agentsope/SkillAlchemy

README.md

• Network accessline 11

  <p align="center">   <a href="https://arxiv.org/abs/2608.23417"><img src="https://img.shields.io/badge/arXiv-2608.23417-b31b1b.svg" alt="arXiv"></a>

• Network accessline 12

  <a href="https://arxiv.org/abs/2608.23417"><img src="https://img.shields.io/badge/arXiv-2608.23417-b31b1b.svg" alt="arXiv"></a>

• Network accessline 13

  <a href="https://github.com/agentsope/SkillAlchemy/stargazers"><img src="https://img.shields.io/github/stars/agentsope/SkillAlchemy?logo=github&color=ffca28" alt="Stars"></a>

• Network accessline 14

  <img src="https://img.shields.io/badge/agents-Claude%20Code%20%7C%20Codex-6E56CF" alt="Supported agents">

• Network accessline 27

  * **2026-08-24** — Our paper, **[SkillAlchemy: Open-World Agent Skill Creation](https://arxiv.org/abs/2608.23417)**, is now available on arXiv.

• Network accessline 139

  Install SkillAlchemy from https://github.com/agentsope/SkillAlchemy and show me how to use it.

• Network accessline 175

  > Hengjun Wang, Shuyue Wei, Boyi Liu, Jun Yang, Yongxin Tong.  > arXiv:2608.23417, 2026 > **[arXiv](https://arxiv.org/abs/2608.23417)** · **[PDF](https://arxiv.org/pdf/2608.23417)**

• Network accessline 177

  > **[arXiv](https://arxiv.org/abs/2608.23417)** · **[PDF](https://arxiv.org/pdf/2608.23417)**

README_CN.md

• Network accessline 11

  <p align="center">   <a href="https://arxiv.org/abs/2608.23417"><img src="https://img.shields.io/badge/arXiv-2608.23417-b31b1b.svg" alt="arXiv"></a>

• Network accessline 12

  <a href="https://arxiv.org/abs/2608.23417"><img src="https://img.shields.io/badge/arXiv-2608.23417-b31b1b.svg" alt="arXiv"></a>

• Network accessline 13

  <a href="https://github.com/agentsope/SkillAlchemy/stargazers"><img src="https://img.shields.io/github/stars/agentsope/SkillAlchemy?logo=github&color=ffca28" alt="Stars"></a>

• Network accessline 14

  <img src="https://img.shields.io/badge/agents-Claude%20Code%20%7C%20Codex-6E56CF" alt="Supported agents">

• Network accessline 25

  * **2026-08-24** — 我们的论文 **[SkillAlchemy: Open-World Agent Skill Creation](https://arxiv.org/abs/2608.23417)** 已上线 arXiv。

• Network accessline 54

  请从 https://github.com/agentsope/SkillAlchemy 安装 SkillAlchemy，并告诉我如何使用。

• Network accessline 172

  > Hengjun Wang, Shuyue Wei, Boyi Liu, Jun Yang, Yongxin Tong > arXiv:2608.23417, 2026 > **[arXiv](https://arxiv.org/abs/2608.23417)** · **[PDF](https://arxiv.org/pdf/2608.23417)**

• Network accessline 174

  > **[arXiv](https://arxiv.org/abs/2608.23417)** · **[PDF](https://arxiv.org/pdf/2608.23417)**

README_JA.md

• Network accessline 11

  <p align="center">   <a href="https://arxiv.org/abs/2608.23417"><img src="https://img.shields.io/badge/arXiv-2608.23417-b31b1b.svg" alt="arXiv"></a>

• Network accessline 12

  <a href="https://arxiv.org/abs/2608.23417"><img src="https://img.shields.io/badge/arXiv-2608.23417-b31b1b.svg" alt="arXiv"></a>

• Network accessline 13

  <a href="https://github.com/agentsope/SkillAlchemy/stargazers"><img src="https://img.shields.io/github/stars/agentsope/SkillAlchemy?logo=github&color=ffca28" alt="Stars"></a>

• Network accessline 14

  <img src="https://img.shields.io/badge/agents-Claude%20Code%20%7C%20Codex-6E56CF" alt="Supported agents">

• Network accessline 27

  * **2026-08-24** — 論文 **[SkillAlchemy: Open-World Agent Skill Creation](https://arxiv.org/abs/2608.23417)** を arXiv で公開しました。

• Network accessline 139

  https://github.com/agentsope/SkillAlchemy から SkillAlchemy をインストールし、使い方を教えてください。

• Network accessline 175

  > Hengjun Wang, Shuyue Wei, Boyi Liu, Jun Yang, Yongxin Tong. > arXiv:2608.23417, 2026 > **[arXiv](https://arxiv.org/abs/2608.23417)** · **[PDF](https://arxiv.org/pdf/2608.23417)**

• Network accessline 177

  > **[arXiv](https://arxiv.org/abs/2608.23417)** · **[PDF](https://arxiv.org/pdf/2608.23417)**

SKILL.md

• Network accessline 33

  > ``` > > Alternatively, search for Lens and LEAP on https://skills.sh and install them there.

• Network accessline 34

  > > Alternatively, search for Lens and LEAP on https://skills.sh and install them there. >

• Network accessline 35

  > Alternatively, search for Lens and LEAP on https://skills.sh and install them there.

package.json

• Executable scriptline 1

  Executable file

skills/LEAP/scripts/build_component_index.py

• Executable scriptline 1

  #!/usr/bin/env python3

skills/LEAP/scripts/build_corpus.py

• Executable scriptline 1

  #!/usr/bin/env python3

• Network accessline 31

  MANIFEST_PATH = ROOT / "corpus" / "manifest.json" SITEMAP_URLS = [     "https://skills.sh/sitemap-skills-1.xml",

• Network accessline 32

  SITEMAP_URLS = [     "https://skills.sh/sitemap-skills-1.xml",     "https://skills.sh/sitemap-skills-2.xml",

• Network accessline 33

  "https://skills.sh/sitemap-skills-1.xml",

• Network accessline 34

  "https://skills.sh/sitemap-skills-2.xml",

• Network accessline 72

  RAW_URL_PATTERNS = [     "https://raw.githubusercontent.com/{owner}/{repo}/main/skills/{name}/SKILL.md",

• Network accessline 73

  "https://raw.githubusercontent.com/{owner}/{repo}/main/skills/{name}/SKILL.md",

• Network accessline 74

  "https://raw.githubusercontent.com/{owner}/{repo}/main/{name}/SKILL.md",

• Network accessline 75

  "https://raw.githubusercontent.com/{owner}/{repo}/master/skills/{name}/SKILL.md",

• Network accessline 76

  "https://raw.githubusercontent.com/{owner}/{repo}/master/{name}/SKILL.md",

• Network accessline 82

  skills = []     try:         req = urllib.request.Request(url, headers={"User-Agent": "LEAP/0.4.0"})

• Network accessline 83

  try:         req = urllib.request.Request(url, headers={"User-Agent": "LEAP/0.4.0"})         with urllib.request.urlopen(req, timeout=REQUEST_TIMEOUT) as resp:

