[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M77X5KYXRBWR4FJ2TS2ZE8X3":3,"skill-same:01M77X5KYXRBWR4FJ2TS2ZE8X3":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},"01M77X5KYXRBWR4FJ2TS2ZE8X3","vision-sft",null,"Fine-tune vision-language models (VLMs) with supervised learning on image+text data. Use when adapting a VLM to a visual domain or task, configuring frozen-vision-tower LoRA, or debugging a VLM fine-tune that trains without learning.","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\u002Fvision-sft\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\ndescription: \"Fine-tune vision-language models (VLMs) with supervised learning on image+text data. Use when adapting a VLM to a visual domain or task, configuring frozen-vision-tower LoRA, or debugging a VLM fine-tune that trains without learning.\"\n---\n# Vision-Language SFT\n\nThis skill assumes `finetuning-method-selection`\nalready routed here: the data shape is\nimage+text demonstrations, not preference pairs\nor a verifiable reward signal, and the base is a\nvision-language model rather than a text-only\none. `lora-qlora-recipes` covers the text-only\nLoRA\u002FQLoRA recipe this skill specializes for the\nvision tower and projector; read that skill first\nif the LoRA fundamentals (rank, alpha, target\nmodules) aren't already familiar.\n\n**Input:** an image+text dataset and a VLM base\nmodel already picked from the model catalog.\n**Output format:** a validated adapter config —\nwhich components are frozen, LoRA target modules,\nand a `min_pixels`\u002F`max_pixels` budget — that\n`llm-finetuning-training-engineer` consumes\ndirectly when it generates a runnable script.\n\n## Quick Reference\n\n| Situation | Default |\n|---|---|\n| Adapting behavior on familiar images | Frozen tower+projector, LoRA r=8–16, α=16–32 |\n| Visual domain shift | Unfreeze last-6 ViT layers, vision LR 5–10x lower |\n| Doesn't fit in bf16 at target rank | QLoRA — frozen vision tower only |\n| `fast_inference=True` | `finetune_vision_layers=False` |\n| Loss normal, eval not improving | Check the Two Silent Killers below first |\n\n## The Consensus Recipe\n\nFreeze the vision tower and the projector. Put\nLoRA on the LLM only, all-linear (the same\nattention + MLP target list as text-only SFT —\nsee `lora-qlora-recipes`), at **r=8–16,\nα=16–32**. This is the settled default for\nadapting a VLM's behavior without disturbing how\nit sees.\n\n- **The vision tower and projector stay frozen by\n  default.** They already encode a general visual\n  representation; retraining them is rarely\n  necessary and adds risk without adding\n  capability for most tasks.\n- **LoRA rank runs lower than the text-only\n  general default** (r=8–16 here vs r=16–32 for\n  text-only SFT) because the LLM-only adapter is\n  adapting behavior, not injecting new visual\n  knowledge.\n- **QLoRA is permitted only with a frozen vision\n  tower.** Quantizing the base while also\n  unfreezing and training vision layers is\n  unsupported and unstable — treat this as a hard\n  pairing rule, not a tunable. If the vision tower\n  needs to unfreeze, drop QLoRA and use bf16 LoRA\n  instead.\n\n```python\n# freeze tower + projector; LoRA on LLM only\nfor name, param in model.named_parameters():\n    if \"vision_tower\" in name or \"projector\" in name:\n        param.requires_grad = False\n\ntarget_modules = [\n    \"q_proj\", \"k_proj\", \"v_proj\", \"o_proj\",\n    \"gate_proj\", \"up_proj\", \"down_proj\",\n]  # LLM-only, all-linear — r=8-16, alpha=16-32\n```\n\n## When to Unfreeze\n\nUnfreezing vision layers is a deliberate\nescalation, not a default decision — reach\nfor it only when the domain shift is\nvisual, not textual.\n\n- **Unfreeze only for visual domain\n  shift.** If the task is teaching new\n  behavior on images the tower already\n  understands (charts, everyday photos),\n  the frozen-tower recipe above is\n  sufficient. Unfreeze when the visual\n  domain itself is unfamiliar to the\n  tower — satellite imagery, medical\n  scans, dense technical diagrams — and\n  the frozen-tower recipe plateaus.\n- **Last-6 ViT layers is the sweet\n  spot.** Unfreezing the final six\n  vision-transformer layers (not the\n  whole tower) measured **+1.7pt DocVQA\n  at ~1.75x training cost** over the\n  frozen baseline. Treat six layers as\n  the ceiling worth paying for; going\n  further spends compute without a\n  matched result.