[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M74MHJVM1GRMTESYA2CC5KD4":3,"skill-same:01M74MHJVM1GRMTESYA2CC5KD4":148},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":16,"sourceHtmlUrl":17,"categories":18,"keywords":19,"publishedAt":20,"linkOnly":21,"markdownContent":22,"markdownBody":23,"frontMatter":24,"refName":25,"commitSha":26,"skillDirectory":27,"forksCount":28,"repositoryHtmlUrl":29,"repositoryHomepageUrl":30,"ownerHtmlUrl":31,"ownerBlogUrl":6,"ownerDescription":32,"updatedAt":33,"install":34,"resources":47},"01M74MHJVM1GRMTESYA2CC5KD4","baseline-selector",null,"Use when selecting, auditing, or justifying experimental baselines for a research idea, paper draft, task, benchmark, or experiment plan. Finds classic methods, recent SOTA, direct competitors, strong simple baselines, ablation baselines, resource-matched comparisons, and reviewer-expected missing baselines, but only admits baselines with non-empty, accessible GitHub repositories that appear reproducible. Records strong papers without usable open-source code as excluded candidates rather than selected baselines.","pengqianhan\u002FAI-Human-Research-OS","pengqianhan","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F4286531?v=4","01M74M3DT9HNZ6HPAK8TX69EX9","PengqianHan","01M74M3DTVSN2GXW0JH2W396FK",3,"MIT","https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fcollected-skills\u002Fbaseline-selector\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fcollected-skills\u002Fbaseline-selector\u002FSKILL.md",[],[],"2026-11-11 15:11:14",false,"---\nname: baseline-selector\ndescription: Use when selecting, auditing, or justifying experimental baselines for a research idea, paper draft, task, benchmark, or experiment plan. Finds classic methods, recent SOTA, direct competitors, strong simple baselines, ablation baselines, resource-matched comparisons, and reviewer-expected missing baselines, but only admits baselines with non-empty, accessible GitHub repositories that appear reproducible. Records strong papers without usable open-source code as excluded candidates rather than selected baselines.\n---\n\n# Baseline Selector\n\nUse this skill to turn a research idea, task, paper, or experiment plan into a defensible baseline set for experiments and paper submission.\n\nCore rule:\n\n```text\nNo usable GitHub implementation, no selected baseline.\n```\n\nA method may be important, classic, or SOTA, but it cannot enter the recommended baseline set unless its GitHub repository is accessible, non-empty, and plausibly reproducible. Record such methods as excluded candidates with the exclusion reason.\n\n## Modes\n\nDecide the mode from the user request:\n\n1. **Idea to baselines**: user gives a new idea or method sketch.\n2. **Task to SOTA**: user asks for baselines for a task, dataset, or benchmark.\n3. **Paper audit**: user provides a draft or existing baseline list to check for gaps.\n4. **Reviewer-risk check**: user asks what reviewers will say is missing.\n5. **Resource-limited selection**: user gives compute\u002Ftime limits and needs a minimal set.\n6. **Reproduction planning**: user already has candidates and needs runnable code choices.\n\n## Output Profiles\n\nChoose an output profile based on the user's real objective. The same task may produce different baseline recommendations under different profiles.\n\n1. **Paper submission mode**: prioritize completeness, reviewer-expected baselines, and stronger defensive coverage.\n2. **Fast prototype mode**: prioritize the smallest meaningful runnable set for quick signal.\n3. **Limited compute mode**: prioritize budget-compatible baselines and explicitly defer expensive ones.\n4. **Industry comparison mode**: prioritize practical open-source systems and deployment-relevant baselines.\n5. **Rebuttal emergency mode**: prioritize the missing baselines a reviewer is most likely to demand immediately.\n\nIf the user does not specify a profile, infer one from the request and state it explicitly in the report.\n\n## Required Inputs\n\nProceed with available information, but first infer or ask only for missing details that materially affect baseline scope:\n\n```text\nresearch idea \u002F task:\ndomain:\ndataset or benchmark:\nmetric:\ntarget venue and year:\ncompute budget \u002F hardware limit:\ntime budget:\nimplementation constraints:\noutput profile:\n```\n\nTarget venue is required for final recommendation quality. Examples: `ICML 2027`, `NeurIPS 2028`, `CVPR 2027`, `ACL 2026`.\nCompute budget is required for final baseline selection. Ask for GPU type\u002Fcount, expected run time, and whether the user needs full retraining, fine-tuning only, or inference-only baselines.