• Network accessline 84

  req = urllib.request.Request(url, headers={"User-Agent": "LEAP/0.4.0"})

• Network accessline 85

  with urllib.request.urlopen(req, timeout=REQUEST_TIMEOUT) as resp:

• Network accessline 86

  tree = ET.parse(resp)         for url_elem in tree.iter("{http://www.sitemaps.org/schemas/sitemap/0.9}url"):             loc = url_elem.find("{http://www.sitemaps.org/schemas/sitemap/0.9}loc")

• Network accessline 87

  for url_elem in tree.iter("{http://www.sitemaps.org/schemas/sitemap/0.9}url"):

• Network accessline 88

  loc = url_elem.find("{http://www.sitemaps.org/schemas/sitemap/0.9}loc")

• Network accessline 89

  lastmod = url_elem.find("{http://www.sitemaps.org/schemas/sitemap/0.9}lastmod")

• Network accessline 111

  url = pattern.format(owner=owner, repo=repo, name=name)         try:             req = urllib.request.Request(url, headers={"User-Agent": "LEAP/0.4.0"})

• Network accessline 112

  try:             req = urllib.request.Request(url, headers={"User-Agent": "LEAP/0.4.0"})             with urllib.request.urlopen(req, timeout=REQUEST_TIMEOUT) as resp:

• Network accessline 113

  req = urllib.request.Request(url, headers={"User-Agent": "LEAP/0.4.0"})

• Network accessline 114

  with urllib.request.urlopen(req, timeout=REQUEST_TIMEOUT) as resp:

skills/LEAP/scripts/download_subtitles.sh

• Executable scriptline 1

  #!/bin/bash

• Network accessline 5

  # Language priority: Chinese > English > other languages. # # Dependency: yt-dlp (https://github.com/yt-dlp/yt-dlp)

• Network accessline 6

  # # Dependency: yt-dlp (https://github.com/yt-dlp/yt-dlp) #   brew install yt-dlp    # macOS

• Network accessline 7

  # Dependency: yt-dlp (https://github.com/yt-dlp/yt-dlp)

skills/LEAP/scripts/score_skill.py

• Executable scriptline 1

  #!/usr/bin/env python3

skills/LEAP/scripts/srt_to_transcript.py

• Executable scriptline 1

  #!/usr/bin/env python3

skills/agentsop-aider/README.md

• Network accessline 3

  A distilled SOP skill for using **Aider** (https://github.com/Aider-AI/aider) as a terminal-based AI pair programmer. Part of a 7-project landscape research (LangGraph, LlamaIndex, DSPy, CrewAI, vLLM, Aider, Dify) extracting the decision OS

skills/agentsop-bio-fraud-forensics/references/R01-misconduct-taxonomy.md

• Network accessline 74

  | source_id | URL | type | confidence | |---|---|---|---| | S1 ORI definition | https://ori.hhs.gov/definition-research-misconduct | regulator (authoritative) | high |

• Network accessline 75

  |---|---|---|---| | S1 ORI definition | https://ori.hhs.gov/definition-research-misconduct | regulator (authoritative) | high | | S2 42 CFR 93 | https://www.ecfr.gov/current/title-42/chapter-I/subchapter-H/part-93 | federal regulation | hig

• Network accessline 76

  | S1 ORI definition | https://ori.hhs.gov/definition-research-misconduct | regulator (authoritative) | high |

• Network accessline 77

  | S2 42 CFR 93 | https://www.ecfr.gov/current/title-42/chapter-I/subchapter-H/part-93 | federal regulation | high |

• Network accessline 78

  | S3 2024 final rule | https://www.federalregister.gov/documents/2024/09/17/2024-20814/public-health-service-policies-on-research-misconduct | federal rulemaking | high |

• Network accessline 79

  | S4 ORI Regs Q&A (burden/honest error) | https://ori.hhs.gov/policies-regulations-qa | regulator | high |

• Network accessline 80

  | S5 NSF OIG "Assessing Intent" | https://oig.nsf.gov/sites/default/files/document/2021-10/Assessing%20Intent%20in%20RM%20Investigations_4.pdf | federal investigator guidance | high |

• Network accessline 81

  | S6 "recklessness" defined | https://www.tandfonline.com/doi/full/10.1080/08989621.2023.2256650 | peer-reviewed | med |

• Network accessline 82

  | S7 Andrade ORI finding | https://retractionwatch.com/2026/02/06/office-research-integrity-2026-ori-finding-researcher-faked-data-grant-applications/ | reported ORI case | high |

• Network accessline 83

  | S8 NIH NOT-OD misconduct notices | https://grants.nih.gov/grants/guide/notice-files/NOT-OD-26-041.html | federal notice | high |

• Network accessline 84

  | S9 Bik 2016 mBio | https://journals.asm.org/doi/10.1128/mbio.00809-16 (PMC: https://pmc.ncbi.nlm.nih.gov/articles/PMC4941872/) | peer-reviewed | high |

• Network accessline 85

  | S10 GRIM (Brown & Heathers) | https://en.wikipedia.org/wiki/GRIM_test ; https://jamesheathers.medium.com/the-grim-test-a-method-for-evaluating-published-research-9a4e5f05e870 | method (peer-reviewed origin) | high |

• Network accessline 86

  | S11 GRIMMER | https://peerj.com/preprints/2400/ | preprint/method | med |

• Network accessline 87

  | S12 Benford forensic | https://www.sciencedirect.com/science/article/abs/pii/S2666281723000161 | peer-reviewed | med |

• Network accessline 88

  | S13 Authorship-for-sale networks | https://pmc.ncbi.nlm.nih.gov/articles/PMC11604963/ | peer-reviewed | high |

• Network accessline 89

  | S14 CIDRE citation cartels | https://www.nature.com/articles/s41598-021-93572-3 | peer-reviewed | high |

• Network accessline 90

  | S15 COPE paper mills | https://publicationethics.org/topic-discussions/systematic-manipulation-publishing-process-paper-mills | community standard | high |

• Network accessline 91

  | S16 COPE salami slicing | https://publicationethics.org/guidance/cope-position/handling-duplicated-or-redundant-content-salami-slicing | community standard | high |

• Network accessline 92

  | S17 Problematic Paper Screener (tortured phrases) | https://arxiv.org/pdf/2402.03370 | peer-reviewed/tool | med |

• Network accessline 93

  | S18 Synthetic western blot forensics | https://arxiv.org/abs/2112.08739 | peer-reviewed | med |

skills/agentsop-bio-fraud-forensics/references/R02-image-forensics.md

• Network accessline 52

  - S1 | https://journals.asm.org/doi/full/10.1128/mbio.00809-16 (and PMC mirror PMC4941872) | Bik, Casadevall, Fang 2016 mBio — foundational duplication-prevalence paper, Category I/II/III definitions | High

• Network accessline 53

  - S2 | https://scienceintegritydigest.com/2020/02/21/western-blots/ | Bik blog — "every band is unique" visual-inspection method | High

