set is thin (< count solid hits), say "找到 N 篇够稳的，剩下的来源都不太可信，先给你这 N 条" — do not pad.
Do NOT ask the user to "确认是否保存到知识库" unless they signalled long-term storage intent. Default is in-chat output only.

Pitfalls

Semantic Scholar 429 is common; honour the Retry-After header and degrade silently to arxiv-only.
DDG / Bing captcha will trigger the v0.45 provider chain in web_search; if all 3 web providers fail, fall back to whatever arxiv + S2 produced even if < count.
"顶刊" sometimes means journal (TPAMI / TMLR) and "顶会" means conference (CVPR / NeurIPS); when the user uses both, include at least one of each.
Don't translate paper titles — keep the originals so the user can search them as-is.

이 저장소의 다른 Skills

같은 저장소

arxiv

Kkkirito-123/ZLAgent

Fetch and summarise recent papers from arxiv.org. Use this skill when the user wants a digest of today's or this week's cs.AI / cs.LG / cs.CL listings, a focused per-topic roundup, or a quick look at a specific submission.

2026-06-1625

code-review-quality

Kkkirito-123/ZLAgent

Lightweight code-review checklist the agent runs before declaring a patch ready. Catches the obvious classes of issue (dead code, missing error handling, weakened tests) without a full security audit. Adapted from addyosmani's "Review" lifecycle phase.

2026-06-1625

systematic-debugging

Kkkirito-123/ZLAgent

Root-cause-first debugging loop for code or runtime errors: understand the failure before proposing a fix. Adapted from Hermes' systematic- debugging — same iron law, smaller surface, ZLAgent-tool-aware.

2026-06-1625

verification-before-completion

Kkkirito-123/ZLAgent

Do not declare a task done without running verification. For code, that means running tests / smoke; for answers, that means checking the claim against a fresh source. Adapted from obra/superpowers and addyosmani's "Verify" lifecycle phase.

2026-06-1625

document-processing

Kkkirito-123/ZLAgent

Umbrella skill for user document workflows — PDF, Word, Excel, PowerPoint, text extraction, summarization. Dispatches to the right MCP / tool; proposes an install via mcp-discovery when no local capability fits. Adapted from Anthropic's docx/pdf/xlsx/pptx skills.

2026-06-1625

mcp-discovery

Kkkirito-123/ZLAgent

When the user wants something the agent doesn't have a tool for yet (travel planning, GitHub issues, Linear, weather, filesystem, ...), search the live MCP ecosystem for an existing server, propose installing it, and on confirm wire it up via mcp_manage. Adapted from Hermes' mcporter skill — same flow, ZLAgent-tool-aware.

2026-06-1625

name	paper-search
description	Find a small list of academic papers on demand — "给我 N 篇目标检测领域顶刊 / 顶会 / SOTA 论文 / 综述" — using the multi-source chain arxiv API → Semantic Scholar → web_search. Distinct from `daily-paper-digest` (scheduled daily delivery) and `arxiv` (single-day category dump): this one is the on-demand "find me N papers about X" surface that the user usually triggers in chat.
version	0.1.0
tags	["research","papers","on-demand"]
metadata	{"zlagent":{"category":"research","triggers":["顶刊","顶会","几篇论文","几个论文","篇论文","找.论文","推荐.论文","SOTA","state-of-the-art","最新进展","综述","survey","对比.*论文","benchmark"],"capabilities":["web_query","read_url"],"related_skills":["arxiv","daily-paper-digest","web-summary"]}}

paper-search

Trigger

Use this skill when the user asks for a small ad-hoc list of academic papers, typically:

"给我目标检测领域的 5 篇顶刊论文，要最新的"
"AI agent 方向最近半年的 SOTA 论文有哪些"
"推荐 3 篇关于 RAG 评估的综述"
"Find me 5 recent papers on diffusion video generation"

Do NOT use this skill for:

Scheduled daily / weekly digests → daily-paper-digest.
"Today's cs.AI listing on arxiv" (category dump, not topic-targeted) → arxiv.
Reading / summarising a single paper the user already linked → web-summary + read_url.

Inputs

Resolve from the user message before fetching:

topic — the research area in plain words ("目标检测 / object detection", "RAG evaluation"). Required; ask one short follow-up if truly ambiguous.
count — how many papers, default 5, hard cap 8.
recency — "最新 / latest" → last 12 months; "近年 / recent" → last 3 years; explicit window like "2024 以来" → respect verbatim.
kind — empirical / SOTA / survey / benchmark / methods. Default: whichever the user phrased; if silent, prefer SOTA + 1 survey when both are obviously useful.

Steps

Source priority (cheapest + highest signal first):
- arxiv API — https://export.arxiv.org/api/query?search_query=ti:<topic>+OR+abs:<topic>&start=0&max_results=20&sortBy=submittedDate&sortOrder=descending via read_url. Returns Atom XML; parse for title / authors / summary / arxiv id. Best for ML / CS / EE.
- Semantic Scholar Graph API — https://api.semanticscholar.org/graph/v1/paper/search?query=<topic>&limit=20&fields=title,authors,year,venue,citationCount,abstract,url via read_url. Better venue + citation metadata than arxiv. If it returns 429, wait and retry once with a 60s gap; then move on without it.
- web_search — only as a last fallback when arxiv + S2 both yielded fewer than count candidates. Query template: "<topic>" survey 2024 site:arxiv.org then site:openreview.net.
Filter + rank:
- Drop preprints whose abstract is empty / "Withdrawn".
- Prefer top-tier venues when the user asked for 顶刊 / 顶会 (CV: CVPR / ICCV / ECCV / TPAMI; ML: NeurIPS / ICML / ICLR; NLP: ACL / EMNLP / NAACL; speech: INTERSPEECH; systems: SOSP / OSDI; security: S&P / USENIX). Surface arxiv-only entries only when no venued match exists.
- Sort: recency desc → citation desc → venue tier.
- Dedupe by arxiv id / DOI / canonical URL.
Render (one bullet per paper, newest first):
- Title (in original language).
- Primary authors (max 3 then et al.).
- Year + venue if known.
- One-line plain-language takeaway (≤ 80 chars).
- Source URL (arxiv abs preferred over PDF).
Tool budget: aim for ≤ 3 tool calls total. List-task adaptive budget already grants up to 10 — reserve the headroom for retries, not for browsing every candidate.

Verification

Every paper has a real source URL.
Title / authors / year are never invented; if unknown, omit the field and say so plainly.
If the result set is thin (< count solid hits), say "找到 N 篇够稳的，剩下的来源都不太可信，先给你这 N 条" — do not pad.
Do NOT ask the user to "确认是否保存到知识库" unless they signalled long-term storage intent. Default is in-chat output only.

Pitfalls

Semantic Scholar 429 is common; honour the Retry-After header and degrade silently to arxiv-only.
DDG / Bing captcha will trigger the v0.45 provider chain in web_search; if all 3 web providers fail, fall back to whatever arxiv + S2 produced even if < count.
"顶刊" sometimes means journal (TPAMI / TMLR) and "顶会" means conference (CVPR / NeurIPS); when the user uses both, include at least one of each.
Don't translate paper titles — keep the originals so the user can search them as-is.