\n\nIf benchmark, metric, or dataset are unknown, state assumptions and mark the baseline set as provisional.\n\n## Response Language Rule\n\nThe language of the final user-facing result should follow the language the user uses as closely as possible. Only proper nouns or standard technical names that are not suitable for translation should remain in English.\n\n- Answer body text in the user's language.\n- Titles, field names, and decision labels should also follow the user's language whenever practical.\n- Proper nouns such as method names, benchmark names, venue names, repository names, and standard technical terms may remain in English when translation would reduce clarity.\n- If the user mixes languages, follow the dominant language unless the user explicitly requests bilingual output.\n- Do not switch to another language for convenience if the user did not ask for it.\n- If a structured report is written to files, the file contents should follow the same rule.\n\n## Main Workflow\n\n0. **Get the current date**\n   - Before searching, obtain the current date from the runtime or environment instead of assuming the model's training cutoff.\n   - Record `as_of_date`, current year, and freshness windows such as last 6 months, last 12 months, and last 24 months.\n   - Use `references\u002Ffreshness-protocol.md` to decide what counts as recent, latest, or classic.\n   - The final report must state the search date and freshness window.\n\n1. **Frame the task**\n   - Convert the user request into a precise comparison problem.\n   - Identify input, output, dataset, metric, training setting, inference setting, and claim type.\n   - Route the task through `references\u002Fdomain-routing.md` to align the search with field-specific benchmark habits and reviewer expectations.\n   - Use `references\u002Fsearch-protocol.md` for search scope and query planning.\n\n2. **Search for candidates**\n   - Collect classic methods, direct competitors, recent SOTA, strong simple methods, benchmark official baselines, and reviewer-expected methods.\n   - Prefer official benchmark pages, survey papers, same-dataset papers, top-conference papers, leaderboards, arXiv, Papers with Code, and GitHub.\n   - Include date-bounded searches for the latest candidates, especially papers and repositories released within the freshness windows.\n   - Record the real publication year and venue for every selected or near-selected method, such as `ICML 2025`, `NeurIPS 2024`, or `CVPR 2026`.\n   - Track search terms and source types.\n\n3. **Apply the GitHub reproducibility gate**\n   - Use `references\u002Fgithub-reproducibility.md`.\n   - A selected baseline must have an accessible GitHub repository, non-empty implementation, runnable instructions, dependency\u002Fenvironment notes, and enough code\u002Fcheckpoints\u002Fconfigs to reproduce or fairly rerun.\n   - Exclude papers with missing code, dead links, empty repositories, placeholder repositories, code-only snippets, non-GitHub code dumps, or repositories that cannot reasonably run the claimed method.\n\n4. **Classify baseline roles**\n   - Use `references\u002Fbaseline-taxonomy.md`.\n   - Assign each candidate one or more roles: classic anchor, direct competitor, current SOTA, simple strong baseline, resource-matched baseline, practical baseline, ablation baseline, or reviewer-expected baseline.\n\n5. **Build evidence tables**\n   - Use `references\u002Fevidence-rules.md`.\n   - Every candidate needs paper source, real publication year, venue, GitHub source, task match, benchmark\u002Fmetric match, code status, reproduction risk, and selection decision.\n   - Never call a method SOTA unless the task, benchmark, metric, date, and source are explicit.\n\n6. **Select baseline sets**\n   - Produce three sets whenever possible:\n     - `minimal`: lowest cost set that still makes the comparison meaningful.\n     - `standard`: expected paper-quality set for the task.\n     - `defensive`: stronger set for top-venue reviewer pressure.\n   - Tailor the recommendation to the user's compute budget, time budget, and target venue\u002Fyear.\n   - Tailor the recommendation to the active output profile.\n   - In `limited compute mode`, explicitly separate `must run`, `nice to run`, and `too expensive for current budget`.\n   - In `rebuttal emergency mode`, prioritize the smallest set of baselines that closes reviewer-visible gaps fastest.\n   - Include only GitHub-reproducible candidates in these sets.