• Network accessline 54

  - S3 | https://scienceintegritydigest.com/2019/11/23/scanning-for-duplications/ | Bik blog — Category I/II/III restated with intent ladder, flipped/shifted/rotated examples | High

• Network accessline 55

  - S4 | https://scienceintegritydigest.com/2024/02/15/the-rat-with-the-big-balls-and-enormous-penis-how-frontiers-published-a-paper-with-botched-ai-generated-images/ | Bik blog — Midjourney/Frontiers AI-figure case | High

• Network accessline 56

  - S5 | https://intercom.help/proofig/en/articles/9414427-analyzing-suspected-issues-in-western-blot-images-with-proofig-best-practices-and-tips | Proofig docs — emboss filter, histogram equalization for blots | High

• Network accessline 57

  - S6 | https://intercom.help/proofig/en/articles/10544884-histogram-equalization-in-image-forensics | Proofig docs — histogram equalization caveat ("visualization tool, not proof") | High

• Network accessline 58

  - S7 | https://29a.ch/photo-forensics/ + https://29a.ch/2015/08/21/noise-analysis-for-image-forensics | Forensically (Jonas Wagner) — ELA, clone detection, noise analysis, level sweep, magnifier | High - S8 | https://pmc.ncbi.nlm.nih.gov/ar

• Network accessline 59

  - S8 | https://pmc.ncbi.nlm.nih.gov/articles/PMC12505991/ (mBio 2025, DOI 10.1128/mbio.01990-25) | ASM/ImageTwin pilot — scale results, 3-stage verification, halftone-only limitation | High - S9 | https://www.science.org/doi/10.1126/science

• Network accessline 60

  - S9 | https://www.science.org/doi/10.1126/science.adn7530 | Science 2024 editorial "Genuine images in 2024" — Proofig adoption (403 on fetch; corroborated via search) | Medium

• Network accessline 61

  - S10 | https://tnqtech.com/imagetwin-launches-ai-generated-image-detection-capability/ | ImageTwin AI-generated-image detection launch (frequency-domain) | Medium

• Network accessline 62

  - S11 | https://www.proofig.com/newsroom/ (Nature AI image-integrity feature) | Proofig AI detection ~98%/0.02% FP — vendor-stated | Low-Medium (vendor claim)

• Network accessline 63

  - S12 | https://arxiv.org/html/2408.13786v1 | "Localization of Synthetic Manipulations in Western Blot Images" — patch-based synthetic-vs-real heatmap | Medium

• Network accessline 64

  - S13 | https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11323046/ | ELA-CNN integration framework for authenticity verification | Medium

• Network accessline 65

  - S14 | https://retractionwatch.com/2026/04/19/elisabeth-bik-scientific-sleuth-image-duplication-mbio-biorxiv-preprint/ | Retraction Watch 10-yr retrospective on Bik preprint | Medium

skills/agentsop-bio-fraud-forensics/references/R03-statistical-forensics.md

• Network accessline 61

  - S1, https://datacolada.org/109, blog (primary investigation, "Clusterfake"), high

• Network accessline 62

  - S2, https://datacolada.org/98, blog (primary investigation, dishonesty field experiment), high

• Network accessline 63

  - S3, https://journals.sagepub.com/doi/abs/10.1177/1948550616673876, peer-reviewed (Brown & Heathers 2017, GRIM, SPPS), high

• Network accessline 64

  - S4, https://en.wikipedia.org/wiki/GRIM_test, encyclopedia (GRIM conditions/limits, case summaries), medium

• Network accessline 65

  - S5, https://cran.r-project.org/web/packages/scrutiny/vignettes/grim.html, software docs (GRIM worked example), high - S6, https://cran.r-project.org/web/packages/scrutiny/vignettes/grimmer.html, software docs (GRIMMER worked example, test

• Network accessline 66

  - S6, https://cran.r-project.org/web/packages/scrutiny/vignettes/grimmer.html, software docs (GRIMMER worked example, test-3 bug), high - S7, https://peerj.com/preprints/2400.pdf, preprint (GRIMMER paper, Anaya/Brown/Heathers) — metadata vi

• Network accessline 67

  - S7, https://peerj.com/preprints/2400.pdf, preprint (GRIMMER paper, Anaya/Brown/Heathers) — metadata via search; PDF 403 on direct fetch, medium

• Network accessline 68

  - S8, https://mbnuijten.com/statcheck/ + https://pmc.ncbi.nlm.nih.gov/articles/PMC7540394/, software/peer-reviewed (statcheck, Nuijten/Epskamp), high

• Network accessline 69

  - S9, https://journals.sagepub.com/doi/10.1177/25152459241258945, peer-reviewed (Nuijten & Wicherts 2024, statcheck in peer review reduces inconsistencies) — abstract via search, medium

• Network accessline 70

  - S10, https://peerj.com/preprints/26968.pdf + https://medium.com/hackernoon/introducing-sprite-and-the-case-of-the-carthorse-child-58683c2bfeb, preprint/author blog (SPRITE method + Carthorse Child worked case), high

• Network accessline 71

  - S11, https://pmc.ncbi.nlm.nih.gov/articles/PMC10088595/, peer-reviewed (Benford's Law for misconduct: second/terminal digits, χ², limits), high

• Network accessline 72

  - S12, https://www.researchgate.net/publication/24083679 ("Not the First Digit!", Diekmann), peer-reviewed (Benford in science; favor higher digits), medium

• Network accessline 73

  - S13, https://github.com/QuentinAndre/pysprite + https://lukaswallrich.github.io/rsprite2/, software (SPRITE implementations Python/R), high

skills/agentsop-bio-fraud-forensics/references/R04-exposure-sites-method.md

• Network accessline 25

  ### Case 1: Data Colada [109] "Clusterfake" — calcChain.xml reveals manually moved rows in a dishonesty study - 情境 (Situation): A field-experiment dataset (the "signing-at-the-top" honesty study) was posted on OSF. Data Colada suspected fab

• Network accessline 26

  - 情境 (Situation): A field-experiment dataset (the "signing-at-the-top" honesty study) was posted on OSF. Data Colada suspected fabrication and obtained the original Excel files. Source: https://datacolada.org/109 ; companion https://datacol

• Network accessline 37

  ### Case 2: PubPeer (commenter "Actinopolyspora biskrensis") — duplicated/flipped blot bands and gel-slice overlap in a 2010 Cancer paper - 情境 (Situation): A 2010 oncology paper ("Adenovirus-mediated expression of truncated E2F-1 suppresses

• Network accessline 38

  - 情境 (Situation): A 2010 oncology paper ("Adenovirus-mediated expression of truncated E2F-1 suppresses tumor growth...", *Cancer*) was flagged on PubPeer. The same pseudonymous sleuth flagged related papers by the same group (a *Virology* p

• Network accessline 48

  ### Case 3 (supporting — paper-mill cluster): Smut Clyde + Tiger BB8 — template-motif clustering of a Chinese paper mill - 情境 (Situation): ~260+ fabricated papers (many in *J BUON*) traced to one Chinese paper mill. Source: https://forbette