\n- **Vision LR must run 5–10x lower than\n  the LLM LR when unfrozen.** The vision\n  tower's pretrained representation is\n  more fragile than the LLM's adapter;\n  the same LR for both risks overwriting\n  the visual representation faster than\n  the LLM adapter can compensate.\n- **High LoRA rank on the patch-\n  embedding layer risks NaN.** If patch\n  embedding is in the unfrozen set, keep\n  its rank low and watch early-step loss\n  closely — one of the most fragile\n  places to apply LoRA in a VLM.\n\n## The Two Silent Killers\n\nBoth produce a run that trains without error and\nwithout learning: the loss curve looks normal,\nthe model doesn't improve, and neither throws an\nexception — both need an explicit pre-training\ncheck, not just a clean training log.\n\n- **Image-tag\u002Fcount mismatch.** Every image\n  placeholder token in the templated text must\n  map 1:1 to a media item actually passed to the\n  collator. A mismatch (one placeholder, zero or\n  two images attached; or an image with no\n  placeholder) doesn't error in most collators —\n  it silently misaligns image and text, and the\n  model \"trains but learns nothing.\" Validate the\n  1:1 placeholder-to-media mapping before training\n  starts, on every example, not just a sample.\n  Full validation-checklist detail:\n  `references\u002Fcollators-and-pitfalls.md`.\n- **`min_pixels`\u002F`max_pixels` resolution budget.**\n  This pair is the single most consequential\n  hyperparameter for quality and memory in VLM\n  SFT — more than rank, alpha, or LR. Too low\n  silently downsamples images below what the task\n  needs (small document text becomes unreadable\n  even though training \"succeeds\"); too high blows\n  the activation memory budget or forces too small\n  a batch to train stably. Set it deliberately per\n  dataset, don't leave it at a framework default.\n\n## Unsloth Specifics\n\n- **`UnslothVisionDataCollator`** is the collator\n  Unsloth expects for VLM SFT — it handles the\n  image-tag alignment and per-architecture\n  processor contract described in\n  `references\u002Fcollators-and-pitfalls.md`. Don't\n  substitute a text-only collator for VLM data.\n- **`finetune_vision_layers=False` is required\n  when `fast_inference=True`.** vLLM cannot serve\n  LoRA adapters on vision layers, so a fast-\n  inference setup that also unfreezes vision\n  layers fails at serve time even if training\n  succeeds. If the recipe calls for unfreezing the\n  last-6 ViT layers (see When to Unfreeze above),\n  fast inference is off the table for that run —\n  choose one or the other, not both.\n\n## Model Choice\n\nBase VLM choice is out of scope for this skill —\nit lives in one place, the model catalog at\n`finetuning-method-selection`'s\n`references\u002Fmodel-catalog.md`. This skill and its\nreferences describe recipes by architecture\nfamily only, never by recommending one model over\nanother.\n\nVLM reinforcement learning (VLM-GRPO) is\nreference-only in this plugin — the fragmented\ntooling and reward-hacking failure modes specific\nto VLM-RL are covered in `grpo-rlvr-training`,\nnot here. This skill's scope stops at supervised\nfine-tuning.\n\n## Failure Modes\n\nThe recurring mistake across every section above\nis treating a clean loss curve as proof the run\nis healthy. A normal-looking curve is consistent\nwith **both** a working run **and** either silent\nkiller, since the model trains on *something*\neither way — just not the aligned image-text\nsignal when a killer is present. A flat eval score\nnext to a normal loss curve means re-run the\nchecklist in `references\u002Fcollators-and-pitfalls.md`\nbefore touching any hyperparameter.\n\n## References\n\n- `references\u002Fcollators-and-pitfalls.md` — per-\n  architecture collator table, dataset-format\n  examples with image placeholders, a pre-\n  training validation checklist, and the two-\n  stage projector-alignment recipe as an advanced\n  pattern.\n\nRelated skills: `finetuning-method-selection`\nroutes here; `lora-qlora-recipes` covers the\ntext-only LoRA fundamentals this skill\nspecializes; `grpo-rlvr-training` covers VLM-RL\n(reference-only); `dataset-curation` covers\nimage+text dataset preparation this skill doesn't.