\n\n7. **Plan reproduction**\n   - For each selected baseline, list repository URL, install path, expected data preprocessing, command\u002Fconfig clues, checkpoint availability, expected compute, and main failure risks.\n   - Prefer official repositories over unofficial ones unless the official repository is unusable and the unofficial repository is clearly maintained and faithful.\n\n8. **Run reviewer-risk audit**\n   - Use `references\u002Freviewer-risk.md`.\n   - Identify missing classic methods, missing recent competitors, metric mismatch, split mismatch, compute unfairness, no significance testing, and omitted simple baselines.\n   - Separate real risks from risks already handled by excluded non-reproducible methods.\n\n9. **Run self-check**\n   - Use `references\u002Fself-check.md`.\n   - Verify that every selected baseline has the correct year, venue, task match, dataset match, metric match, and GitHub status.\n   - Verify that no selected baseline violates the compute budget or target venue strategy.\n   - Verify that major expected baseline buckets were considered and that any omitted bucket is explicitly justified.\n   - Verify that important non-reproducible or non-matching methods are recorded outside the selected set.\n\n10. **Write final report**\n   - Use `templates\u002Fbaseline-report.md`.\n   - If writing files, prefer `baseline_selection\u002F` with numbered outputs.\n   - If answering in chat, follow the same section order but keep it concise.\n\n## Output Files\n\nWhen the user wants files, use this structure:\n\n```text\nbaseline_selection\u002F\n|- 01_task_definition.md\n|- 02_search_strategy.md\n|- 03_candidate_baselines.md\n|- 04_excluded_nonreproducible.md\n|- 05_recommended_baseline_sets.md\n|- 06_reproduction_plan.md\n|- 07_reviewer_risk_check.md\n|- 08_self_check.md\n`- 09_final_decision.md\n```\n\n## Source Label Sets\n\nChoose one source label set at the start of the response and use it consistently.\n\n### English Label Set\n\nUse this set when the user asks in English:\n\n```text\nDecision Labels\nselected\nselected-if-budget-allows\nselected-if-run-check-passes\nselected-as-harness\nexcluded-no-github\nexcluded-empty-repo\nexcluded-unrunnable-repo\nexcluded-task-mismatch\nexcluded-metric-mismatch\nexcluded-superseded\nwatchlist\nunknown-needs-verification\n\nFinal Verdict Labels\nstrong baseline set\nacceptable baseline set\nrisky baseline set\nincomplete baseline set\n\nSection Titles\nTask Definition\nSearch Strategy\nCandidate Baselines\nGitHub Evidence Chain\nExcluded Non-Reproducible or Non-Selectable Candidates\nRecommended Baseline Sets\nProfile-Aware Recommendation Lens\nMinimal Set\nStandard Set\nDefensive Set\nRecommendation for This User\nReproduction Plan\nReviewer Risk Check\nSelf-Check\nPer-Baseline Verification\nCoverage Check\nFinal Verdict\nMust Run\nNice To Run\n```\n\n### Chinese Label Set\n\nUse this set when the user asks in Chinese:\n\n```text\n判定标签\n入选\n预算允许时入选\n运行检查通过后入选\n作为辅助框架纳入\n因无 GitHub 仓库排除\n因空仓库排除\n因仓库不可运行排除\n因任务不匹配排除\n因指标不匹配排除\n因已被替代排除\n观察名单\n待核验\n\n最终结论标签\n强 baseline 集合\n可接受的 baseline 集合\n风险较高的 baseline 集合\n不完整的 baseline 集合\n\n章节标题\n任务定义\n搜索策略\n候选 Baseline\nGitHub 证据链\n排除项与未纳入候选\n推荐 Baseline 集合\n画像化推荐视角\n最小集合\n标准集合\n防守型集合\n面向当前用户的推荐\n复现计划\n审稿风险检查\n自检\n逐个 Baseline 核验\n覆盖检查\n最终结论\n必须运行\n建议补充\n```\n\nDo not output the English source labels directly in a Chinese response unless the label itself is a proper noun or the user explicitly asks for the English wording.\n\n## Final Verdict\n\nAlways give one final verdict from the active source label set.\n\nExplain the verdict using:\n\n- coverage of classic anchors;\n- coverage of recent SOTA\u002Fdirect competitors;\n- GitHub reproducibility status;\n- resource feasibility;\n- reviewer risk;\n- exact next actions.\n\n## Non-Negotiables\n\n- Always get and record the current date before deciding what counts as latest or recent.\n- Always record the real publication year and venue for every selected baseline.\n- Always require a target venue\u002Fyear and a compute budget before finalizing the recommended set.\n- Always state the active output profile before recommending the final set.\n- Always self-check task fit, benchmark fit, metric fit, venue\u002Fyear fields, GitHub status, and compute feasibility before the final recommendation.\n- Do not recommend a baseline without a usable GitHub repository.\n- Do not treat a project page, PDF, leaderboard row, or citation count as reproducible code.\n- Do not include empty, placeholder, dead-link, or non-runnable repositories in selected baselines.\n- Do not hide non-reproducible SOTA papers; record them in the excluded table with reasons.