• Network accessline 49

  - 情境 (Situation): ~260+ fabricated papers (many in *J BUON*) traced to one Chinese paper mill. Source: https://forbetterscience.com/2021/05/26/the-chinese-paper-mill-industry-interview-with-smut-clyde-and-tiger-bb8/

• Network accessline 91

  - S1, https://datacolada.org/109, exposure-method (Clusterfake calcChain protocol), high

• Network accessline 92

  - S2, https://datacolada.org/98, exposure-method (companion fraud-evidence post), high

• Network accessline 93

  - S3, https://retractionwatch.com/2024/07/19/mistakes-were-made-paper-by-department-chair-earns-expression-of-concern-as-more-questioned/, exposure-method + status (Actinopolyspora biskrensis PubPeer case, EoC), high

• Network accessline 94

  - S4, https://scienceintegritydigest.com/2019/11/23/scanning-for-duplications/, exposure-method (Bik Type I/II/III + workflow), high

• Network accessline 95

  - S5, https://forbetterscience.com/2021/05/26/the-chinese-paper-mill-industry-interview-with-smut-clyde-and-tiger-bb8/, exposure-method (paper-mill motif clustering; sleuth specialties), high

• Network accessline 96

  - S6, https://retractionwatch.com/retraction-watch-database-user-guide/, database (RW DB search fields/usage), high

• Network accessline 97

  - S7, https://www.crossref.org/documentation/retrieve-metadata/retraction-watch/, database (RW fields, Crossref hosting/API), high

• Network accessline 98

  - S8, https://ori.hhs.gov/content/case_summary, database (ORI adjudicated findings index), high

• Network accessline 99

  - S9, https://scite.ai/blog/2022-10-19_automated-notice-detection, database (scite retraction/notice detection), medium-high

• Network accessline 100

  - S10, https://en.wikipedia.org/wiki/GRIM_test, method (GRIM, Brown & Heathers 2016), medium-high

• Network accessline 101

  - S11, https://peerj.com/preprints/2400.pdf, method (GRIMMER, variability test), medium

• Network accessline 102

  - S12, https://retractionwatch.com/2026/04/19/elisabeth-bik-scientific-sleuth-image-duplication-mbio-biorxiv-preprint/, context (Bik 20,621-paper screen, 3.8% rate), high

skills/agentsop-bio-fraud-forensics/references/R05-evidence-red-lines.md

• Network accessline 140

  - S1 — https://scienceintegritydigest.com/2019/07/16/pubpeer-a-website-to-comment-on-scientific-papers/ — Bik's own blog explaining PubPeer norms ("similar" yes / "fraud" no) — high - S2 — https://www.pubpeer.com/static/faq — PubPeer offici

• Network accessline 141

  - S2 — https://www.pubpeer.com/static/faq — PubPeer official commenting/moderation rules (facts, publicly verifiable, no fraud accusations) — high (content confirmed via secondary quotes; direct fetch 403)

• Network accessline 142

  - S3 — https://reason.com/volokh/2024/09/11/prof-francesca-ginos-libel-claims-against-harvard-business-school-and-data-colada-dismissed/ — Volokh/Reason legal analysis of the Gino dismissal incl. quoted opinion-doctrine reasoning — high

• Network accessline 143

  - S4 — https://www.science.org/content/article/honesty-researcher-s-lawsuit-against-data-sleuths-dismissed — Science news on Gino suit dismissal — high (headline/summary; full text paywalled/403)

• Network accessline 144

  - S5 — https://www.thecrimson.com/article/2024/9/12/judge-dismisses-gino-lawsuit-defamation-charges/ — Harvard Crimson detail on Judge Joun's ruling (First Amendment, actual malice, partial denial for Harvard) — high

• Network accessline 145

  - S6 — https://www.aclu.org/cases/sarkar-v-doe-pubpeer-subpoena-challenge — ACLU case page, Sarkar v. Doe — high - S7 — https://www.nachtlaw.com/blog/2016/12/court-of-appeals-decides-internet-anonymity-case/ — Michigan COA ruling: opinion o

• Network accessline 146

  - S7 — https://www.nachtlaw.com/blog/2016/12/court-of-appeals-decides-internet-anonymity-case/ — Michigan COA ruling: opinion on disclosed facts not defamatory — high - S8 — https://publicationethics.org/guidance/flowchart/inappropriate-ima

• Network accessline 147

  - S8 — https://publicationethics.org/guidance/flowchart/inappropriate-image-manipulation-published-article — COPE flowchart: contact author → ask institution to investigate → expression of concern during investigation → correction vs retrac

• Network accessline 148

  - S9 — https://scholarlykitchen.sspnet.org/2024/04/23/guest-post-the-future-of-image-integrity-in-scientific-research/ — AACR/Proofig numbers (1,367 screened; 207 contacted; 204 honest mistakes; 4 withdrawn); microscopy tiling overlap as be

• Network accessline 149

  - S10 — https://pmc.ncbi.nlm.nih.gov/articles/PMC4941872/ — Bik et al. mBio prevalence study; loading-control same-experiment reuse not a problem; simple vs repositioned/altered duplication likelihood — high

• Network accessline 150

  - S11 — https://thepublicationplan.com/2022/11/29/spotting-fake-images-in-scientific-research-insights-from-science-integrity-consultant-elisabeth-bik/ — Bik's duplication categories and "half deliberate" estimate, report-to-journals stance

• Network accessline 151

  - S12 — https://www.statnews.com/2024/02/28/elisabeth-bik-scientific-integrity-research-misconduct/ — Bik: "don't want to falsely accuse anybody," needs "real proof," intent beyond image analysis — high

• Network accessline 152

  - S13 — https://pmc.ncbi.nlm.nih.gov/articles/PMC11573918/ (Cell Patterns) — image-splicing detection; splicing legitimate when declared; honest lab artifacts confused with splicing — medium-high

• Network accessline 153

  - S14 — https://arxiv.org/pdf/2108.12947 / https://arxiv.org/pdf/2412.03261 — JPEG compression artifacts cause forensic false positives — medium-high

• Network accessline 154

  - S15 — https://pmc.ncbi.nlm.nih.gov/articles/PMC4430488/ — PLOS ONE study: exoneration largely restores credibility (relevant to reputational stakes of premature accusation) — medium

skills/agentsop-bio-fraud-forensics/references/R06-screening-workflow.md

• Network accessline 92

  - S1, https://scienceintegritydigest.com/2019/11/23/scanning-for-duplications/ , practitioner blog (Bik), high — Bik's own scan order, "don't read text," 1–15 min/paper, image-type priorities. - S2, https://journals.asm.org/doi/10.1128/mbio

• Network accessline 93

  - S2, https://journals.asm.org/doi/10.1128/mbio.00809-16 , peer-reviewed (Bik mBio 2016), high — Category I/II/III duplication taxonomy. - S3, https://www.pubpeer.com/static/faq , primary platform guidance, high (retrieved via search snippe

• Network accessline 94

  - S3, https://www.pubpeer.com/static/faq , primary platform guidance, high (retrieved via search snippet; direct fetch 403) — perform-transformation-yourself rule, compression-artifact rejection, reprobe/loading-control innocent explanation

• Network accessline 95

  - S4, https://www.editage.com/insights/identifying-problematic-images-tips-and-resources-for-peer-reviewers , publisher-services guide, medium-high — image red-flag checklist (edges, backgrounds, halos), tools (ImageJ, Forensically, Adobe B

• Network accessline 96

  - S5, https://www.csescienceeditor.org/article/the-stm-integrity-hub/ , industry/society article, high — ~20 independent signals: duplicate submission, references, tortured phrases, metadata, AI-content, submission patterns.

