[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M77ME8VTDS2QEBBZ0JXHG7F5":3,"skill-same:01M77ME8VTDS2QEBBZ0JXHG7F5":56},{"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},"01M77ME8VTDS2QEBBZ0JXHG7F5","finetuning-method-selection",null,"Decide whether to fine-tune at all, and route to the right method (SFT, DPO\u002FORPO\u002FKTO, GRPO\u002FRLVR, continued pretraining) and base model. Use when starting any fine-tuning effort, when unsure whether RAG or prompting would suffice, or when choosing between preference-optimization and reinforcement methods.","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\u002F.agents\u002Fskills\u002Ffinetuning-method-selection\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: finetuning-method-selection\ndescription: Decide whether to fine-tune at all, and route to the right method (SFT, DPO\u002FORPO\u002FKTO, GRPO\u002FRLVR, continued pretraining) and base model. Use when starting any fine-tuning effort, when unsure whether RAG or prompting would suffice, or when choosing between preference-optimization and reinforcement methods.\n---\n\n# Fine-Tuning Method Selection\n\nThis is the router skill for the fine-tuning\nlifecycle: it decides whether fine-tuning is the\nright tool at all, and if so, which method and\nwhich base-model size class. Every other skill\nin this plugin assumes this routing already\nhappened — start here before opening\n`lora-qlora-recipes`, `preference-optimization`,\nor `grpo-rlvr-training`.\n\n## When to Use This Skill\n\n- Starting any fine-tuning effort, before a\n  framework or base model has been chosen.\n- Unsure whether RAG or prompt engineering would\n  solve the problem more cheaply than training.\n- Choosing between preference optimization (DPO\n  family) and a reinforcement method (GRPO\u002FRLVR)\n  for the same underlying task.\n- Sizing a candidate model\u002Fmethod combination\n  before committing to a run.\n\n## Quick Reference\n\n| Situation | Route |\n|---|---|\n| Facts change often (prices, docs, news) | RAG, not fine-tuning |\n| Desired behavior still being figured out | Prompt engineering |\n| Stable domain knowledge, ≥500MB text | CPT then SFT — see Off-Ramps First |\n| Have input\u002Foutput demonstrations | SFT — see `lora-qlora-recipes` |\n| Have preference pairs or thumbs-up\u002Fdown | DPO\u002FORPO\u002FKTO — see `preference-optimization` |\n| Have a verifiable pass\u002Ffail signal | GRPO+RLVR — see `grpo-rlvr-training` |\n| No eval harness yet | Stop — see `eval-harness-first` |\n\n## Off-Ramps First\n\nMost requests that sound like \"fine-tune this\"\nare served better and cheaper elsewhere. Check\nthese off-ramps before opening a training run:\n\n- **Knowledge-bound and volatile** (the gap is\n  facts that change — prices, docs, current\n  events): route to RAG, not fine-tuning. A\n  fine-tuned model bakes in a snapshot; volatile\n  facts go stale immediately.\n- **Behavior-bound and shifting** (the desired\n  behavior is still being figured out, or\n  changes per request): route to prompt\n  engineering. Fine-tuning locks in a behavior;\n  don't lock in one that hasn't stabilized yet.\n- **Stable, dense domain knowledge**: this is\n  where continued pretraining (CPT) enters, sized\n  by how much domain text exists:\n\n| Domain text volume | Route |\n|---|---|\n| \u003C10MB | RAG only |\n| 10MB–500MB | RAG + fine-tune |\n| 500MB–10GB | CPT, then SFT |\n| >10GB | CPT required |\n\nCPT learning rate ≈ **10% of the pretraining\nLR**. CPT is guidance-only in this plugin —\nsizing and LR guidance live here, but this\nplugin does not execute a CPT run.\n\n## Method Router\n\nOnce the off-ramps are ruled out, this is the\nfull decision tree (verbatim from the research\nthis plugin is built on):\n\n```\nNew FACTS?  