\n- Do not compare methods across incompatible datasets, splits, metrics, or training settings without marking the mismatch.\n- Do not claim exhaustive coverage when search evidence is incomplete.\n\n","# Baseline Selector\n\nUse this skill to turn a research idea, task, paper, or experiment plan into a defensible baseline set for experiments and paper submission.\n\nCore rule:\n\n```text\nNo usable GitHub implementation, no selected baseline.\n```\n\nA method may be important, classic, or SOTA, but it cannot enter the recommended baseline set unless its GitHub repository is accessible, non-empty, and plausibly reproducible. Record such methods as excluded candidates with the exclusion reason.\n\n## Modes\n\nDecide the mode from the user request:\n\n1. **Idea to baselines**: user gives a new idea or method sketch.\n2. **Task to SOTA**: user asks for baselines for a task, dataset, or benchmark.\n3. **Paper audit**: user provides a draft or existing baseline list to check for gaps.\n4. **Reviewer-risk check**: user asks what reviewers will say is missing.\n5. **Resource-limited selection**: user gives compute\u002Ftime limits and needs a minimal set.\n6. **Reproduction planning**: user already has candidates and needs runnable code choices.\n\n## Output Profiles\n\nChoose an output profile based on the user's real objective. The same task may produce different baseline recommendations under different profiles.\n\n1. **Paper submission mode**: prioritize completeness, reviewer-expected baselines, and stronger defensive coverage.\n2. **Fast prototype mode**: prioritize the smallest meaningful runnable set for quick signal.\n3. **Limited compute mode**: prioritize budget-compatible baselines and explicitly defer expensive ones.\n4. **Industry comparison mode**: prioritize practical open-source systems and deployment-relevant baselines.\n5. **Rebuttal emergency mode**: prioritize the missing baselines a reviewer is most likely to demand immediately.\n\nIf the user does not specify a profile, infer one from the request and state it explicitly in the report.\n\n## Required Inputs\n\nProceed with available information, but first infer or ask only for missing details that materially affect baseline scope:\n\n```text\nresearch idea \u002F task:\ndomain:\ndataset or benchmark:\nmetric:\ntarget venue and year:\ncompute budget \u002F hardware limit:\ntime budget:\nimplementation constraints:\noutput profile:\n```\n\nTarget venue is required for final recommendation quality. Examples: `ICML 2027`, `NeurIPS 2028`, `CVPR 2027`, `ACL 2026`.\nCompute budget is required for final baseline selection. Ask for GPU type\u002Fcount, expected run time, and whether the user needs full retraining, fine-tuning only, or inference-only baselines.\n\nIf benchmark, metric, or dataset are unknown, state assumptions and mark the baseline set as provisional.\n\n## Response Language Rule\n\nThe language of the final user-facing result should follow the language the user uses as closely as possible. Only proper nouns or standard technical names that are not suitable for translation should remain in English.\n\n- Answer body text in the user's language.\n- Titles, field names, and decision labels should also follow the user's language whenever practical.\n- Proper nouns such as method names, benchmark names, venue names, repository names, and standard technical terms may remain in English when translation would reduce clarity.\n- If the user mixes languages, follow the dominant language unless the user explicitly requests bilingual output.\n- Do not switch to another language for convenience if the user did not ask for it.\n- If a structured report is written to files, the file contents should follow the same rule.\n\n## Main Workflow\n\n0. **Get the current date**\n   - Before searching, obtain the current date from the runtime or environment instead of assuming the model's training cutoff.\n   - Record `as_of_date`, current year, and freshness windows such as last 6 months, last 12 months, and last 24 months.\n   - Use `references\u002Ffreshness-protocol.md` to decide what counts as recent, latest, or classic.\n   - The final report must state the search date and freshness window.\n\n1. **Frame the task**\n   - Convert the user request into a precise comparison problem.\n   - Identify input, output, dataset, metric, training setting, inference setting, and claim type.\n   - Route the task through `references\u002Fdomain-routing.md` to align the search with field-specific benchmark habits and reviewer expectations.\n   - Use `references\u002Fsearch-protocol.md` for search scope and query planning.