• Network accessline 97

  - S6, https://www.editage.com/insights/paper-mills-and-the-erosion-of-research-credibility-researchers-beware , publisher-services guide, medium-high — affiliation/email/ORCID first-screen red flags, irrelevant recurring references.

• Network accessline 98

  - S7, https://scienceintegritydigest.com/2021/08/27/cassava-sciences-of-stocks-and-blots/ , practitioner blog (Bik), high — Case 1 specifics (lane counts, splice positions, boxed bands, background similarity).

• Network accessline 99

  - S8, https://www.enago.com/academy/grim-and-sprite-simple-tools-to-identify-errors-in-research/ , explainer, medium — GRIM/SPRITE mechanics.

• Network accessline 100

  - S9, https://pmc.ncbi.nlm.nih.gov/articles/PMC12722777/ , peer-reviewed/PMC, medium-high — multi-tool stats+image workflow with mandatory human confirmation.

• Network accessline 101

  - S10, https://www.alzforum.org/news/community-news/data-fabrication-ousted-nia-neuroscience-director-eliezer-masliah , news, high — Masliah image-duplication findings corroborate elite-lab fabrication (used to justify "don't exonerate on c

• Network accessline 102

  - S11, https://www.stm-assoc.org/papermillchecker/ , industry, medium — STM Papermill checker MVP; tool-as-accelerator framing.

skills/agentsop-bio-fraud-forensics/references/R07-paper-mill-signals.md

• Network accessline 50

  - S1 | https://thebulletin.org/2022/01/bosom-peril-is-not-breast-cancer-how-weird-computer-generated-phrases-help-researchers-find-scientific-publishing-fraud/ | journalism/expert (Cabanac/Labbé/Magazinov) | high — tortured-phrase examples,

• Network accessline 51

  - S2 | https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0213266 | peer-reviewed (Labbé, Grima, Byrne et al., PLOS ONE 2019) | high — Seek & Blastn primary methods paper

• Network accessline 52

  - S3 | https://link.springer.com/article/10.1007/s00210-023-02846-2 | peer-reviewed (Naunyn-Schmiedeberg's Arch Pharmacol 2023) | high — reagent-error rates in cancer journals

• Network accessline 53

  - S4 | https://scienceintegritydigest.com/2020/02/21/the-tadpole-paper-mill/ | expert blog (Elisabeth Bik) | high — Tadpole mill figure fingerprints, 400+ papers, institution counts

• Network accessline 54

  - S5 | https://febs.onlinelibrary.wiley.com/doi/10.1002/1873-3468.14143 | peer-reviewed editorial (FEBS Letters 2021) | high — 12-manuscript batch, templated figure recognition

• Network accessline 55

  - S6 | https://retractionwatch.com/2024/03/14/up-to-one-in-seven-of-submissions-to-hundreds-of-wiley-journals-show-signs-of-paper-mill-activity/ | journalism (Retraction Watch) | high — Wiley Papermill Detection, six signals, one-in-seven s

• Network accessline 56

  - S7 | https://retractionwatch.com/2023/12/19/hindawi-reveals-process-for-retracting-more-than-8000-paper-mill-articles/ | journalism (Retraction Watch) | high — Hindawi 8,000+ retractions, special-issue concentration

• Network accessline 57

  - S8 | https://www.nature.com/articles/s41598-024-71230-8 (open mirror: https://pmc.ncbi.nlm.nih.gov/articles/PMC11604963/ ; preprint https://arxiv.org/pdf/2401.04022) | peer-reviewed (Scientific Reports 2024) | high — authorship-for-sale n

• Network accessline 58

  - S9 | https://www.eurekalert.org/news-releases/1114511 | press release (QUT, 2024-2025) | medium-high — BERT cancer "spam filter", 250k/2.6M papers, 91% accuracy, 16% peak

• Network accessline 59

  - S10 | https://www.stm-assoc.org/stm-integrity-hub-incorporates-clear-skies-papermill-alarm-screening-tool/ | industry (STM/Clear Skies) | high — Papermill Alarm traffic-light, Integrity Hub >10 signals

• Network accessline 60

  - S11 | https://arxiv.org/pdf/2602.05930 (NeurIPS 2025 fabricated-citation taxonomy); https://arxiv.org/pdf/2602.06718 (GhostCite) | preprint | medium — AI fabricated-citation prevalence; recent, not yet peer-reviewed

• Network accessline 61

  - S12 | https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10028415/ | peer-reviewed | medium-high — AI-generated fabrication/plagiarism, "as an AI language model" leakage

• Network accessline 62

  - S13 | https://www.dimensions.ai/blog/detecting-tortured-phrases-to-unmask-fake-science/ | industry blog | medium — corroborates PPS scale and method

skills/agentsop-bounded-loop/SKILL.md

• Network accessline 98

  > "If you are not expecting your graph to go through many iterations, you > likely have a cycle. Check your logic for infinite loops." > — `[lc-docs/errors]` `https://docs.langchain.com/oss/python/langgraph/errors/GRAPH_RECURSION_LIMIT`

• Network accessline 100

  > — `[lc-docs/errors]` `https://docs.langchain.com/oss/python/langgraph/errors/GRAPH_RECURSION_LIMIT`

• Network accessline 369

  - **Source**: `[gh/6731]`   (https://github.com/langchain-ai/langgraph/issues/6731), maintainer

• Network accessline 370

  (https://github.com/langchain-ai/langgraph/issues/6731), maintainer

• Network accessline 417

  - **Source**:   [github.com/crewAIInc/crewAI/issues/330](https://github.com/crewAIInc/crewAI/issues/330)

• Network accessline 418

  [github.com/crewAIInc/crewAI/issues/330](https://github.com/crewAIInc/crewAI/issues/330)

• Network accessline 420

  ["the delegation ping-pong"](https://azguards.com/technical/the-delegation-ping-pong-breaking-infinite-handoff-loops-in-crewai-hierarchical-topologies/).