volatile → RAG | stable+dense → CPT (LR ~10% of pretrain) → SFT\nNew BEHAVIOR? shifting → prompt-engineering | stable:\n  demos → SFT (LoRA\u002FQLoRA, all-linear, α=2r)\n  preference pairs → DPO (SimPO if length-bias, ORPO if memory-bound)\n  unpaired 👍\u002F👎 → KTO\n  verifiable success → RLVR + GRPO (DAPO\u002FGSPO\u002FDr.GRPO per failure mode)\nDeploy: FP8 (Hopper+) | NVFP4 (Blackwell scale) | AWQ (older) | GGUF+imatrix (edge)\nBEFORE ANY OF THIS: the eval harness must exist first.\n```\n\nRead the tree top-down: answer \"new facts or new\nbehavior,\" then follow the branch that matches\nthe data shape in hand (demos, preference pairs,\nthumbs up\u002Fdown, or verifiable success\u002Ffailure).\nThe data shape picks the method — not the other\nway around.\n\n### Worked Routing Examples\n\n- *\"Users want the assistant to follow our\n  support macros exactly.\"* Behavior is stable\n  and demonstrable from transcripts → demos →\n  **SFT**.\n- *\"We have pairs of good\u002Fbad responses from\n  reviewer thumbs-up\u002Fdown, unpaired.\"* → unpaired\n  signal → **KTO**, not DPO (DPO needs paired\n  preferences).\n- *\"The model can already solve some of these\n  math problems and we can grade correctness\n  automatically.\"* → verifiable success signal →\n  **GRPO+RLVR**, and only after confirming the\n  model succeeds at least sometimes (see Key\n  Routing Facts below).\n- *\"We want the model to know this week's\n  pricing page.\"* → volatile facts → **RAG**, no\n  training run at all.\n\n## Key Routing Facts\n\n- **Loss-function choice is low-leverage.** A\n  240-H100-run study found method choice worth\n  ~1 percentage point versus ~50 points for model\n  scale, and zero of 20 DPO variants beat vanilla\n  DPO. Don't spend a routing decision agonizing\n  over DPO-variant selection — spend it on\n  getting the data shape and scale right.\n- **DPO is for taste, GRPO+RLVR is for\n  reasoning.** Preference pairs that encode a\n  subjective judgment (tone, style, \"which answer\n  is better\") route to DPO. Tasks with a\n  verifiable pass\u002Ffail signal (math, code, tool\n  calls) route to GRPO+RLVR instead.\n- **RL is not the fix for a model that never\n  succeeds.** GRPO and other RL methods sharpen\n  an existing capability — they don't teach one\n  from zero. If the model doesn't yet understand\n  the task or output format, run SFT first; only\n  bring in RL once the model succeeds at least\n  sometimes.\n\n### Common Routing Mistakes\n\n- Reaching for fine-tuning to fix facts that\n  change weekly — that's a RAG problem, and\n  fine-tuning will just go stale faster than the\n  source data does.\n- Picking a DPO variant before checking whether\n  the actual bottleneck is data quality or model\n  scale — variant choice is the ~1pp lever, not\n  the ~50pp one.\n- Starting an RL run on a model that fails every\n  rollout — route to SFT first so RL has\n  something to sharpen.\n- Treating CPT as the default for \"the model\n  doesn't know our domain\" — check the data\n  volume thresholds first; under 500MB, RAG or\n  RAG+fine-tune iterates faster than a CPT run.\n\n## Model Selection\n\nBase-model choice is size-class first, family\nsecond, and it goes stale fast — so it lives in\nexactly one place: `references\u002Fmodel-catalog.md`.\nThat file is the only place in this plugin (and\nin the DGX Spark ops plugin) that names a base\nmodel family. Neither this skill nor\n`references\u002Fmemory-math.md` names one; both\ndescribe models by size class only (for example,\n\"8B-class LoRA,\" not a model name).\n\nThe catalog is dated on purpose — model rankings\nturn over quarterly. It carries a \"last verified\"\ndate and a refresh checklist. Before trusting a\nrow, check that date; if stale, work the refresh\nchecklist in the catalog before recommending a\nmodel from it.