\n\n2. **Search for candidates**\n   - Collect classic methods, direct competitors, recent SOTA, strong simple methods, benchmark official baselines, and reviewer-expected methods.\n   - Prefer official benchmark pages, survey papers, same-dataset papers, top-conference papers, leaderboards, arXiv, Papers with Code, and GitHub.\n   - Include date-bounded searches for the latest candidates, especially papers and repositories released within the freshness windows.\n   - Record the real publication year and venue for every selected or near-selected method, such as `ICML 2025`, `NeurIPS 2024`, or `CVPR 2026`.\n   - Track search terms and source types.\n\n3. **Apply the GitHub reproducibility gate**\n   - Use `references\u002Fgithub-reproducibility.md`.\n   - A selected baseline must have an accessible GitHub repository, non-empty implementation, runnable instructions, dependency\u002Fenvironment notes, and enough code\u002Fcheckpoints\u002Fconfigs to reproduce or fairly rerun.\n   - Exclude papers with missing code, dead links, empty repositories, placeholder repositories, code-only snippets, non-GitHub code dumps, or repositories that cannot reasonably run the claimed method.\n\n4. **Classify baseline roles**\n   - Use `references\u002Fbaseline-taxonomy.md`.\n   - Assign each candidate one or more roles: classic anchor, direct competitor, current SOTA, simple strong baseline, resource-matched baseline, practical baseline, ablation baseline, or reviewer-expected baseline.\n\n5. **Build evidence tables**\n   - Use `references\u002Fevidence-rules.md`.\n   - Every candidate needs paper source, real publication year, venue, GitHub source, task match, benchmark\u002Fmetric match, code status, reproduction risk, and selection decision.\n   - Never call a method SOTA unless the task, benchmark, metric, date, and source are explicit.\n\n6. **Select baseline sets**\n   - Produce three sets whenever possible:\n     - `minimal`: lowest cost set that still makes the comparison meaningful.\n     - `standard`: expected paper-quality set for the task.\n     - `defensive`: stronger set for top-venue reviewer pressure.\n   - Tailor the recommendation to the user's compute budget, time budget, and target venue\u002Fyear.\n   - Tailor the recommendation to the active output profile.\n   - In `limited compute mode`, explicitly separate `must run`, `nice to run`, and `too expensive for current budget`.\n   - In `rebuttal emergency mode`, prioritize the smallest set of baselines that closes reviewer-visible gaps fastest.\n   - Include only GitHub-reproducible candidates in these sets.\n\n7. **Plan reproduction**\n   - For each selected baseline, list repository URL, install path, expected data preprocessing, command\u002Fconfig clues, checkpoint availability, expected compute, and main failure risks.\n   - Prefer official repositories over unofficial ones unless the official repository is unusable and the unofficial repository is clearly maintained and faithful.\n\n8. **Run reviewer-risk audit**\n   - Use `references\u002Freviewer-risk.md`.\n   - Identify missing classic methods, missing recent competitors, metric mismatch, split mismatch, compute unfairness, no significance testing, and omitted simple baselines.\n   - Separate real risks from risks already handled by excluded non-reproducible methods.\n\n9. **Run self-check**\n   - Use `references\u002Fself-check.md`.\n   - Verify that every selected baseline has the correct year, venue, task match, dataset match, metric match, and GitHub status.\n   - Verify that no selected baseline violates the compute budget or target venue strategy.\n   - Verify that major expected baseline buckets were considered and that any omitted bucket is explicitly justified.\n   - Verify that important non-reproducible or non-matching methods are recorded outside the selected set.\n\n10. **Write final report**\n   - Use `templates\u002Fbaseline-report.md`.\n   - If writing files, prefer `baseline_selection\u002F` with numbered outputs.\n   - If answering in chat, follow the same section order but keep it concise.\n\n## Output Files\n\nWhen the user wants files, use this structure:\n\n```text\nbaseline_selection\u002F\n|- 01_task_definition.md\n|- 02_search_strategy.md\n|- 03_candidate_baselines.md\n|- 04_excluded_nonreproducible.md\n|- 05_recommended_baseline_sets.md\n|- 06_reproduction_plan.md\n|- 07_reviewer_risk_check.md\n|- 08_self_check.md\n`- 09_final_decision.md\n```\n\n## Source Label Sets\n\nChoose one source label set at the start of the response and use it consistently.