• Network accessline 601

  moves the failure further away while burning more tokens. The LM in the loop is **never** a reliable terminator. Source-of-truth case: LangGraph issue [#6731](https://github.com/langchain-ai/langgraph/issues/6731)

• Network accessline 602

  the loop is **never** a reliable terminator. Source-of-truth case: LangGraph issue [#6731](https://github.com/langchain-ai/langgraph/issues/6731) marked "not planned" — the framework will not save you; the discipline

• Network accessline 603

  LangGraph issue [#6731](https://github.com/langchain-ai/langgraph/issues/6731)

skills/agentsop-bounded-loop/references/R1-source-evidence.md

• Network accessline 12

  - **Primary**: [github.com/langchain-ai/langgraph/issues/6731](https://github.com/langchain-ai/langgraph/issues/6731)

• Network accessline 18

  > "If you are not expecting your graph to go through many iterations, you   > likely have a cycle. Check your logic for infinite loops."   > — [docs.langchain.com/oss/python/langgraph/errors/GRAPH_RECURSION_LIMIT](https://docs.langchain.com

• Network accessline 19

  > likely have a cycle. Check your logic for infinite loops."   > — [docs.langchain.com/oss/python/langgraph/errors/GRAPH_RECURSION_LIMIT](https://docs.langchain.com/oss/python/langgraph/errors/GRAPH_RECURSION_LIMIT) - **Cheatsheet reinforce

• Network accessline 20

  > — [docs.langchain.com/oss/python/langgraph/errors/GRAPH_RECURSION_LIMIT](https://docs.langchain.com/oss/python/langgraph/errors/GRAPH_RECURSION_LIMIT)

• Network accessline 23

  > recursion limit is a safety net for runaway code, not a primary control   > flow mechanism."   > — [sumanmichael.github.io/langgraph-cheatsheet — FAQs & Gotchas](https://sumanmichael.github.io/langgraph-cheatsheet/cheatsheet/faqs-gotchas/

• Network accessline 24

  > flow mechanism."   > — [sumanmichael.github.io/langgraph-cheatsheet — FAQs & Gotchas](https://sumanmichael.github.io/langgraph-cheatsheet/cheatsheet/faqs-gotchas/) - **Implication used in SKILL.md**:

• Network accessline 25

  > — [sumanmichael.github.io/langgraph-cheatsheet — FAQs & Gotchas](https://sumanmichael.github.io/langgraph-cheatsheet/cheatsheet/faqs-gotchas/)

• Network accessline 35

  - **Primary issue**: [github.com/crewAIInc/crewAI/issues/330](https://github.com/crewAIInc/crewAI/issues/330)

• Network accessline 37

  ("A delegates to B, B delegates back to A, …"). Token cost blew past budget   before any timeout fired. - **Related issues**: [#4783](https://github.com/crewAIInc/crewAI/issues/4783),

• Network accessline 38

  before any timeout fired. - **Related issues**: [#4783](https://github.com/crewAIInc/crewAI/issues/4783),   [#2606](https://github.com/crewAIInc/crewAI/issues/2606).

• Network accessline 39

  - **Related issues**: [#4783](https://github.com/crewAIInc/crewAI/issues/4783),

• Network accessline 40

  [#2606](https://github.com/crewAIInc/crewAI/issues/2606).

• Network accessline 41

  - **Root-cause writeup**:   [azguards.com — "The delegation ping-pong: breaking infinite handoff loops   in CrewAI hierarchical topologies"](https://azguards.com/technical/the-delegation-ping-pong-breaking-infinite-handoff-loops-in-crewai-h

• Network accessline 42

  [azguards.com — "The delegation ping-pong: breaking infinite handoff loops   in CrewAI hierarchical topologies"](https://azguards.com/technical/the-delegation-ping-pong-breaking-infinite-handoff-loops-in-crewai-hierarchical-topologies/)   d

• Network accessline 43

  in CrewAI hierarchical topologies"](https://azguards.com/technical/the-delegation-ping-pong-breaking-infinite-handoff-loops-in-crewai-hierarchical-topologies/)

• Network accessline 62

  - **Primary**: [aider.chat/docs/usage/lint-test.html](https://aider.chat/docs/usage/lint-test.html)

• Network accessline 83

  recommends an explicit `max_iterations` / `tool_use_budget` controlled by     the caller. See     [docs.anthropic.com/en/api/messages](https://docs.anthropic.com/en/api/messages)

• Network accessline 84

  the caller. See     [docs.anthropic.com/en/api/messages](https://docs.anthropic.com/en/api/messages)     and the Claude Agent SDK pattern docs.

• Network accessline 85

  [docs.anthropic.com/en/api/messages](https://docs.anthropic.com/en/api/messages)

• Network accessline 104

  — the budget is *declared* up front. No notion of "let the optimiser   decide when to stop." - **Source**: [dspy.ai/docs/building-blocks/optimizers](https://dspy.ai/docs/building-blocks/optimizers)

• Network accessline 105

  decide when to stop." - **Source**: [dspy.ai/docs/building-blocks/optimizers](https://dspy.ai/docs/building-blocks/optimizers)   and [github.com/stanfordnlp/dspy](https://github.com/stanfordnlp/dspy).

• Network accessline 106

  - **Source**: [dspy.ai/docs/building-blocks/optimizers](https://dspy.ai/docs/building-blocks/optimizers)

• Network accessline 107

  and [github.com/stanfordnlp/dspy](https://github.com/stanfordnlp/dspy).

skills/agentsop-bounded-loop/references/R2-cross-framework.md

• Network accessline 11

  | Framework | Generic safety net | Recommended explicit bound | Where the counter lives | Known footgun | |---|---|---|---|---| | **LangGraph** | `recursion_limit=25` (`GRAPH_RECURSION_LIMIT`) | retry counter in `TypedDict` state + conditio

• Network accessline 12

  |---|---|---|---|---| | **LangGraph** | `recursion_limit=25` (`GRAPH_RECURSION_LIMIT`) | retry counter in `TypedDict` state + conditional edge to `END` | `state["retries"]: Annotated[int, operator.add]` | Raising `recursion_limit` to 100 to

• Network accessline 13

  | **LangGraph** | `recursion_limit=25` (`GRAPH_RECURSION_LIMIT`) | retry counter in `TypedDict` state + conditional edge to `END` | `state["retries"]: Annotated[int, operator.add]` | Raising `recursion_limit` to 100 to "fix" a loop — issue 

• Network accessline 14

  | **CrewAI** | `max_iter=20` per Agent, `max_rpm` per Crew | `max_iter` per agent **plus** `allow_delegation=False` on workers + outer timeout | Per-Agent (does NOT cross delegation handoffs — issue [#330](https://github.com/crewAIInc/crewA

skills/agentsop-code-execution-decision/SKILL.md

• Dangerous shell operationline 165

  | 3 | **Sandbox choice** | Decided to emit code | Pick interpreter by trust + capability: DSPy PoT (Python sandbox), OpenAI Code Interpreter (managed container), Anthropic code-exec (VM), LangChain `PythonREPLTool` (**unsandboxed, in-proces

skills/agentsop-conventions-pinning/SKILL.md

• Network accessline 366

  | 1 | Rewrite the rule to be more specific. "Format properly" → "Use 2-space indent, double quotes, no trailing comma in JSON" | | 2 | Move it earlier in the file (recency in context matters) and bold it | | 3 | Add a counter-example: "Wron

• Network accessline 367

  | 2 | Move it earlier in the file (recency in context matters) and bold it | | 3 | Add a counter-example: "Wrong: `requests.get(url)`. Right: `httpx.get(url, timeout=10)`" | | 4 | If the rule MUST hold, demote it to a hook / lint / CI gate.

