[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M77VBN8KD97FTFCA06RKSDB7":3,"skill-same:01M77VBN8KD97FTFCA06RKSDB7":52},{"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":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":24,"commitSha":25,"skillDirectory":26,"forksCount":27,"repositoryHtmlUrl":28,"repositoryHomepageUrl":6,"ownerHtmlUrl":29,"ownerBlogUrl":30,"ownerDescription":6,"updatedAt":31,"install":32,"resources":45},"01M77VBN8KD97FTFCA06RKSDB7","preference-optimization",null,"Align a fine-tuned model with preference data using DPO, ORPO, KTO, or SimPO. Use when preference pairs or thumbs-up\u002Fdown feedback exist, when choosing between preference-optimization methods, or when a DPO run needs hyperparameters or debugging.","Reihantt6\u002Fsns-myagent","Reihantt6","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F219822095?v=4","01M77KK94PV8B7C2D9XWXNS8DJ","Mohamad Reihan tegar prasetya","01M77KK958QFMTHYXGXFV7KZ79",2,"MIT","https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent\u002Fblob\u002Fmain\u002Fagent\u002Fskills\u002Fpreference-optimization\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\ndescription: \"Align a fine-tuned model with preference data using DPO, ORPO, KTO, or SimPO. Use when preference pairs or thumbs-up\u002Fdown feedback exist, when choosing between preference-optimization methods, or when a DPO run needs hyperparameters or debugging.\"\n---\n# Preference Optimization\n\nThis skill assumes `finetuning-method-selection`\nalready routed here because the data shape is\npreference pairs or unpaired thumbs-up\u002Fdown\nfeedback, not demonstrations (that's\n`lora-qlora-recipes`) or a verifiable reward\nsignal (that's `grpo-rlvr-training`). What\nfollows is method selection among the DPO family,\nthe evidence for how much that selection actually\nmatters, the production training pattern, and how\nto build the pairs in the first place.\n\n**Input:** a routing decision (preference\noptimization) plus preference pairs or unpaired\nfeedback, usually from an SFT checkpoint.\n**Output format:** a validated method choice plus\na config — the kwarg values in\n`references\u002Fmethod-configs.md`, not free-form\nadvice — that `llm-finetuning-training-engineer`\nconsumes directly.\n\n## Method Selection\n\n| Data shape | Method | Key parameters |\n|---|---|---|\n| Preference pairs, default case | **DPO** | β=0.1, LR 5e-7–1e-6, 1–2 epochs |\n| Memory-bound or no SFT checkpoint | **ORPO** | reference-free, fused SFT+preference in one loss |\n| Unpaired thumbs-up\u002Fdown | **KTO** | binary label per example, no pairing needed |\n| Length bias observed, sweep budget available | **SimPO** | reference-free; see sweep grid below |\n\n- **DPO is the safe default.** Use β=0.1 and a\n  learning rate of 5e-7 to 1e-6 for 1–2 epochs.\n  This LR is *lower* than the SFT LR that produced\n  the checkpoint being aligned — porting an SFT-\n  scale LR into a DPO run is the most common\n  misconfiguration here, not an edge case.\n- **ORPO** routes in when memory is the\n  constraint, or when there's no separate SFT\n  checkpoint to start from — it's reference-free\n  and fuses the SFT and preference objectives into\n  one loss, skipping the separate SFT pass and the\n  reference-model memory cost DPO carries.\n- **KTO** routes in when feedback is unpaired\n  binary signal (thumbs-up\u002Fdown) rather than\n  matched preference pairs — don't force unpaired\n  feedback into synthetic pairs to use DPO instead.\n- **SimPO** fixes DPO's length bias but only pays\n  off with disciplined sweeping — its published\n  gains are a ceiling reported under a tuned sweep,\n  not a baseline any single config will reproduce.\n  Route here only when there's sweep budget; use\n  DPO instead if there isn't.\n- **Classic RLHF (reward model + PPO) is retired**\n  outside frontier labs. Don't reach for it in a\n  production pipeline — every method above is\n  cheaper and better-supported for the same data\n  shapes.\n\n### Worked Examples\n\n- *\"We have an SFT checkpoint and clean paired\n  preference data, no length-bias complaints yet.\"*\n  → default case → **DPO** at β=0.1.\n- *\"Reviewers click thumbs-up\u002Fdown per response;\n  nothing is paired.\"* → unpaired signal →\n  **KTO**, not DPO — don't synthesize pairs to\n  force DPO onto unpaired data.\n- *\"GPU budget doesn't cover a separate SFT pass\n  plus a DPO reference model.\"* → memory-bound,\n  no separate checkpoint → **ORPO**.