\n\n### English Label Set\n\nUse this set when the user asks in English:\n\n```text\nDecision Labels\nselected\nselected-if-budget-allows\nselected-if-run-check-passes\nselected-as-harness\nexcluded-no-github\nexcluded-empty-repo\nexcluded-unrunnable-repo\nexcluded-task-mismatch\nexcluded-metric-mismatch\nexcluded-superseded\nwatchlist\nunknown-needs-verification\n\nFinal Verdict Labels\nstrong baseline set\nacceptable baseline set\nrisky baseline set\nincomplete baseline set\n\nSection Titles\nTask Definition\nSearch Strategy\nCandidate Baselines\nGitHub Evidence Chain\nExcluded Non-Reproducible or Non-Selectable Candidates\nRecommended Baseline Sets\nProfile-Aware Recommendation Lens\nMinimal Set\nStandard Set\nDefensive Set\nRecommendation for This User\nReproduction Plan\nReviewer Risk Check\nSelf-Check\nPer-Baseline Verification\nCoverage Check\nFinal Verdict\nMust Run\nNice To Run\n```\n\n### Chinese Label Set\n\nUse this set when the user asks in Chinese:\n\n```text\n判定标签\n入选\n预算允许时入选\n运行检查通过后入选\n作为辅助框架纳入\n因无 GitHub 仓库排除\n因空仓库排除\n因仓库不可运行排除\n因任务不匹配排除\n因指标不匹配排除\n因已被替代排除\n观察名单\n待核验\n\n最终结论标签\n强 baseline 集合\n可接受的 baseline 集合\n风险较高的 baseline 集合\n不完整的 baseline 集合\n\n章节标题\n任务定义\n搜索策略\n候选 Baseline\nGitHub 证据链\n排除项与未纳入候选\n推荐 Baseline 集合\n画像化推荐视角\n最小集合\n标准集合\n防守型集合\n面向当前用户的推荐\n复现计划\n审稿风险检查\n自检\n逐个 Baseline 核验\n覆盖检查\n最终结论\n必须运行\n建议补充\n```\n\nDo not output the English source labels directly in a Chinese response unless the label itself is a proper noun or the user explicitly asks for the English wording.\n\n## Final Verdict\n\nAlways give one final verdict from the active source label set.\n\nExplain the verdict using:\n\n- coverage of classic anchors;\n- coverage of recent SOTA\u002Fdirect competitors;\n- GitHub reproducibility status;\n- resource feasibility;\n- reviewer risk;\n- exact next actions.\n\n## Non-Negotiables\n\n- Always get and record the current date before deciding what counts as latest or recent.\n- Always record the real publication year and venue for every selected baseline.\n- Always require a target venue\u002Fyear and a compute budget before finalizing the recommended set.\n- Always state the active output profile before recommending the final set.\n- Always self-check task fit, benchmark fit, metric fit, venue\u002Fyear fields, GitHub status, and compute feasibility before the final recommendation.\n- Do not recommend a baseline without a usable GitHub repository.\n- Do not treat a project page, PDF, leaderboard row, or citation count as reproducible code.\n- Do not include empty, placeholder, dead-link, or non-runnable repositories in selected baselines.\n- Do not hide non-reproducible SOTA papers; record them in the excluded table with reasons.\n- Do not compare methods across incompatible datasets, splits, metrics, or training settings without marking the mismatch.\n- Do not claim exhaustive coverage when search evidence is incomplete.\n\n",{"name":5,"description":7},"main","e71245365df78e06cd166602744bf2491f79910e","research-skills-hub\u002Fcollected-skills\u002Fbaseline-selector",0,"https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS","https:\u002F\u002Fpengqianhan.github.io\u002FAI-Human-Research-OS\u002F","https:\u002F\u002Fgithub.com\u002Fpengqianhan","Ph.D. candidate of The University of Auckland, New Zealand","2026-11-14 18:08:16",{"source":8,"skillName":5,"commands":35,"prompt":45,"zipAvailable":46},[36,39,42],{"tool":37,"command":38},"npx","npx skills add pengqianhan\u002FAI-Human-Research-OS --skill baseline-selector",{"tool":40,"command":41},"bunx","bunx skills add pengqianhan\u002FAI-Human-Research-OS --skill baseline-selector",{"tool":43,"command":44},"pnpm","pnpm dlx skills add pengqianhan\u002FAI-Human-Research-OS --skill baseline-selector","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fcollected-skills\u002Fbaseline-selector\u002FSKILL.md",true,[48,54,58,62,66,70,74,78,82,86,90,94,98,102,106,110,114,118,122,126,130,134,138,142],{"path":49,"resourceType":50,"resourceTypeDesc":51,"sizeBytes":52,"sourceHtmlUrl":53},"agents\u002Fopenai.yaml","REFERENCE","参考文档",260,"https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fcollected-skills\u002Fbaseline-selector\u002Fagents\u002Fopenai.yaml",{"path":55,"resourceType":50,"resourceTypeDesc":51,"sizeBytes":56,"sourceHtmlUrl":57},"CHANGELOG.md",920,"https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fcollected-skills\u002Fbaseline-selector\u002FCHANGELOG.md",{"path":59,"resourceType":50,"resourceTypeDesc":51,"sizeBytes":60,"sourceHtmlUrl":61},"docs\u002FARCHITECTURE.md",2026,"https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fcollected-skills\u002Fbaseline-selector\u002Fdocs\u002FARCHITECTURE.md",{"path":