[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M77WEX3WSXEC2V5WCGDP95D0":3,"skill-same:01M77WEX3WSXEC2V5WCGDP95D0":58},{"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},"01M77WEX3WSXEC2V5WCGDP95D0","spark-memory-thermal-ops",null,"Manage unified memory and thermals during long-running ML jobs on NVIDIA DGX Spark. Use when planning memory headroom for a training run on GB10, when a job OOMs on unified memory, or when monitoring temperature and power during multi-hour training.","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\u002Fspark-memory-thermal-ops\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\ndescription: \"Manage unified memory and thermals during long-running ML jobs on NVIDIA DGX Spark. Use when planning memory headroom for a training run on GB10, when a job OOMs on unified memory, or when monitoring temperature and power during multi-hour training.\"\n---\n# Spark Memory & Thermal Ops\n\nDGX Spark's GB10 chip has one 128GB unified\nmemory (UMA) pool shared by CPU and GPU, and a\nsustained power ceiling well below its rated\nfigure. Both break discrete-GPU assumptions:\nheadroom isn't what `nvidia-smi` reports, and a\nrun that starts fast will slow down mid-job\nwith nothing misconfigured. This skill covers\nplanning memory headroom, working an actual\nOOM, and watching thermals across a long job.\nFor launch-time failure modes (ABI mismatches,\nflash-attn, playbook breakage), see\n`spark-training-gotchas` — this skill assumes\nthe job starts.\n\n## Common Issues Quick Reference\n\n| Situation | Do this |\n|---|---|\n| Planning headroom before launch | Budget against `free -g`, not `nvidia-smi` — see UMA Memory Model |\n| Job OOMs on unified memory | Work the OOM Ladder in order: flush, then batch\u002Fpack, then method downgrade |\n| Throughput drops mid-run | Check the power\u002Ftemp log before assuming a config bug — see Thermal Monitoring |\n| Trainer + inference server both wanted | Run one at a time — see Concurrent Workloads |\n\n## When to Use This Skill\n\n- Sizing a training run against the 128GB pool\n  before launch — will this model, method, and\n  batch\u002Fpack combination fit.\n- A run OOMs mid-load or mid-step and the\n  remediation order matters — what to try first,\n  second, third.\n- Watching temperature and power during a\n  multi-hour job, deciding whether a slowdown is\n  thermal throttling or something else.\n- Planning to run a trainer alongside an\n  inference server (vLLM, Ollama) on the same box.\n\n## UMA Memory Model\n\nSpark has no separate GPU VRAM — the GPU and\nCPU share one 128GB pool. Two consequences:\n\n- **`nvidia-smi` and `cudaMemGetInfo`\n  underreport pressure — or report nothing at\n  all.** Both report CUDA-allocator-visible\n  memory, not the pool's actual state — a box can\n  show headroom in `nvidia-smi` and still OOM,\n  because page-cache and mmap'd pages the\n  allocator doesn't see consume the same pool. On\n  some driver\u002Fsetups, the memory query returns\n  `[N\u002FA], [N\u002FA]` outright instead of a number — a\n  script grepping for a numeric value there gets\n  nothing, not a misleading undercount (see\n  `spark-training-gotchas` gotcha G3).\n\n- **Model load is a transient peak, not the\n  steady state.** Loading safetensors weights\n  mmaps the file, then copies into CUDA\n  tensors — for a window during load, both the\n  mmap'd pages and the CUDA copy count against\n  the pool at once. A model that fits while\n  training can still OOM during load if headroom\n  was sized for the post-load footprint instead\n  of this doubled transient.\n\nPlan and diagnose with `free -g`, not\n`nvidia-smi`:\n\n```bash\nfree -g | awk 'NR==2 {print \"free:\", $4, \"GB\"}'\n```\n\nRule of thumb: take that free figure, subtract a\nfew GB for OS\u002Fdriver overhead, and budget against\nthe result — not the 128GB spec number.