\n","# Vision-Language SFT\n\nThis skill assumes `finetuning-method-selection`\nalready routed here: the data shape is\nimage+text demonstrations, not preference pairs\nor a verifiable reward signal, and the base is a\nvision-language model rather than a text-only\none. `lora-qlora-recipes` covers the text-only\nLoRA\u002FQLoRA recipe this skill specializes for the\nvision tower and projector; read that skill first\nif the LoRA fundamentals (rank, alpha, target\nmodules) aren't already familiar.\n\n**Input:** an image+text dataset and a VLM base\nmodel already picked from the model catalog.\n**Output format:** a validated adapter config —\nwhich components are frozen, LoRA target modules,\nand a `min_pixels`\u002F`max_pixels` budget — that\n`llm-finetuning-training-engineer` consumes\ndirectly when it generates a runnable script.\n\n## Quick Reference\n\n| Situation | Default |\n|---|---|\n| Adapting behavior on familiar images | Frozen tower+projector, LoRA r=8–16, α=16–32 |\n| Visual domain shift | Unfreeze last-6 ViT layers, vision LR 5–10x lower |\n| Doesn't fit in bf16 at target rank | QLoRA — frozen vision tower only |\n| `fast_inference=True` | `finetune_vision_layers=False` |\n| Loss normal, eval not improving | Check the Two Silent Killers below first |\n\n## The Consensus Recipe\n\nFreeze the vision tower and the projector. Put\nLoRA on the LLM only, all-linear (the same\nattention + MLP target list as text-only SFT —\nsee `lora-qlora-recipes`), at **r=8–16,\nα=16–32**. This is the settled default for\nadapting a VLM's behavior without disturbing how\nit sees.\n\n- **The vision tower and projector stay frozen by\n  default.** They already encode a general visual\n  representation; retraining them is rarely\n  necessary and adds risk without adding\n  capability for most tasks.\n- **LoRA rank runs lower than the text-only\n  general default** (r=8–16 here vs r=16–32 for\n  text-only SFT) because the LLM-only adapter is\n  adapting behavior, not injecting new visual\n  knowledge.\n- **QLoRA is permitted only with a frozen vision\n  tower.** Quantizing the base while also\n  unfreezing and training vision layers is\n  unsupported and unstable — treat this as a hard\n  pairing rule, not a tunable. If the vision tower\n  needs to unfreeze, drop QLoRA and use bf16 LoRA\n  instead.\n\n```python\n# freeze tower + projector; LoRA on LLM only\nfor name, param in model.named_parameters():\n    if \"vision_tower\" in name or \"projector\" in name:\n        param.requires_grad = False\n\ntarget_modules = [\n    \"q_proj\", \"k_proj\", \"v_proj\", \"o_proj\",\n    \"gate_proj\", \"up_proj\", \"down_proj\",\n]  # LLM-only, all-linear — r=8-16, alpha=16-32\n```\n\n## When to Unfreeze\n\nUnfreezing vision layers is a deliberate\nescalation, not a default decision — reach\nfor it only when the domain shift is\nvisual, not textual.\n\n- **Unfreeze only for visual domain\n  shift.** If the task is teaching new\n  behavior on images the tower already\n  understands (charts, everyday photos),\n  the frozen-tower recipe above is\n  sufficient. Unfreeze when the visual\n  domain itself is unfamiliar to the\n  tower — satellite imagery, medical\n  scans, dense technical diagrams — and\n  the frozen-tower recipe plateaus.\n- **Last-6 ViT layers is the sweet\n  spot.** Unfreezing the final six\n  vision-transformer layers (not the\n  whole tower) measured **+1.7pt DocVQA\n  at ~1.75x training cost** over the\n  frozen baseline. Treat six layers as\n  the ceiling worth paying for; going\n  further spends compute without a\n  matched result.\n- **Vision LR must run 5–10x lower than\n  the LLM LR when unfrozen.** The vision\n  tower's pretrained representation is\n  more fragile than the LLM's adapter;\n  the same LR for both risks overwriting\n  the visual representation faster than\n  the LLM adapter can compensate.\n- **High LoRA rank on the patch-\n  embedding layer risks NaN.** If patch\n  embedding is in the unfrozen set, keep\n  its rank low and watch early-step loss\n  closely — one of the most fragile\n  places to apply LoRA in a VLM.\n\n## The Two Silent Killers\n\nBoth produce a run that trains without error and\nwithout learning: the loss curve looks normal,\nthe model doesn't improve, and neither throws an\nexception — both need an explicit pre-training\ncheck, not just a clean training log.