\n\n**Precedence when the catalog and a method skill\ndisagree:** the catalog's per-row Notes column\nstates hardware\u002Fsize-class *feasibility*, not a\nmethod recommendation — `lora-qlora-recipes`'s\nLoRA vs QLoRA vs Full FT table (routed by task\nshape) governs the actual method choice.\n\n## Memory Feasibility\n\nBefore committing to a method, size it: total\nmemory ≈ **params × dtype bytes + optimizer\nstate + gradients + activations**. Work each\nterm for the chosen dtype and method (full\nfine-tune, LoRA, or QLoRA) — worked worksheets\nand size-class examples live in\n`references\u002Fmemory-math.md`.\n\nOn DGX Spark specifically, unified-memory\nbehavior breaks the naive estimate (transient\nload peaks, `nvidia-smi` underreporting, thermal\nthrottling on long runs). Once the\n`dgx-spark-ops` plugin is installed, defer\nSpark-specific feasibility calls to its\n`spark-memory-thermal-ops` skill rather than\nre-deriving them here.\n\n## Related Skills\n\nOnce this skill has picked a method, hand off to\nthe skill that executes it:\n\n- `lora-qlora-recipes` — SFT via LoRA\u002FQLoRA\n- `preference-optimization` — DPO, ORPO, KTO\n- `grpo-rlvr-training` — GRPO with verifiable\n  rewards\n\nNo method is selected before the eval harness\nexists — see `eval-harness-first`.\n","# Fine-Tuning Method Selection\n\nThis is the router skill for the fine-tuning\nlifecycle: it decides whether fine-tuning is the\nright tool at all, and if so, which method and\nwhich base-model size class. Every other skill\nin this plugin assumes this routing already\nhappened — start here before opening\n`lora-qlora-recipes`, `preference-optimization`,\nor `grpo-rlvr-training`.\n\n## When to Use This Skill\n\n- Starting any fine-tuning effort, before a\n  framework or base model has been chosen.\n- Unsure whether RAG or prompt engineering would\n  solve the problem more cheaply than training.\n- Choosing between preference optimization (DPO\n  family) and a reinforcement method (GRPO\u002FRLVR)\n  for the same underlying task.\n- Sizing a candidate model\u002Fmethod combination\n  before committing to a run.\n\n## Quick Reference\n\n| Situation | Route |\n|---|---|\n| Facts change often (prices, docs, news) | RAG, not fine-tuning |\n| Desired behavior still being figured out | Prompt engineering |\n| Stable domain knowledge, ≥500MB text | CPT then SFT — see Off-Ramps First |\n| Have input\u002Foutput demonstrations | SFT — see `lora-qlora-recipes` |\n| Have preference pairs or thumbs-up\u002Fdown | DPO\u002FORPO\u002FKTO — see `preference-optimization` |\n| Have a verifiable pass\u002Ffail signal | GRPO+RLVR — see `grpo-rlvr-training` |\n| No eval harness yet | Stop — see `eval-harness-first` |\n\n## Off-Ramps First\n\nMost requests that sound like \"fine-tune this\"\nare served better and cheaper elsewhere. Check\nthese off-ramps before opening a training run:\n\n- **Knowledge-bound and volatile** (the gap is\n  facts that change — prices, docs, current\n  events): route to RAG, not fine-tuning. A\n  fine-tuned model bakes in a snapshot; volatile\n  facts go stale immediately.\n- **Behavior-bound and shifting** (the desired\n  behavior is still being figured out, or\n  changes per request): route to prompt\n  engineering. Fine-tuning locks in a behavior;\n  don't lock in one that hasn't stabilized yet.\n- **Stable, dense domain knowledge**: this is\n  where continued pretraining (CPT) enters, sized\n  by how much domain text exists:\n\n| Domain text volume | Route |\n|---|---|\n| \u003C10MB | RAG only |\n| 10MB–500MB | RAG + fine-tune |\n| 500MB–10GB | CPT, then SFT |\n| >10GB | CPT required |\n\nCPT learning rate ≈ **10% of the pretraining\nLR**. CPT is guidance-only in this plugin —\nsizing and LR guidance live here, but this\nplugin does not execute a CPT run.