• Network accessline 368

  | 3 | Add a counter-example: "Wrong: `requests.get(url)`. Right: `httpx.get(url, timeout=10)`" |

skills/agentsop-cost-tiered-models/SKILL.md

• Network accessline 15

  audience: coder-agents and engineers designing multi-call LM pipelines sources:   - https://aider.chat/2024/09/26/architect.html

• Network accessline 16

  sources:   - https://aider.chat/2024/09/26/architect.html   - https://github.com/stanfordnlp/dspy/issues/1596

• Network accessline 17

  - https://aider.chat/2024/09/26/architect.html

• Network accessline 18

  - https://github.com/stanfordnlp/dspy/issues/1596

• Network accessline 19

  - https://docs.vllm.ai/en/latest/features/speculative_decoding/

• Network accessline 20

  - https://langchain-ai.github.io/langgraph/concepts/multi_agent/

skills/agentsop-cost-tiered-models/intermediate/operation_candidates.json

• Network accessline 38

  "output": "Two-step pipeline: propose (strong, free text) -> apply (cheap, strict format).",       "evidence": [         "https://aider.chat/2024/09/26/architect.html",

• Network accessline 39

  "evidence": [         "https://aider.chat/2024/09/26/architect.html",         "Polyglot Pass@2: o1-preview 79.7% -> o1-preview+Sonnet 82.7% -> o1-preview+o1-mini 85%; Sonnet self-paired 77.4% -> 80.5%"

• Network accessline 40

  "https://aider.chat/2024/09/26/architect.html",

• Network accessline 53

  "output": "Execution tier with an escalation valve; escalation-rate metric.",       "evidence": [         "https://docs.vllm.ai/en/latest/features/speculative_decoding/ (draft proposes, target verifies, reject->correct)",

• Network accessline 54

  "evidence": [         "https://docs.vllm.ai/en/latest/features/speculative_decoding/ (draft proposes, target verifies, reject->correct)",         "https://aider.chat/docs/troubleshooting/edit-errors.html (escalate model / change format on e

• Network accessline 55

  "https://docs.vllm.ai/en/latest/features/speculative_decoding/ (draft proposes, target verifies, reject->correct)",

• Network accessline 56

  "https://aider.chat/docs/troubleshooting/edit-errors.html (escalate model / change format on edit failure)"

• Network accessline 67

  "output": "Per-Tier-E-call format choice.",       "evidence": [         "https://aider.chat/2024/08/14/code-in-json.html (all models, incl. Sonnet, do worse wrapped in JSON)"

• Network accessline 68

  "evidence": [         "https://aider.chat/2024/08/14/code-in-json.html (all models, incl. Sonnet, do worse wrapped in JSON)"       ]

• Network accessline 69

  "https://aider.chat/2024/08/14/code-in-json.html (all models, incl. Sonnet, do worse wrapped in JSON)"

• Network accessline 81

  "output": "optimizer-LM (cheap) vs task-LM (expensive) assignment.",       "evidence": [         "https://github.com/stanfordnlp/dspy/issues/1596 (gpt-4o-mini optimizer + gpt-4o task: parity, lower cost)"

• Network accessline 82

  "evidence": [         "https://github.com/stanfordnlp/dspy/issues/1596 (gpt-4o-mini optimizer + gpt-4o task: parity, lower cost)"       ]

• Network accessline 83

  "https://github.com/stanfordnlp/dspy/issues/1596 (gpt-4o-mini optimizer + gpt-4o task: parity, lower cost)"

• Network accessline 94

  "output": "Fine-tuned small executor + strong decision model combo.",       "evidence": [         "https://dspy.ai/api/optimizers/BootstrapFinetune/ (student=Llama-3.2-1B, teacher=gpt-4o-mini)",

• Network accessline 95

  "evidence": [         "https://dspy.ai/api/optimizers/BootstrapFinetune/ (student=Llama-3.2-1B, teacher=gpt-4o-mini)",         "output/dspy-sop-skill/SKILL.md Case B (GPT-4 pipeline breaks on Llama-3-8B)"

• Network accessline 96

  "https://dspy.ai/api/optimizers/BootstrapFinetune/ (student=Llama-3.2-1B, teacher=gpt-4o-mini)",

• Network accessline 109

  "output": "supervisor=Tier-S, workers=Tier-E orchestration.",       "evidence": [         "https://langchain-ai.github.io/langgraph/concepts/multi_agent/",

• Network accessline 110

  "evidence": [         "https://langchain-ai.github.io/langgraph/concepts/multi_agent/",         "output/langgraph-sop-skill/SKILL.md Case 2 (supervisor uses more tokens than swarm due to translation)"

• Network accessline 111

  "https://langchain-ai.github.io/langgraph/concepts/multi_agent/",

• Network accessline 119

  "title": "Is Aider architect mode worth the cost?",       "rule": "Gain = (quality delta) x (task value) - (extra call cost). Reasoner already clean at editing -> delta ~= 0, don't split. Reasoner editing dirty (o1/o3) -> delta large (79.7%

• Network accessline 120

  "rule": "Gain = (quality delta) x (task value) - (extra call cost). Reasoner already clean at editing -> delta ~= 0, don't split. Reasoner editing dirty (o1/o3) -> delta large (79.7%->82.7%->85%), split.",       "evidence": "https://aider.c

• Network accessline 121

  "evidence": "https://aider.chat/2024/09/26/architect.html"

• Network accessline 126

  "rule": "Add detectable failure signals; escalate failing steps only (not the whole tier). Track escalation rate: <10% healthy; 10-30% watch; >30% the step belongs in Tier-S, reclassify. Never probabilistically escalate everything (= back t

• Network accessline 127

  "evidence": [         "https://docs.vllm.ai/en/latest/features/speculative_decoding/ (high QPS -> speculation steals capacity -> disable)",         "https://aider.chat/docs/troubleshooting/edit-errors.html"

• Network accessline 128

  "https://docs.vllm.ai/en/latest/features/speculative_decoding/ (high QPS -> speculation steals capacity -> disable)",

• Network accessline 129

  "https://aider.chat/docs/troubleshooting/edit-errors.html"

skills/agentsop-crewai/SKILL.md

• Network accessline 479

  ## Sources (核心引用) - [docs.crewai.com/en/concepts/agents](https://docs.crewai.com/en/concepts/agents)

• Network accessline 480

  - [docs.crewai.com/en/concepts/agents](https://docs.crewai.com/en/concepts/agents)

• Network accessline 481

  - [docs.crewai.com/en/concepts/tasks](https://docs.crewai.com/en/concepts/tasks)

• Network accessline 482

  - [docs.crewai.com/en/concepts/flows](https://docs.crewai.com/en/concepts/flows)

• Network accessline 483

  - [docs.crewai.com/en/concepts/memory](https://docs.crewai.com/en/concepts/memory)