\nThe worksheet in `references\u002Fuma-accounting.md`\naccepts parameter count, dtype, and method as\ninput, and returns a memory estimate to compare\nagainst known anchors.\n\n### Planning Sequence\n\nBefore launch, work through these in order:\n\n1. Read `free -g`; subtract OS\u002Fdriver overhead\n   for the budget.\n2. Estimate weights + optimizer + gradients +\n   activations from `references\u002Fuma-accounting.md`.\n3. Compare against the closest anchor (70B\n   QLoRA, 27B LoRA, 9B full FT), not the\n   estimate alone.\n4. If the estimate is close to the budget, start\n   with shorter packing or a smaller batch —\n   cheaper than hitting the OOM Ladder mid-run.\n\n### Example: Sizing a 70B QLoRA Run\n\nA sanity check of the worksheet formula against\nthe ≈40GB anchor:\n\n```python\nparams = 70e9\nweights_gb = params * 0.5 \u002F 1e9      # NF4, step 1\nadapter_gb = 0.5                     # step 5, negligible\ntotal_gb = weights_gb + adapter_gb   # + activations\nprint(f\"{total_gb:.0f}GB before activations\")\n```\n\nWeights alone land near the ≈40GB anchor — a plan\nestimating far above that for the same model\nclass is a signal to recheck dtype and method.\n\n## The OOM Ladder\n\nWhen a job OOMs on unified memory, work this\nladder in order. Each step is more disruptive\nthan the last — don't skip ahead:\n**reducing batch size is never step 1.**\n\n1. **Flush the buffer cache.** Page cache from a\n   previous run or a large dataset read often\n   accounts for GB of the \"missing\" headroom.\n   This costs nothing but a rerun and doesn't\n   touch the job's configuration:\n\n   ```bash\n   sync; echo 3 > \u002Fproc\u002Fsys\u002Fvm\u002Fdrop_caches\n   ```\n\n   Needs root; a between-run reset, not a\n   mid-training step. See\n   `spark-training-gotchas` (gotcha G3) for the\n   full diagnostic behind this step.\n\n2. **Reduce batch size or packing length.** Only\n   after a flush fails to free enough headroom,\n   cut batch size or packing length — the first\n   step that changes what the run does. Prefer\n   packing length first; it drives activation\n   footprint more directly at long context.\n\n3. **Downgrade the method: bf16 LoRA before\n   QLoRA.** If flushing and shrinking batch\u002Fpack\n   still OOM, drop the method a tier — bf16 LoRA\n   is next, not the reverse. QLoRA's bitsandbytes\n   dequantization buffers are transient CUDA-side\n   allocations that can OOM before an equivalent\n   bf16 LoRA run would, even though QLoRA's\n   steady-state footprint is smaller. A QLoRA OOM\n   is not proof the model doesn't fit.\n\nFall back further (smaller model, multi-Spark)\nonly after all three steps and the job still\nwon't fit.\n\n## Thermal Monitoring\n\nMulti-hour runs push into Spark's sustained\npower ceiling, well under the rated figure —\nexpected platform behavior, not a symptom to\nexplain away:\n\n- Sample temperature and power alongside the\n  training logs, not after a slowdown is\n  noticed — every 30-60 seconds correlates a\n  throughput drop with a thermal event. Keep\n  the CSV output format `assets\u002Fthermal-sample.sh`\n  writes, so timestamps line up against the log:\n\n  ```bash\n  bash assets\u002Fthermal-sample.sh 30 thermal.log\n  ```\n\n- **A sustained ~100W power draw is the platform\n  cap, not a configuration bug.** Don't re-tune\n  batch size or precision to \"fix\" a plateau\n  that's the box behaving normally under load.\n  If temperature climbs while power stays flat\n  under the rated 240W figure, that's the\n  signature to recognize.\n\n- Log throttle events explicitly instead of\n  letting a run silently slow down unrecorded. A\n  run whose per-step time doubles two hours in\n  should show that in the log, correlated against\n  the thermal sample at that timestamp. Full\n  throttling diagnostics: `spark-training-gotchas`\n  (gotcha G4).