\n- *\"DPO output favors longer answers regardless of\n  quality, and there's time to run a sweep.\"* →\n  length bias plus sweep budget → **SimPO**. Skip\n  it if the sweep budget isn't actually there.\n\n## The Low-Leverage Truth\n\nA 2026 240-H100-run study (arXiv 2603.19335) is\nthe load-bearing evidence behind the table above:\n**loss-function choice is worth roughly 1\npercentage point of leverage, model scale is\nworth roughly 50.** Zero of 20 DPO variants tested\nbeat vanilla DPO. Rankings also **invert with\nscale** — a variant that wins in a small pilot can\nlose at deployment size.\n\nTwo practical consequences:\n\n- Don't spend a routing decision agonizing over\n  DPO-variant bake-offs. The table above is\n  sufficient; deeper variant selection is\n  low-leverage compared to data quality and scale.\n- **Validate at deployment scale before trusting a\n  ranking.** A method comparison run on a small\n  pilot model doesn't transfer to the production\n  size class — re-check the winner once scale\n  changes.\n\nThis is also why the Method Selection table above\nis deliberately short: it encodes the ~1pp lever,\nnot a ranking of DPO variants that the same study\nshows doesn't hold up across scale. Treat any\nvariant-selection advice that isn't in that table\n— including advice that claims a specific variant\n\"wins\" — as unproven until it's been validated at\nthe target deployment size.\n\n## Production Pattern: Iterative On-Policy DPO\n\nA single offline DPO pass on a static preference\ndataset is a starting point, not the production\npattern. The policy drifts away from the\ndistribution the pairs were sampled from as\ntraining proceeds, and a static dataset goes stale\nagainst that drift. Production pipelines run DPO\niteratively and on-policy instead:\n\n1. Sample completions from the current policy\n   checkpoint.\n2. Score or rank the completions (reward model,\n   judge, or task grader).\n3. Run a DPO pass using the current checkpoint as\n   the reference model.\n4. The resulting checkpoint becomes both the new\n   policy *and* the new reference for the next\n   round.\n\nRepeat. Each round's reference model is the prior\nround's output, not a fixed initial checkpoint —\nthat's what keeps the preference signal on-policy\ninstead of scoring against an increasingly stale\ndistribution.\n\nA single-pass DPO run is still a reasonable first\niteration — it just isn't the whole pipeline. Plan\nfor at least one more round once the first\ncheckpoint exists, rather than treating pass one\nas the finished artifact.\n\n## Pair Construction\n\nBuild DPO\u002FORPO pairs from **same-task\npassing-vs-failing trajectories** — two attempts\nat the same underlying task, not unrelated\nbest-and-worst examples pulled from different\ntasks. Within that trajectory set, select the\nrejected member at **μ−2σ of the reward\ndistribution, never the minimum**. Naive\nbest-vs-worst pair construction (max reward vs.\nabsolute minimum) degrades as scale increases; the\nμ−2σ selection is more robust to the same scale\nsensitivity the low-leverage study surfaced above.\n\n```\nsorted_by_reward = sort(trajectories, key=reward)\nchosen   = sorted_by_reward[-1]                # highest reward\nmu, sigma = mean(rewards), stdev(rewards)\nrejected = closest(sorted_by_reward, mu - 2 * sigma)\n# NOT sorted_by_reward[0] — the absolute minimum\n# is the naive best-vs-worst construction that\n# degrades as scale increases.\n```\n\nFor the mechanics of turning graded traces into\nthese pairs — including rejection sampling and\njudge-scored delta selection — see\n`trace-to-training-data`.\n\n## References\n\nComplete TRL config blocks per method —\n`DPOConfig`, `ORPOConfig`, `KTOConfig`, and the\nSimPO sweep grid — plus Unsloth wrappers and a\ncatastrophic-forgetting note live in\n`references\u002Fmethod-configs.md`. Those configs use\nthe same current-TRL API conventions established\nin `lora-qlora-recipes`'s\n`references\u002Funsloth-trl-mapping.md`\n(`processing_class`, not `tokenizer=`).\n\n`references\u002Fmethod-configs.md` also carries the\ncatastrophic-forgetting note: a too-high learning\nrate is the usual cause when a preference-tuned\ncheckpoint loses general capability, and the fix\nis almost always to drop the LR toward the low end\nof the range in the Method Selection table above\nbefore reaching for any other remediation.