• Network accessline 484

  - [docs.crewai.com/en/learn/hierarchical-process](https://docs.crewai.com/en/learn/hierarchical-process)

• Network accessline 485

  - [github.com/crewAIInc/crewAI](https://github.com/crewAIInc/crewAI)

• Network accessline 486

  - [github.com/crewAIInc/crewAI/discussions/1220](https://github.com/crewAIInc/crewAI/discussions/1220) — manager_agent 正确用法讨论

• Network accessline 487

  - [github.com/crewAIInc/crewAI/issues/330](https://github.com/crewAIInc/crewAI/issues/330) — allow_delegation 无限循环

• Network accessline 488

  - [towardsdatascience.com Manager-Worker fails](https://towardsdatascience.com/why-crewais-manager-worker-architecture-fails-and-how-to-fix-it/)

• Network accessline 489

  - [azguards.com/the-delegation-ping-pong](https://azguards.com/technical/the-delegation-ping-pong-breaking-infinite-handoff-loops-in-crewai-hierarchical-topologies/)

• Network accessline 490

  - [softwareengineeringdaily.com/2025/06/03/crew-ai-with-joao-moura](https://softwareengineeringdaily.com/2025/06/03/crew-ai-with-joao-moura/) — João Moura "agency" 哲学

• Network accessline 491

  - [callsphere.ai/blog/crewai-process-types](https://callsphere.ai/blog/crewai-process-types-sequential-hierarchical-consensual-workflows)

• Network accessline 492

  - [community.crewai.com/t/5710](https://community.crewai.com/t/choosing-between-sequential-and-hierarchical-processes-in-crewai-for-a-shopping-chatbot/5710)

• Network accessline 493

  - [mem0.ai/blog/crewai-memory-production-setup-with-mem0](https://mem0.ai/blog/crewai-memory-production-setup-with-mem0)

• Network accessline 494

  - [datacamp.com/tutorial/crewai-vs-langgraph-vs-autogen](https://www.datacamp.com/tutorial/crewai-vs-langgraph-vs-autogen)

skills/agentsop-crewai/references/R5-ecosystem-context.md

• Network accessline 124

  ## 6. Sources - [datacamp.com/tutorial/crewai-vs-langgraph-vs-autogen](https://www.datacamp.com/tutorial/crewai-vs-langgraph-vs-autogen)

• Network accessline 125

  - [datacamp.com/tutorial/crewai-vs-langgraph-vs-autogen](https://www.datacamp.com/tutorial/crewai-vs-langgraph-vs-autogen)

• Network accessline 126

  - [latenode.com 2026 comparison](https://latenode.com/blog/platform-comparisons-alternatives/automation-platform-comparisons/langgraph-vs-autogen-vs-crewai-complete-ai-agent-framework-comparison-architecture-analysis-2025)

• Network accessline 127

  - [galileo.ai/blog/autogen-vs-crewai-vs-langgraph-vs-openai-agents-framework](https://galileo.ai/blog/autogen-vs-crewai-vs-langgraph-vs-openai-agents-framework)

• Network accessline 128

  - [aaronyuqi.medium.com firsthand comparison](https://aaronyuqi.medium.com/first-hand-comparison-of-langgraph-crewai-and-autogen-30026e60b563)

• Network accessline 129

  - [python.plainenglish.io honest comparison](https://python.plainenglish.io/autogen-vs-langgraph-vs-crewai-a-production-engineers-honest-comparison-d557b3b9262c)

skills/agentsop-dify/SKILL.md

• Network accessline 176

  git clone https://github.com/langgenius/dify

• Access to secretsline 178

  [Credential-related source omitted]

• Network accessline 504

  - 主页 / Repo: https://github.com/langgenius/dify

• Network accessline 505

  - 文档: https://docs.dify.ai/

• Network accessline 506

  - Marketplace: https://marketplace.dify.ai/

• Network accessline 507

  - 官方插件: https://github.com/langgenius/dify-official-plugins

• Network accessline 508

  - Cloud: https://cloud.dify.ai/

• Network accessline 509

  - 中文文档镜像: https://hellodify.com/

skills/agentsop-dify/intermediate/operation_candidates.json

• Network accessline 166

  ],   "key_sources": [     "https://github.com/langgenius/dify",

• Network accessline 167

  "key_sources": [     "https://github.com/langgenius/dify",     "https://docs.dify.ai/",

• Network accessline 168

  "https://github.com/langgenius/dify",

• Network accessline 169

  "https://docs.dify.ai/",

• Network accessline 170

  "https://marketplace.dify.ai/",

• Network accessline 171

  "https://dify.ai/blog/",

• Network accessline 172

  "https://memo.d.foundation/breakdown/dify",

• Network accessline 173

  "https://blog.elest.io/dify-vs-langflow-vs-flowise-which-open-source-llm-app-builder-actually-ships-to-production/",

• Network accessline 174

  "https://jimmysong.io/blog/open-source-ai-agent-workflow-comparison/",

• Network accessline 175

  "https://sider.ai/blog/ai-tools/dify-vs-ragflow-which-rag-platform-should-you-build-on-in-2025",

• Network accessline 176

  "https://hellodify.com/en/docs/workflow/workflow-chatflow-difference",

• Network accessline 177

  "https://zediot.com/blog/dify-difference-between-agent-and-workflow/",

• Network accessline 178

  "https://learnwithparam.com/blog/batteries-included-rag-platforms-dify-ragflow-onyx",

• Network accessline 179

  "https://zhuanlan.zhihu.com/p/1947389040702781389",

• Network accessline 180

  "https://github.com/langgenius/dify/issues/21455",

• Network accessline 181

  "https://github.com/langgenius/dify/issues/28245",

• Network accessline 182

  "https://github.com/langgenius/dify/discussions/26138",

• Network accessline 183

  "https://github.com/langgenius/dify/discussions/27053",

• Network accessline 184

  "https://github.com/langgenius/dify/discussions/32254",

• Network accessline 185

  "https://architjn.com/blog/dify-cloud-pricing-plans-free-tier-when-to-self-host"

skills/agentsop-dify/references/R1-architecture.md

• Access to secretsline 122

  [Credential-related source omitted]

skills/agentsop-dify/references/R2-sop-workflow.md

• Network accessline 36

  git clone https://github.com/langgenius/dify

• Access to secretsline 38

  [Credential-related source omitted]

skills/agentsop-http-tool-wrapping/SKILL.md

• Access to secretsline 189

  [Credential-related source omitted]

skills/agentsop-http-tool-wrapping/intermediate/operation_candidates.json

• Access to secretsline 38

  [Credential-related source omitted]

skills/agentsop-http-tool-wrapping/references/R2-pattern-library.md

• Network accessline 115

  [Credential-related source omitted]

• Network accessline 116

  [Credential-related source omitted]

• Access to secretsline 116

  [Credential-related source omitted]

• Network accessline 117

  return httpx.Client(base_url="https://api.example.com",

• Access to secretsline 121

  [Credential-related source omitted]

skills/agentsop-hybrid-retrieval/SKILL.md