[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M77VQAK9FH7APDWW8J7PPK98":3,"skill-same:01M77VQAK9FH7APDWW8J7PPK98":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},"01M77VQAK9FH7APDWW8J7PPK98","quantized-export",null,"Export a promoted fine-tuned model in the right deployment format — merged safetensors, LoRA-only, GGUF with imatrix, or FP8. Use after a checkpoint passes promotion, when choosing a quantization format for a target device, or when an exported model fails its smoke test.","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\u002Fquantized-export\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\ndescription: \"Export a promoted fine-tuned model in the right deployment format — merged safetensors, LoRA-only, GGUF with imatrix, or FP8. Use after a checkpoint passes promotion, when choosing a quantization format for a target device, or when an exported model fails its smoke test.\"\n---\n# Quantized Export\n\nThe last stop after `checkpoint-promotion`\nhands off a `PROMOTE` verdict: a checkpoint\nthat cleared the four-stage gate still isn't\ndeployed until it's exported in the right\nformat for its target runtime and proven to\nstill work post-export. A `REJECT` verdict\nnever reaches this skill — export starts only\nfrom a promoted checkpoint.\n\n**Input:** a promoted checkpoint (or LoRA\nadapter) plus the target deployment surface —\nGPU class, serving stack, and whether\nlong-context\u002Fcode\u002Fmath workloads are in\nscope.\n**Output format:** an exported artifact in\nthe chosen format plus a smoke-test diff\nreport comparing 3–5 golden outputs\npre-export and post-export.\n\n## Format Map\n\nPick format by hardware and deployment shape,\nnot by habit — the wrong pick either wastes\nthroughput headroom or breaks silently on\nspecific workloads (see Workload Overrides).\n\n- **FP8 is the default on Hopper-class GPUs\n  and newer.** It preserves near-bf16 quality\n  at roughly half the memory, and it's the\n  safe first choice whenever the target GPU\n  supports it and no edge-device constraint\n  applies.\n- **AWQ INT4 targets older GPUs** that predate\n  FP8 hardware support. **GPTQ is superseded\n  for new deployments** — don't reach for it\n  on a fresh export; AWQ has better accuracy\n  retention at the same bit width and wider\n  current tooling support.\n- **GGUF with Q4_K_M quantization, built from\n  an imatrix, is the edge\u002Fllama.cpp format.**\n  Use it for local or CPU-adjacent\n  deployment, not for GPU-serving\n  throughput — it optimizes for footprint,\n  not tokens\u002Fsec on a datacenter GPU.\n- **NVFP4 is for Blackwell-at-scale\n  deployments only — and explicitly NOT on\n  GB10.** NVFP4 on SM121 (GB10) runs **~32%\n  slower than FP8** because the hardware\n  lacks a native `cvt.e2m1x2` path unless the\n  kernel is compiled `sm_121a`. Choosing\n  NVFP4 on a GB10 target is a regression, not\n  an upgrade — pick FP8 there instead.\n- **Merged vs. LoRA-only is a separate axis\n  from quant format.** A merged export folds\n  the adapter into the base weights: larger\n  artifact, no base-model dependency at serve\n  time. LoRA-only keeps the adapter separate:\n  much smaller artifact, but the serving stack\n  must load the exact same base model\n  alongside it — a mismatched or\n  wrong-revision base silently changes\n  outputs. Pick merged when artifact\n  portability matters more than storage; pick\n  LoRA-only when disk footprint or multi-adapter\n  serving matters more.\n\n### Worked Picks\n\nThe core format-selection tradeoff, read as a\nlookup table for common scenarios:\n\n| Target | Workload | Format |\n|---|---|---|\n| Datacenter GPU | generic chat | FP8 |\n| Datacenter GPU | long-context\u002Fcode\u002Fmath | FP8 or W8A8 — never INT4 |\n| Older GPU generation | generic | AWQ INT4 |\n| Edge device \u002F laptop | llama.cpp serving | GGUF Q4_K_M + imatrix |\n| GB10 | any workload | FP8 via vLLM nightly, or GGUF via llama.cpp locally — skip NVFP4 |\n\n```yaml\n# quick decision snippet — see the table above for the full map\nhopper_or_newer: fp8\nolder_gpu: awq-int4\nedge_llama_cpp: gguf-q4_k_m+imatrix\ngb10_any_workload: fp8-vllm-nightly   # never nvfp4 on GB10\n```\n\n## Workload Overrides\n\nThe Format Map above is a default, not a rule\nthat survives every workload. **Long-context,\ncode, and math workloads break at INT4** —\nquantization error compounds across long\nsequences and precise token-level reasoning in\nways that don't show up on short, generic\nprompts. For any of these three workload\nclasses, **stay on FP8 or W8A8** even if the\ntarget hardware would otherwise justify INT4\non cost grounds.