\n\nRelated skills: `finetuning-method-selection`\nroutes here once preference pairs or unpaired\nfeedback exist; `lora-qlora-recipes` produces the\nSFT checkpoint DPO\u002FKTO\u002FSimPO align (ORPO's\nfused path can skip it); `trace-to-training-data`\nconverts passing\u002Ffailing trajectories into the\npairs this skill's Pair Construction section\nconsumes.\n","# Preference Optimization\n\nThis skill assumes `finetuning-method-selection`\nalready routed here because the data shape is\npreference pairs or unpaired thumbs-up\u002Fdown\nfeedback, not demonstrations (that's\n`lora-qlora-recipes`) or a verifiable reward\nsignal (that's `grpo-rlvr-training`). What\nfollows is method selection among the DPO family,\nthe evidence for how much that selection actually\nmatters, the production training pattern, and how\nto build the pairs in the first place.\n\n**Input:** a routing decision (preference\noptimization) plus preference pairs or unpaired\nfeedback, usually from an SFT checkpoint.\n**Output format:** a validated method choice plus\na config — the kwarg values in\n`references\u002Fmethod-configs.md`, not free-form\nadvice — that `llm-finetuning-training-engineer`\nconsumes directly.\n\n## Method Selection\n\n| Data shape | Method | Key parameters |\n|---|---|---|\n| Preference pairs, default case | **DPO** | β=0.1, LR 5e-7–1e-6, 1–2 epochs |\n| Memory-bound or no SFT checkpoint | **ORPO** | reference-free, fused SFT+preference in one loss |\n| Unpaired thumbs-up\u002Fdown | **KTO** | binary label per example, no pairing needed |\n| Length bias observed, sweep budget available | **SimPO** | reference-free; see sweep grid below |\n\n- **DPO is the safe default.** Use β=0.1 and a\n  learning rate of 5e-7 to 1e-6 for 1–2 epochs.\n  This LR is *lower* than the SFT LR that produced\n  the checkpoint being aligned — porting an SFT-\n  scale LR into a DPO run is the most common\n  misconfiguration here, not an edge case.\n- **ORPO** routes in when memory is the\n  constraint, or when there's no separate SFT\n  checkpoint to start from — it's reference-free\n  and fuses the SFT and preference objectives into\n  one loss, skipping the separate SFT pass and the\n  reference-model memory cost DPO carries.\n- **KTO** routes in when feedback is unpaired\n  binary signal (thumbs-up\u002Fdown) rather than\n  matched preference pairs — don't force unpaired\n  feedback into synthetic pairs to use DPO instead.\n- **SimPO** fixes DPO's length bias but only pays\n  off with disciplined sweeping — its published\n  gains are a ceiling reported under a tuned sweep,\n  not a baseline any single config will reproduce.\n  Route here only when there's sweep budget; use\n  DPO instead if there isn't.\n- **Classic RLHF (reward model + PPO) is retired**\n  outside frontier labs. Don't reach for it in a\n  production pipeline — every method above is\n  cheaper and better-supported for the same data\n  shapes.\n\n### Worked Examples\n\n- *\"We have an SFT checkpoint and clean paired\n  preference data, no length-bias complaints yet.\"*\n  → default case → **DPO** at β=0.1.\n- *\"Reviewers click thumbs-up\u002Fdown per response;\n  nothing is paired.\"* → unpaired signal →\n  **KTO**, not DPO — don't synthesize pairs to\n  force DPO onto unpaired data.\n- *\"GPU budget doesn't cover a separate SFT pass\n  plus a DPO reference model.\"* → memory-bound,\n  no separate checkpoint → **ORPO**.\n- *\"DPO output favors longer answers regardless of\n  quality, and there's time to run a sweep.\"* →\n  length bias plus sweep budget → **SimPO**. Skip\n  it if the sweep budget isn't actually there.\n\n## The Low-Leverage Truth\n\nA 2026 240-H100-run study (arXiv 2603.19335) is\nthe load-bearing evidence behind the table above:\n**loss-function choice is worth roughly 1\npercentage point of leverage, model scale is\nworth roughly 50.** Zero of 20 DPO variants tested\nbeat vanilla DPO. Rankings also **invert with\nscale** — a variant that wins in a small pilot can\nlose at deployment size.\n\nTwo practical consequences:\n\n- Don't spend a routing decision agonizing over\n  DPO-variant bake-offs. The table above is\n  sufficient; deeper variant selection is\n  low-leverage compared to data quality and scale.\n- **Validate at deployment scale before trusting a\n  ranking.** A method comparison run on a small\n  pilot model doesn't transfer to the production\n  size class — re-check the winner once scale\n  changes.\n\nThis is also why the Method Selection table above\nis deliberately short: it encodes the ~1pp lever,\nnot a ranking of DPO variants that the same study\nshows doesn't hold up across scale. Treat any\nvariant-selection advice that isn't in that table\n— including advice that claims a specific variant\n\"wins\" — as unproven until it's been validated at\nthe target deployment size.