\n\n## Concurrent Workloads\n\nBecause the 128GB pool is global, eviction\nhappens without either process's logs showing\nan OOM:\n\n- The one-heavy-job rule applies to **uncapped or\n  near-capacity** workloads — an uncapped trainer\n  and inference server (vLLM, Ollama) compete for\n  the same pool. A small, capped workload doesn't:\n  a \u003C4GB LoRA fine-tune coexists fine alongside\n  vLLM capped at `gpu-memory-utilization\u003C=0.5` —\n  check the other process's cap, not just its\n  presence, before stopping it.\n\n- Inference servers evict trainer pages silently\n  under uncapped\u002Fnear-capacity contention, and\n  vice versa — neither logs an error, so a slow\n  run or lost KV cache is a contention symptom to\n  check for. Stop unrelated *uncapped* servers\n  before a long or full-pool run.\n\nCheck for GPU-resident processes first:\n\n```bash\nps aux | grep -E 'vllm|ollama|trl|axolotl' | grep -v grep\n```\n\nThis procedure complements `spark-training-gotchas`\n(gotchas G3, G4, G6) — that skill covers launch-time\nfailures; this one, the running job.\n\nMemory math worksheets:\n`references\u002Fuma-accounting.md`.\n","# Spark Memory & Thermal Ops\n\nDGX Spark's GB10 chip has one 128GB unified\nmemory (UMA) pool shared by CPU and GPU, and a\nsustained power ceiling well below its rated\nfigure. Both break discrete-GPU assumptions:\nheadroom isn't what `nvidia-smi` reports, and a\nrun that starts fast will slow down mid-job\nwith nothing misconfigured. This skill covers\nplanning memory headroom, working an actual\nOOM, and watching thermals across a long job.\nFor launch-time failure modes (ABI mismatches,\nflash-attn, playbook breakage), see\n`spark-training-gotchas` — this skill assumes\nthe job starts.\n\n## Common Issues Quick Reference\n\n| Situation | Do this |\n|---|---|\n| Planning headroom before launch | Budget against `free -g`, not `nvidia-smi` — see UMA Memory Model |\n| Job OOMs on unified memory | Work the OOM Ladder in order: flush, then batch\u002Fpack, then method downgrade |\n| Throughput drops mid-run | Check the power\u002Ftemp log before assuming a config bug — see Thermal Monitoring |\n| Trainer + inference server both wanted | Run one at a time — see Concurrent Workloads |\n\n## When to Use This Skill\n\n- Sizing a training run against the 128GB pool\n  before launch — will this model, method, and\n  batch\u002Fpack combination fit.\n- A run OOMs mid-load or mid-step and the\n  remediation order matters — what to try first,\n  second, third.\n- Watching temperature and power during a\n  multi-hour job, deciding whether a slowdown is\n  thermal throttling or something else.\n- Planning to run a trainer alongside an\n  inference server (vLLM, Ollama) on the same box.\n\n## UMA Memory Model\n\nSpark has no separate GPU VRAM — the GPU and\nCPU share one 128GB pool. Two consequences:\n\n- **`nvidia-smi` and `cudaMemGetInfo`\n  underreport pressure — or report nothing at\n  all.** Both report CUDA-allocator-visible\n  memory, not the pool's actual state — a box can\n  show headroom in `nvidia-smi` and still OOM,\n  because page-cache and mmap'd pages the\n  allocator doesn't see consume the same pool. On\n  some driver\u002Fsetups, the memory query returns\n  `[N\u002FA], [N\u002FA]` outright instead of a number — a\n  script grepping for a numeric value there gets\n  nothing, not a misleading undercount (see\n  `spark-training-gotchas` gotcha G3).\n\n- **Model load is a transient peak, not the\n  steady state.** Loading safetensors weights\n  mmaps the file, then copies into CUDA\n  tensors — for a window during load, both the\n  mmap'd pages and the CUDA copy count against\n  the pool at once. A model that fits while\n  training can still OOM during load if headroom\n  was sized for the post-load footprint instead\n  of this doubled transient.