\n\n- **Image-tag\u002Fcount mismatch.** Every image\n  placeholder token in the templated text must\n  map 1:1 to a media item actually passed to the\n  collator. A mismatch (one placeholder, zero or\n  two images attached; or an image with no\n  placeholder) doesn't error in most collators —\n  it silently misaligns image and text, and the\n  model \"trains but learns nothing.\" Validate the\n  1:1 placeholder-to-media mapping before training\n  starts, on every example, not just a sample.\n  Full validation-checklist detail:\n  `references\u002Fcollators-and-pitfalls.md`.\n- **`min_pixels`\u002F`max_pixels` resolution budget.**\n  This pair is the single most consequential\n  hyperparameter for quality and memory in VLM\n  SFT — more than rank, alpha, or LR. Too low\n  silently downsamples images below what the task\n  needs (small document text becomes unreadable\n  even though training \"succeeds\"); too high blows\n  the activation memory budget or forces too small\n  a batch to train stably. Set it deliberately per\n  dataset, don't leave it at a framework default.\n\n## Unsloth Specifics\n\n- **`UnslothVisionDataCollator`** is the collator\n  Unsloth expects for VLM SFT — it handles the\n  image-tag alignment and per-architecture\n  processor contract described in\n  `references\u002Fcollators-and-pitfalls.md`. Don't\n  substitute a text-only collator for VLM data.\n- **`finetune_vision_layers=False` is required\n  when `fast_inference=True`.** vLLM cannot serve\n  LoRA adapters on vision layers, so a fast-\n  inference setup that also unfreezes vision\n  layers fails at serve time even if training\n  succeeds. If the recipe calls for unfreezing the\n  last-6 ViT layers (see When to Unfreeze above),\n  fast inference is off the table for that run —\n  choose one or the other, not both.\n\n## Model Choice\n\nBase VLM choice is out of scope for this skill —\nit lives in one place, the model catalog at\n`finetuning-method-selection`'s\n`references\u002Fmodel-catalog.md`. This skill and its\nreferences describe recipes by architecture\nfamily only, never by recommending one model over\nanother.\n\nVLM reinforcement learning (VLM-GRPO) is\nreference-only in this plugin — the fragmented\ntooling and reward-hacking failure modes specific\nto VLM-RL are covered in `grpo-rlvr-training`,\nnot here. This skill's scope stops at supervised\nfine-tuning.\n\n## Failure Modes\n\nThe recurring mistake across every section above\nis treating a clean loss curve as proof the run\nis healthy. A normal-looking curve is consistent\nwith **both** a working run **and** either silent\nkiller, since the model trains on *something*\neither way — just not the aligned image-text\nsignal when a killer is present. A flat eval score\nnext to a normal loss curve means re-run the\nchecklist in `references\u002Fcollators-and-pitfalls.md`\nbefore touching any hyperparameter.\n\n## References\n\n- `references\u002Fcollators-and-pitfalls.md` — per-\n  architecture collator table, dataset-format\n  examples with image placeholders, a pre-\n  training validation checklist, and the two-\n  stage projector-alignment recipe as an advanced\n  pattern.\n\nRelated skills: `finetuning-method-selection`\nroutes here; `lora-qlora-recipes` covers the\ntext-only LoRA fundamentals this skill\nspecializes; `grpo-rlvr-training` covers VLM-RL\n(reference-only); `dataset-curation` covers\nimage+text dataset preparation this skill doesn't.\n",{"description":7},"main","58843b159d7a4050d6b785a98b908cce505c0b92","agent\u002Fskills\u002Fvision-sft",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 vision-sft",{"tool":38,"command":39},"bunx","bunx skills add Reihantt6\u002Fsns-myagent --skill vision-sft",{"tool":41,"command":42},"pnpm","pnpm dlx skills add Reihantt6\u002Fsns-myagent --skill vision-sft","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent\u002Fblob\u002Fmain\u002Fagent\u002Fskills\u002Fvision-sft\u002FSKILL.md",true,[46],{"path":47,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":50,"sourceHtmlUrl":51},"references\u002Fcollators-and-pitfalls.md","REFERENCE","参考文档",5969,"https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent\u002Fblob\u002Fmain\u002Fagent\u002Fskills\u002Fvision-sft\u002Freferences\u002Fcollators-and-pitfalls.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",[],[]]