\n\n## Production Pattern: Iterative On-Policy DPO\n\nA single offline DPO pass on a static preference\ndataset is a starting point, not the production\npattern. The policy drifts away from the\ndistribution the pairs were sampled from as\ntraining proceeds, and a static dataset goes stale\nagainst that drift. Production pipelines run DPO\niteratively and on-policy instead:\n\n1. Sample completions from the current policy\n   checkpoint.\n2. Score or rank the completions (reward model,\n   judge, or task grader).\n3. Run a DPO pass using the current checkpoint as\n   the reference model.\n4. The resulting checkpoint becomes both the new\n   policy *and* the new reference for the next\n   round.\n\nRepeat. Each round's reference model is the prior\nround's output, not a fixed initial checkpoint —\nthat's what keeps the preference signal on-policy\ninstead of scoring against an increasingly stale\ndistribution.\n\nA single-pass DPO run is still a reasonable first\niteration — it just isn't the whole pipeline. Plan\nfor at least one more round once the first\ncheckpoint exists, rather than treating pass one\nas the finished artifact.\n\n## Pair Construction\n\nBuild DPO\u002FORPO pairs from **same-task\npassing-vs-failing trajectories** — two attempts\nat the same underlying task, not unrelated\nbest-and-worst examples pulled from different\ntasks. Within that trajectory set, select the\nrejected member at **μ−2σ of the reward\ndistribution, never the minimum**. Naive\nbest-vs-worst pair construction (max reward vs.\nabsolute minimum) degrades as scale increases; the\nμ−2σ selection is more robust to the same scale\nsensitivity the low-leverage study surfaced above.\n\n```\nsorted_by_reward = sort(trajectories, key=reward)\nchosen   = sorted_by_reward[-1]                # highest reward\nmu, sigma = mean(rewards), stdev(rewards)\nrejected = closest(sorted_by_reward, mu - 2 * sigma)\n# NOT sorted_by_reward[0] — the absolute minimum\n# is the naive best-vs-worst construction that\n# degrades as scale increases.\n```\n\nFor the mechanics of turning graded traces into\nthese pairs — including rejection sampling and\njudge-scored delta selection — see\n`trace-to-training-data`.\n\n## References\n\nComplete TRL config blocks per method —\n`DPOConfig`, `ORPOConfig`, `KTOConfig`, and the\nSimPO sweep grid — plus Unsloth wrappers and a\ncatastrophic-forgetting note live in\n`references\u002Fmethod-configs.md`. Those configs use\nthe same current-TRL API conventions established\nin `lora-qlora-recipes`'s\n`references\u002Funsloth-trl-mapping.md`\n(`processing_class`, not `tokenizer=`).\n\n`references\u002Fmethod-configs.md` also carries the\ncatastrophic-forgetting note: a too-high learning\nrate is the usual cause when a preference-tuned\ncheckpoint loses general capability, and the fix\nis almost always to drop the LR toward the low end\nof the range in the Method Selection table above\nbefore reaching for any other remediation.\n\nRelated skills: `finetuning-method-selection`\nroutes here once preference pairs or unpaired\nfeedback exist; `lora-qlora-recipes` produces the\nSFT checkpoint DPO\u002FKTO\u002FSimPO align (ORPO's\nfused path can skip it); `trace-to-training-data`\nconverts passing\u002Ffailing trajectories into the\npairs this skill's Pair Construction section\nconsumes.\n",{"description":7},"main","58843b159d7a4050d6b785a98b908cce505c0b92","agent\u002Fskills\u002Fpreference-optimization",0,"https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent","https:\u002F\u002Fgithub.com\u002FReihantt6","reihsns.web.id","2026-08-19 20:08:18",{"source":8,"skillName":5,"commands":33,"prompt":43,"zipAvailable":44},[34,37,40],{"tool":35,"command":36},"npx","npx skills add Reihantt6\u002Fsns-myagent --skill preference-optimization",{"tool":38,"command":39},"bunx","bunx skills add Reihantt6\u002Fsns-myagent --skill preference-optimization",{"tool":41,"command":42},"pnpm","pnpm dlx skills add Reihantt6\u002Fsns-myagent --skill preference-optimization","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent\u002Fblob\u002Fmain\u002Fagent\u002Fskills\u002Fpreference-optimization\u002FSKILL.md",true,[46],{"path":47,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":50,"sourceHtmlUrl":51},"references\u002Fmethod-configs.md","REFERENCE","参考文档",6602,"https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent\u002Fblob\u002Fmain\u002Fagent\u002Fskills\u002Fpreference-optimization\u002Freferences\u002Fmethod-configs.md",[53,60,67,74,81,88],{"skillNo":54,"name":55,"nameZh":6,"description":56,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":57,"categories":58,"keywords":59,"publishedAt":19},"01M77XDCYZ407XYMF2T4P6F7VT","youtube-transcript","Extract