\n\nPlan and diagnose with `free -g`, not\n`nvidia-smi`:\n\n```bash\nfree -g | awk 'NR==2 {print \"free:\", $4, \"GB\"}'\n```\n\nRule of thumb: take that free figure, subtract a\nfew GB for OS\u002Fdriver overhead, and budget against\nthe result — not the 128GB spec number.\nThe worksheet in `references\u002Fuma-accounting.md`\naccepts parameter count, dtype, and method as\ninput, and returns a memory estimate to compare\nagainst known anchors.\n\n### Planning Sequence\n\nBefore launch, work through these in order:\n\n1. Read `free -g`; subtract OS\u002Fdriver overhead\n   for the budget.\n2. Estimate weights + optimizer + gradients +\n   activations from `references\u002Fuma-accounting.md`.\n3. Compare against the closest anchor (70B\n   QLoRA, 27B LoRA, 9B full FT), not the\n   estimate alone.\n4. If the estimate is close to the budget, start\n   with shorter packing or a smaller batch —\n   cheaper than hitting the OOM Ladder mid-run.\n\n### Example: Sizing a 70B QLoRA Run\n\nA sanity check of the worksheet formula against\nthe ≈40GB anchor:\n\n```python\nparams = 70e9\nweights_gb = params * 0.5 \u002F 1e9      # NF4, step 1\nadapter_gb = 0.5                     # step 5, negligible\ntotal_gb = weights_gb + adapter_gb   # + activations\nprint(f\"{total_gb:.0f}GB before activations\")\n```\n\nWeights alone land near the ≈40GB anchor — a plan\nestimating far above that for the same model\nclass is a signal to recheck dtype and method.\n\n## The OOM Ladder\n\nWhen a job OOMs on unified memory, work this\nladder in order. Each step is more disruptive\nthan the last — don't skip ahead:\n**reducing batch size is never step 1.**\n\n1. **Flush the buffer cache.** Page cache from a\n   previous run or a large dataset read often\n   accounts for GB of the \"missing\" headroom.\n   This costs nothing but a rerun and doesn't\n   touch the job's configuration:\n\n   ```bash\n   sync; echo 3 > \u002Fproc\u002Fsys\u002Fvm\u002Fdrop_caches\n   ```\n\n   Needs root; a between-run reset, not a\n   mid-training step. See\n   `spark-training-gotchas` (gotcha G3) for the\n   full diagnostic behind this step.\n\n2. **Reduce batch size or packing length.** Only\n   after a flush fails to free enough headroom,\n   cut batch size or packing length — the first\n   step that changes what the run does. Prefer\n   packing length first; it drives activation\n   footprint more directly at long context.\n\n3. **Downgrade the method: bf16 LoRA before\n   QLoRA.** If flushing and shrinking batch\u002Fpack\n   still OOM, drop the method a tier — bf16 LoRA\n   is next, not the reverse. QLoRA's bitsandbytes\n   dequantization buffers are transient CUDA-side\n   allocations that can OOM before an equivalent\n   bf16 LoRA run would, even though QLoRA's\n   steady-state footprint is smaller. A QLoRA OOM\n   is not proof the model doesn't fit.\n\nFall back further (smaller model, multi-Spark)\nonly after all three steps and the job still\nwon't fit.\n\n## Thermal Monitoring\n\nMulti-hour runs push into Spark's sustained\npower ceiling, well under the rated figure —\nexpected platform behavior, not a symptom to\nexplain away:\n\n- Sample temperature and power alongside the\n  training logs, not after a slowdown is\n  noticed — every 30-60 seconds correlates a\n  throughput drop with a thermal event. Keep\n  the CSV output format `assets\u002Fthermal-sample.sh`\n  writes, so timestamps line up against the log:\n\n  ```bash\n  bash assets\u002Fthermal-sample.sh 30 thermal.log\n  ```\n\n- **A sustained ~100W power draw is the platform\n  cap, not a configuration bug.** Don't re-tune\n  batch size or precision to \"fix\" a plateau\n  that's the box behaving normally under load.\n  If temperature climbs while power stays flat\n  under the rated 240W figure, that's the\n  signature to recognize.