\n\n- Don't validate this override with MMLU or\n  similar broad-knowledge benchmarks — they\n  don't stress the failure mode. **Measure\n  with the actual task evals** — the goldens\n  and graders from `eval-harness-first`, run\n  through the exported artifact — because\n  INT4 degradation on long-context, code, or\n  math shows up as task-specific failures\n  (dropped context, broken syntax, arithmetic\n  errors) well before it moves a knowledge\n  benchmark.\n- If a task eval regresses after an INT4\n  export on one of these three workload\n  classes, the fix is switching format, not\n  re-tuning the quantization recipe — AWQ\n  and GPTQ variants at the same bit width\n  share the same compounding-error failure\n  mode on these workloads.\n\n## The Smoke Test\n\nExport bugs are silent at the file level — a\nmalformed export still produces a\nloadable artifact, so file-existence checks\nprove nothing. **The smoke test is\nmandatory for every export, with no\nexception for a format that \"should just\nwork\":**\n\n1. **Load the exported artifact in its actual\n   target runtime** — vLLM for FP8\u002FAWQ,\n   llama.cpp for GGUF, not a quick\n   sanity load in a different framework than\n   the one that will serve it in production.\n2. **Run 3–5 golden prompts through it** —\n   pull these from the same `eval\u002Fgoldens.jsonl`\n   `eval-harness-first` maintains, not a fresh\n   ad hoc set.\n3. **Compare each output against the\n   pre-export generation** for the same\n   prompt, same deterministic sampling\n   settings — greedy decoding (temperature 0)\n   and a fixed seed, persisted and reused\n   between the pre- and post-export runs, not\n   just nominally identical config. **For a\n   lossless export, byte match is the gate —\n   any diff is a bug.** For a **lossy**\n   (quantized) export, byte match is expected\n   to fail; the gate is task-grader verdict\n   agreement instead — see\n   `references\u002Fexport-commands.md`'s\n   Smoke-Test Script Skeleton.\n\nRun this as a gate, not a manual check:\n\n```bash\npython smoke_test.py \"$EXPORT_PATH\" \\\n    eval\u002Fgoldens.jsonl pre-export-outputs.jsonl\n# non-zero exit on any pre\u002Fpost mismatch\n```\n\n### Failure Signatures\n\nWhat export bugs actually look like, not a\nclean pass\u002Ffail flag:\n\n- **Template mismatch** presents as garbled or\n  run-on output — the chat template baked\n  into the export doesn't match the one the\n  checkpoint was trained and evaluated\n  against, so turn boundaries or special\n  tokens land in the wrong place.\n- **Wrong quantization applied to `lm_head`**\n  presents as off-template or semantically\n  nonsensical output that still looks\n  fluent — the output head lost precision it\n  needed even though the rest of the network\n  quantized cleanly.\n\nNever ship an export that skipped this step —\na checkpoint's `PROMOTE` verdict says the\nun-exported checkpoint is good; it says\nnothing about the export pipeline. Re-run on\nany quant-method or runtime version bump, not\nonly after the first export. Runnable command\nsequences for every format plus the\nsmoke-test script skeleton:\n`references\u002Fexport-commands.md`.\n\n## Related Skills\n\n- `checkpoint-promotion` — the only valid\n  upstream source for this skill. A checkpoint\n  without a `PROMOTE` verdict doesn't reach\n  export.\n- `eval-harness-first` — owns the\n  `eval\u002Fgoldens.jsonl` this skill's smoke test\n  draws its 3–5 prompts from, and the task\n  evals the Workload Overrides section\n  requires for long-context\u002Fcode\u002Fmath\n  validation.\n- `finetuning-method-selection` — its\n  `references\u002Fmodel-catalog.md` is the place\n  to check hardware-class assumptions (which\n  GPU generations a base model targets) before\n  picking a format off the Format Map above.\n\n**Spark users:** on GB10, GGUF via llama.cpp\nworks well for local serving, and FP8 serving\nvia vLLM nightly builds is the other proven\npath — NVFP4 is the one format to avoid there\n(see the Format Map exception above). Once the\n`dgx-spark-ops` plugin is installed, defer\nSpark-specific serving and thermal questions to\nits skills rather than re-deriving them here.