63,"resourceType":50,"resourceTypeDesc":51,"sizeBytes":64,"sourceHtmlUrl":65},"docs\u002FCOMPATIBILITY.md",1306,"https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fcollected-skills\u002Fbaseline-selector\u002Fdocs\u002FCOMPATIBILITY.md",{"path":67,"resourceType":50,"resourceTypeDesc":51,"sizeBytes":68,"sourceHtmlUrl":69},"docs\u002FDESIGN_PRINCIPLES.md",2066,"https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fcollected-skills\u002Fbaseline-selector\u002Fdocs\u002FDESIGN_PRINCIPLES.md",{"path":71,"resourceType":50,"resourceTypeDesc":51,"sizeBytes":72,"sourceHtmlUrl":73},"docs\u002FINSTALL.md",1953,"https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fcollected-skills\u002Fbaseline-selector\u002Fdocs\u002FINSTALL.md",{"path":75,"resourceType":50,"resourceTypeDesc":51,"sizeBytes":76,"sourceHtmlUrl":77},"docs\u002FROADMAP.md",1190,"https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fcollected-skills\u002Fbaseline-selector\u002Fdocs\u002FROADMAP.md",{"path":79,"resourceType":50,"resourceTypeDesc":51,"sizeBytes":80,"sourceHtmlUrl":81},"evals\u002Fmanual-eval-suite.md",2126,"https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fcollected-skills\u002Fbaseline-selector\u002Fevals\u002Fmanual-eval-suite.md",{"path":83,"resourceType":50,"resourceTypeDesc":51,"sizeBytes":84,"sourceHtmlUrl":85},"examples\u002F3d-detection-nuscenes.md",902,"https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fcollected-skills\u002Fbaseline-selector\u002Fexamples\u002F3d-detection-nuscenes.md",{"path":87,"resourceType":50,"resourceTypeDesc":51,"sizeBytes":88,"sourceHtmlUrl":89},"examples\u002Fllm-agent-memory.md",12479,"https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fcollected-skills\u002Fbaseline-selector\u002Fexamples\u002Fllm-agent-memory.md",{"path":91,"resourceType":91,"resourceTypeDesc":92,"sizeBytes":93,"sourceHtmlUrl":16},"LICENSE","许可证",1056,{"path":95,"resourceType":50,"resourceTypeDesc":51,"sizeBytes":96,"sourceHtmlUrl":97},"manifest.json",1171,"https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fcollected-skills\u002Fbaseline-selector\u002Fmanifest.json",{"path":99,"resourceType":50,"resourceTypeDesc":51,"sizeBytes":100,"sourceHtmlUrl":101},"README.md",6195,"https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fcollected-skills\u002Fbaseline-selector\u002FREADME.md",{"path":103,"resourceType":50,"resourceTypeDesc":51,"sizeBytes":104,"sourceHtmlUrl":105},"references\u002Fbaseline-taxonomy.md",1903,"https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fcollected-skills\u002Fbaseline-selector\u002Freferences\u002Fbaseline-taxonomy.md",{"path":107,"resourceType":50,"resourceTypeDesc":51,"sizeBytes":108,"sourceHtmlUrl":109},"references\u002Fdomain-routing.md",5123,"https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fcollected-skills\u002Fbaseline-selector\u002Freferences\u002Fdomain-routing.md",{"path":111,"resourceType":50,"resourceTypeDesc":51,"sizeBytes":112,"sourceHtmlUrl":113},"references\u002Fevidence-rules.md",1397,"https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fcollected-skills\u002Fbaseline-selector\u002Freferences\u002Fevidence-rules.md",{"path":115,"resourceType":50,"resourceTypeDesc":51,"sizeBytes":116,"sourceHtmlUrl":117},"references\u002Ffreshness-protocol.md",2286,"https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fcollected-skills\u002Fbaseline-selector\u002Freferences\u002Ffreshness-protocol.md",{"path":119,"resourceType":50,"resourceTypeDesc":51,"sizeBytes":120,"sourceHtmlUrl":121},"references\u002Fgithub-reproducibility.md",5951,"https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fcollected-skills\u002Fbaseline-selector\u002Freferences\u002Fgithub-reproducibility.md",{"path":123,"resourceType":50,"resourceTypeDesc":51,"sizeBytes":124,"sourceHtmlUrl":125},"references\u002Foutput-schema.md",2207,"https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fcollected-skills\u002Fbaseline-selector\u002Freferences\u002Foutput-schema.md",{"path":127,"resourceType":50,"resourceTypeDesc":51,"sizeBytes":128,"sourceHtmlUrl":129},"references\u002Freviewer-risk.md",1374,"https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fcollected-skills\u002Fbaseline-selector\u002Freferences\u002Freviewer-risk.md",{"path":131,"resourceType":50,"resourceTypeDesc":51,"sizeBytes":132,"sourceHtmlUrl":133},"references\u002Fsearch-protocol.md",1905,"https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fcollected-skills\u002Fbaseline-selector\u002Freferences\u002Fsearch-protocol.md",{"path":135,"resourceType":50,"resourceTypeDesc":51,"sizeBytes":136,"sourceHtmlUrl":137},