\n\n- Log throttle events explicitly instead of\n  letting a run silently slow down unrecorded. A\n  run whose per-step time doubles two hours in\n  should show that in the log, correlated against\n  the thermal sample at that timestamp. Full\n  throttling diagnostics: `spark-training-gotchas`\n  (gotcha G4).\n\n## Concurrent Workloads\n\nBecause the 128GB pool is global, eviction\nhappens without either process's logs showing\nan OOM:\n\n- The one-heavy-job rule applies to **uncapped or\n  near-capacity** workloads — an uncapped trainer\n  and inference server (vLLM, Ollama) compete for\n  the same pool. A small, capped workload doesn't:\n  a \u003C4GB LoRA fine-tune coexists fine alongside\n  vLLM capped at `gpu-memory-utilization\u003C=0.5` —\n  check the other process's cap, not just its\n  presence, before stopping it.\n\n- Inference servers evict trainer pages silently\n  under uncapped\u002Fnear-capacity contention, and\n  vice versa — neither logs an error, so a slow\n  run or lost KV cache is a contention symptom to\n  check for. Stop unrelated *uncapped* servers\n  before a long or full-pool run.\n\nCheck for GPU-resident processes first:\n\n```bash\nps aux | grep -E 'vllm|ollama|trl|axolotl' | grep -v grep\n```\n\nThis procedure complements `spark-training-gotchas`\n(gotchas G3, G4, G6) — that skill covers launch-time\nfailures; this one, the running job.\n\nMemory math worksheets:\n`references\u002Fuma-accounting.md`.\n",{"description":7},"main","58843b159d7a4050d6b785a98b908cce505c0b92","agent\u002Fskills\u002Fspark-memory-thermal-ops",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 spark-memory-thermal-ops",{"tool":38,"command":39},"bunx","bunx skills add Reihantt6\u002Fsns-myagent --skill spark-memory-thermal-ops",{"tool":41,"command":42},"pnpm","pnpm dlx skills add Reihantt6\u002Fsns-myagent --skill spark-memory-thermal-ops","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent\u002Fblob\u002Fmain\u002Fagent\u002Fskills\u002Fspark-memory-thermal-ops\u002FSKILL.md",true,[46,52],{"path":47,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":50,"sourceHtmlUrl":51},"assets\u002Fthermal-sample.sh","SCRIPT","脚本",1017,"https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent\u002Fblob\u002Fmain\u002Fagent\u002Fskills\u002Fspark-memory-thermal-ops\u002Fassets\u002Fthermal-sample.sh",{"path":53,"resourceType":54,"resourceTypeDesc":55,"sizeBytes":56,"sourceHtmlUrl":57},"references\u002Fuma-accounting.md","REFERENCE","参考文档",4407,"https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent\u002Fblob\u002Fmain\u002Fagent\u002Fskills\u002Fspark-memory-thermal-ops\u002Freferences\u002Fuma-accounting.md",[59,66,73,80,87,94],{"skillNo":60,"name":61,"nameZh":6,"description":62,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":63,"categories":64,"keywords":65,"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":67,"name":68,"nameZh":6,"description":69,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":70,"categories":71,"keywords":72,"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":74,"name":75,"nameZh":6,"description":76,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":77,"categories":78,"keywords":79,"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":81,"name":82,"nameZh":6,"description":83,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":84,"categories":85,"keywords":86,"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":88,"name":89,"nameZh":6,"description":90,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":91,"categories":92,"keywords":93,"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":95,"name":96,"nameZh":6,"description":97,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":98,"categories":99,"keywords":100,"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",[],[]]