\n","# Quantized Export\n\nThe last stop after `checkpoint-promotion`\nhands off a `PROMOTE` verdict: a checkpoint\nthat cleared the four-stage gate still isn't\ndeployed until it's exported in the right\nformat for its target runtime and proven to\nstill work post-export. A `REJECT` verdict\nnever reaches this skill — export starts only\nfrom a promoted checkpoint.\n\n**Input:** a promoted checkpoint (or LoRA\nadapter) plus the target deployment surface —\nGPU class, serving stack, and whether\nlong-context\u002Fcode\u002Fmath workloads are in\nscope.\n**Output format:** an exported artifact in\nthe chosen format plus a smoke-test diff\nreport comparing 3–5 golden outputs\npre-export and post-export.\n\n## Format Map\n\nPick format by hardware and deployment shape,\nnot by habit — the wrong pick either wastes\nthroughput headroom or breaks silently on\nspecific workloads (see Workload Overrides).\n\n- **FP8 is the default on Hopper-class GPUs\n  and newer.** It preserves near-bf16 quality\n  at roughly half the memory, and it's the\n  safe first choice whenever the target GPU\n  supports it and no edge-device constraint\n  applies.\n- **AWQ INT4 targets older GPUs** that predate\n  FP8 hardware support. **GPTQ is superseded\n  for new deployments** — don't reach for it\n  on a fresh export; AWQ has better accuracy\n  retention at the same bit width and wider\n  current tooling support.\n- **GGUF with Q4_K_M quantization, built from\n  an imatrix, is the edge\u002Fllama.cpp format.**\n  Use it for local or CPU-adjacent\n  deployment, not for GPU-serving\n  throughput — it optimizes for footprint,\n  not tokens\u002Fsec on a datacenter GPU.\n- **NVFP4 is for Blackwell-at-scale\n  deployments only — and explicitly NOT on\n  GB10.** NVFP4 on SM121 (GB10) runs **~32%\n  slower than FP8** because the hardware\n  lacks a native `cvt.e2m1x2` path unless the\n  kernel is compiled `sm_121a`. Choosing\n  NVFP4 on a GB10 target is a regression, not\n  an upgrade — pick FP8 there instead.\n- **Merged vs. LoRA-only is a separate axis\n  from quant format.** A merged export folds\n  the adapter into the base weights: larger\n  artifact, no base-model dependency at serve\n  time. LoRA-only keeps the adapter separate:\n  much smaller artifact, but the serving stack\n  must load the exact same base model\n  alongside it — a mismatched or\n  wrong-revision base silently changes\n  outputs. Pick merged when artifact\n  portability matters more than storage; pick\n  LoRA-only when disk footprint or multi-adapter\n  serving matters more.\n\n### Worked Picks\n\nThe core format-selection tradeoff, read as a\nlookup table for common scenarios:\n\n| Target | Workload | Format |\n|---|---|---|\n| Datacenter GPU | generic chat | FP8 |\n| Datacenter GPU | long-context\u002Fcode\u002Fmath | FP8 or W8A8 — never INT4 |\n| Older GPU generation | generic | AWQ INT4 |\n| Edge device \u002F laptop | llama.cpp serving | GGUF Q4_K_M + imatrix |\n| GB10 | any workload | FP8 via vLLM nightly, or GGUF via llama.cpp locally — skip NVFP4 |\n\n```yaml\n# quick decision snippet — see the table above for the full map\nhopper_or_newer: fp8\nolder_gpu: awq-int4\nedge_llama_cpp: gguf-q4_k_m+imatrix\ngb10_any_workload: fp8-vllm-nightly   # never nvfp4 on GB10\n```\n\n## Workload Overrides\n\nThe Format Map above is a default, not a rule\nthat survives every workload. **Long-context,\ncode, and math workloads break at INT4** —\nquantization error compounds across long\nsequences and precise token-level reasoning in\nways that don't show up on short, generic\nprompts. For any of these three workload\nclasses, **stay on FP8 or W8A8** even if the\ntarget hardware would otherwise justify INT4\non cost grounds.\n\n- Don't validate this override with MMLU or\n  similar broad-knowledge benchmarks — they\n  don't stress the failure mode. **Measure\n  with the actual task evals** — the goldens\n  and graders from `eval-harness-first`, run\n  through the exported artifact — because\n  INT4 degradation on long-context, code, or\n  math shows up as task-specific failures\n  (dropped context, broken syntax, arithmetic\n  errors) well before it moves a knowledge\n  benchmark.\n- If a task eval regresses after an INT4\n  export on one of these three workload\n  classes, the fix is switching format, not\n  re-tuning the quantization recipe — AWQ\n  and GPTQ variants at the same bit width\n  share the same compounding-error failure\n  mode on these workloads.