"references\u002Fself-check.md",2377,"https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fcollected-skills\u002Fbaseline-selector\u002Freferences\u002Fself-check.md",{"path":139,"resourceType":50,"resourceTypeDesc":51,"sizeBytes":140,"sourceHtmlUrl":141},"templates\u002Fbaseline-report.md",7172,"https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fcollected-skills\u002Fbaseline-selector\u002Ftemplates\u002Fbaseline-report.md",{"path":143,"resourceType":144,"resourceTypeDesc":145,"sizeBytes":146,"sourceHtmlUrl":147},"VERSION","OTHER","其他",6,"https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fcollected-skills\u002Fbaseline-selector\u002FVERSION",[149,156,163,170,177,184],{"skillNo":150,"name":151,"nameZh":6,"description":152,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":153,"categories":154,"keywords":155,"publishedAt":20},"01M74MWYTHT498SVMN823896T7","workflow-skill-creator","Distills a completed user workflow or interaction into a reusable agent skill. Use when the user asks to turn their workflow, interaction, or multi-step process into a skill, or when they say \"make this a skill\", \"create a skill from what we just did\", \"package this workflow\" or similar. Do not use for creating skills from scratch without an existing workflow (use a generic skill-creator for that).","https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fscience-skills\u002Fworkflow_skill_creator\u002FSKILL.md",[],[],{"skillNo":157,"name":158,"nameZh":6,"description":159,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":160,"categories":161,"keywords":162,"publishedAt":20},"01M74MWWVCFZ8HQS8840F1V1FT","uv","Checks whether the uv Python package manager is installed and installs it if missing. Ensures uv is on PATH. Use when another skill requires uv as a prerequisite.","https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fscience-skills\u002Fuv\u002FSKILL.md",[],[],{"skillNo":164,"name":165,"nameZh":6,"description":166,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":167,"categories":168,"keywords":169,"publishedAt":20},"01M74MWRH5BTSWZ2C5PCTD5FV9","uniprot-database","Access protein metadata, function, taxonomy, and sequences across UniProtKB, UniParc, and UniRef. Use when searching for proteins, mapping identifiers, or retrieving functional annotations and publications. Don't use for sequence alignment, protein folding, or sequence similarity search (use specialized skills for those tasks).","https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fscience-skills\u002Funiprot_database\u002FSKILL.md",[],[],{"skillNo":171,"name":172,"nameZh":6,"description":173,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":174,"categories":175,"keywords":176,"publishedAt":20},"01M74MWN1D3ER9F1FZPWH8HQC1","unibind-database","Queries the UniBind database for experimentally validated transcription factor (TF) binding sites. Use when retrieving direct TF-DNA interaction datasets, downloading binding site coordinates (BED\u002FFASTA) for local analysis, or listing available datasets by species, cell line, or TF name. Don't use to query specific intervals, locations, genes, motif models or expression data.","https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fscience-skills\u002Funibind_database\u002FSKILL.md",[],[],{"skillNo":178,"name":179,"nameZh":6,"description":180,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":181,"categories":182,"keywords":183,"publishedAt":20},"01M74MWHHB5QZY3NR2FYA1EK4N","ucsc-conservation-and-tfbs","Fetch Evolutionary Conservation scores (phyloP, phastCons) and Transcription Factor Binding Sites (TFBS) from the UCSC Genome Browser. Use when analyzing whether genomic variants or regions are evolutionarily conserved, functionally important, or bounded by TF regulators across major projects (ENCODE, JASPAR, ReMap).","https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fscience-skills\u002Fucsc_conservation_and_tfbs\u002FSKILL.md",[],[],{"skillNo":185,"name":186,"nameZh":6,"description":187,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":188,"categories":189,"keywords":190,"publishedAt":20},"01M74MWERYHSY87P21FDMVE6BA","string-database","Query the STRING database for protein-protein interactions (PPIs), functional enrichment, and homology. Use when the user asks about interactions between specific proteins, interaction evidence, confidence scores, protein interaction partners, or pathway enrichments.","https:\u002F\u002Fgithub.com\u002Fpengqianhan\u002FAI-Human-Research-OS\u002Fblob\u002Fmain\u002Fresearch-skills-hub\u002Fscience-skills\u002Fstring_database\u002FSKILL.md",[],[]]