transcripts from YouTube videos. Use when the user asks for a transcript, subtitles, or captions of a YouTube video and provides a YouTube URL (youtube.com\u002Fwatch?v=, youtu.be\u002F, or similar). Supports output with or without timestamps.","https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent\u002Fblob\u002Fmain\u002Fagent\u002Fskills\u002Fyoutube-transcript\u002FSKILL.md",[],[],{"skillNo":61,"name":62,"nameZh":6,"description":63,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":64,"categories":65,"keywords":66,"publishedAt":19},"01M77XD7PBC98J6MXGJJ2TJ8DX","youtube-thumbnail-design","YouTube thumbnail design with specific dimensions, contrast rules, and mobile preview optimization. Covers safe zones, text placement, face expression psychology, and A\u002FB testing. Use for: YouTube thumbnails, video cover images, click-through optimization. Triggers: youtube thumbnail, thumbnail design, video thumbnail, click through rate, ctr optimization, youtube cover, video cover image, thumbnail maker, thumbnail tips, youtube design, video preview image","https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent\u002Fblob\u002Fmain\u002Fagent\u002Fskills\u002Fyoutube-thumbnail-design\u002FSKILL.md",[],[],{"skillNo":68,"name":69,"nameZh":6,"description":70,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":71,"categories":72,"keywords":73,"publishedAt":19},"01M77XD2541C8HXJ6RHYZP4NGJ","youtube-short-maker","Make YouTube Shorts that hold attention with Pexo — vertical 9:16, fast-paced, hook in the first 3 seconds. Describe your topic and Pexo writes the script, generates the shots, picks the models, and assembles a finished Short with music and captions. Use for YouTube Shorts: \"youtube short\", \"make a youtube short\", \"shorts video\", \"short-form video\". NOT for long-form or landscape video.","https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent\u002Fblob\u002Fmain\u002Fagent\u002Fskills\u002Fyoutube-short-maker\u002FSKILL.md",[],[],{"skillNo":75,"name":76,"nameZh":6,"description":77,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":78,"categories":79,"keywords":80,"publishedAt":19},"01M77XCXKVSYK7KTZ3PQBVVGFD","xcode-project-setup","Safely modifies Xcode projects (.pbxproj) to add Swift Packages and link files. Use this skill whenever an iOS project needs dependencies installed (e.g. Firebase, Alamofire).","https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent\u002Fblob\u002Fmain\u002Fagent\u002Fskills\u002Fxcode-project-setup\u002FSKILL.md",[],[],{"skillNo":82,"name":83,"nameZh":6,"description":84,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":85,"categories":86,"keywords":87,"publishedAt":19},"01M77XCR1DV8ZQK2NHWKVY4EWX","x402","Set up Browser Use Cloud payments with x402 — pay per request from a crypto wallet (USDC on Base mainnet), no signup or API key. Two setups it works out up front — \"just use it\" (set up a wallet so you or Claude Code can run cloud browser tasks paid from the wallet — Claude writes and runs throwaway scripts, nothing touches your codebase) or \"build it in\" (install the SDK and write the key + code into your project). Walks through wallet setup, funding, .env, and a ~$1 test run. Use when the user asks about x402, pay-per-use, USDC payments, or wants Browser Use Cloud without an API key. For the free-tier signup (reverse-CAPTCHA → API key), use `browser-use cloud signup` or the `cloud` skill instead.","https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent\u002Fblob\u002Fmain\u002Fagent\u002Fskills\u002Fx402\u002FSKILL.md",[],[],{"skillNo":89,"name":90,"nameZh":6,"description":91,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":92,"categories":93,"keywords":94,"publishedAt":19},"01M77XCMD3R3ZBZQQSW6E0793B","x-twitter-scraper","Build GitHub Copilot workflows with Xquik X API SDKs, REST endpoints, hosted Apify Actor runs, MCP tools, TweetClaw OpenClaw plugin installs, signed webhooks, tweet search, user lookup, follower exports, media actions, and agent automation.","https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent\u002Fblob\u002Fmain\u002Fagent\u002Fskills\u002Fx-twitter-scraper\u002FSKILL.md",[],[]]