\n\n## The Smoke Test\n\nExport bugs are silent at the file level — a\nmalformed export still produces a\nloadable artifact, so file-existence checks\nprove nothing. **The smoke test is\nmandatory for every export, with no\nexception for a format that \"should just\nwork\":**\n\n1. **Load the exported artifact in its actual\n   target runtime** — vLLM for FP8\u002FAWQ,\n   llama.cpp for GGUF, not a quick\n   sanity load in a different framework than\n   the one that will serve it in production.\n2. **Run 3–5 golden prompts through it** —\n   pull these from the same `eval\u002Fgoldens.jsonl`\n   `eval-harness-first` maintains, not a fresh\n   ad hoc set.\n3. **Compare each output against the\n   pre-export generation** for the same\n   prompt, same deterministic sampling\n   settings — greedy decoding (temperature 0)\n   and a fixed seed, persisted and reused\n   between the pre- and post-export runs, not\n   just nominally identical config. **For a\n   lossless export, byte match is the gate —\n   any diff is a bug.** For a **lossy**\n   (quantized) export, byte match is expected\n   to fail; the gate is task-grader verdict\n   agreement instead — see\n   `references\u002Fexport-commands.md`'s\n   Smoke-Test Script Skeleton.\n\nRun this as a gate, not a manual check:\n\n```bash\npython smoke_test.py \"$EXPORT_PATH\" \\\n    eval\u002Fgoldens.jsonl pre-export-outputs.jsonl\n# non-zero exit on any pre\u002Fpost mismatch\n```\n\n### Failure Signatures\n\nWhat export bugs actually look like, not a\nclean pass\u002Ffail flag:\n\n- **Template mismatch** presents as garbled or\n  run-on output — the chat template baked\n  into the export doesn't match the one the\n  checkpoint was trained and evaluated\n  against, so turn boundaries or special\n  tokens land in the wrong place.\n- **Wrong quantization applied to `lm_head`**\n  presents as off-template or semantically\n  nonsensical output that still looks\n  fluent — the output head lost precision it\n  needed even though the rest of the network\n  quantized cleanly.\n\nNever ship an export that skipped this step —\na checkpoint's `PROMOTE` verdict says the\nun-exported checkpoint is good; it says\nnothing about the export pipeline. Re-run on\nany quant-method or runtime version bump, not\nonly after the first export. Runnable command\nsequences for every format plus the\nsmoke-test script skeleton:\n`references\u002Fexport-commands.md`.\n\n## Related Skills\n\n- `checkpoint-promotion` — the only valid\n  upstream source for this skill. A checkpoint\n  without a `PROMOTE` verdict doesn't reach\n  export.\n- `eval-harness-first` — owns the\n  `eval\u002Fgoldens.jsonl` this skill's smoke test\n  draws its 3–5 prompts from, and the task\n  evals the Workload Overrides section\n  requires for long-context\u002Fcode\u002Fmath\n  validation.\n- `finetuning-method-selection` — its\n  `references\u002Fmodel-catalog.md` is the place\n  to check hardware-class assumptions (which\n  GPU generations a base model targets) before\n  picking a format off the Format Map above.\n\n**Spark users:** on GB10, GGUF via llama.cpp\nworks well for local serving, and FP8 serving\nvia vLLM nightly builds is the other proven\npath — NVFP4 is the one format to avoid there\n(see the Format Map exception above). Once the\n`dgx-spark-ops` plugin is installed, defer\nSpark-specific serving and thermal questions to\nits skills rather than re-deriving them here.\n",{"description":7},"main","58843b159d7a4050d6b785a98b908cce505c0b92","agent\u002Fskills\u002Fquantized-export",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 quantized-export",{"tool":38,"command":39},"bunx","bunx skills add Reihantt6\u002Fsns-myagent --skill quantized-export",{"tool":41,"command":42},"pnpm","pnpm dlx skills add Reihantt6\u002Fsns-myagent --skill quantized-export","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent\u002Fblob\u002Fmain\u002Fagent\u002Fskills\u002Fquantized-export\u002FSKILL.md",true,[46],{"path":47,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":50,"sourceHtmlUrl":51},"references\u002Fexport-commands.md","REFERENCE","参考文档",12736,"https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent\u002Fblob\u002Fmain\u002Fagent\u002Fskills\u002Fquantized-export\u002Freferences\u002Fexport-commands.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",[],[]]