\n\n## Method Router\n\nOnce the off-ramps are ruled out, this is the\nfull decision tree (verbatim from the research\nthis plugin is built on):\n\n```\nNew FACTS?  volatile → RAG | stable+dense → CPT (LR ~10% of pretrain) → SFT\nNew BEHAVIOR? shifting → prompt-engineering | stable:\n  demos → SFT (LoRA\u002FQLoRA, all-linear, α=2r)\n  preference pairs → DPO (SimPO if length-bias, ORPO if memory-bound)\n  unpaired 👍\u002F👎 → KTO\n  verifiable success → RLVR + GRPO (DAPO\u002FGSPO\u002FDr.GRPO per failure mode)\nDeploy: FP8 (Hopper+) | NVFP4 (Blackwell scale) | AWQ (older) | GGUF+imatrix (edge)\nBEFORE ANY OF THIS: the eval harness must exist first.\n```\n\nRead the tree top-down: answer \"new facts or new\nbehavior,\" then follow the branch that matches\nthe data shape in hand (demos, preference pairs,\nthumbs up\u002Fdown, or verifiable success\u002Ffailure).\nThe data shape picks the method — not the other\nway around.\n\n### Worked Routing Examples\n\n- *\"Users want the assistant to follow our\n  support macros exactly.\"* Behavior is stable\n  and demonstrable from transcripts → demos →\n  **SFT**.\n- *\"We have pairs of good\u002Fbad responses from\n  reviewer thumbs-up\u002Fdown, unpaired.\"* → unpaired\n  signal → **KTO**, not DPO (DPO needs paired\n  preferences).\n- *\"The model can already solve some of these\n  math problems and we can grade correctness\n  automatically.\"* → verifiable success signal →\n  **GRPO+RLVR**, and only after confirming the\n  model succeeds at least sometimes (see Key\n  Routing Facts below).\n- *\"We want the model to know this week's\n  pricing page.\"* → volatile facts → **RAG**, no\n  training run at all.\n\n## Key Routing Facts\n\n- **Loss-function choice is low-leverage.** A\n  240-H100-run study found method choice worth\n  ~1 percentage point versus ~50 points for model\n  scale, and zero of 20 DPO variants beat vanilla\n  DPO. Don't spend a routing decision agonizing\n  over DPO-variant selection — spend it on\n  getting the data shape and scale right.\n- **DPO is for taste, GRPO+RLVR is for\n  reasoning.** Preference pairs that encode a\n  subjective judgment (tone, style, \"which answer\n  is better\") route to DPO. Tasks with a\n  verifiable pass\u002Ffail signal (math, code, tool\n  calls) route to GRPO+RLVR instead.\n- **RL is not the fix for a model that never\n  succeeds.** GRPO and other RL methods sharpen\n  an existing capability — they don't teach one\n  from zero. If the model doesn't yet understand\n  the task or output format, run SFT first; only\n  bring in RL once the model succeeds at least\n  sometimes.\n\n### Common Routing Mistakes\n\n- Reaching for fine-tuning to fix facts that\n  change weekly — that's a RAG problem, and\n  fine-tuning will just go stale faster than the\n  source data does.\n- Picking a DPO variant before checking whether\n  the actual bottleneck is data quality or model\n  scale — variant choice is the ~1pp lever, not\n  the ~50pp one.\n- Starting an RL run on a model that fails every\n  rollout — route to SFT first so RL has\n  something to sharpen.\n- Treating CPT as the default for \"the model\n  doesn't know our domain\" — check the data\n  volume thresholds first; under 500MB, RAG or\n  RAG+fine-tune iterates faster than a CPT run.\n\n## Model Selection\n\nBase-model choice is size-class first, family\nsecond, and it goes stale fast — so it lives in\nexactly one place: `references\u002Fmodel-catalog.md`.\nThat file is the only place in this plugin (and\nin the DGX Spark ops plugin) that names a base\nmodel family. Neither this skill nor\n`references\u002Fmemory-math.md` names one; both\ndescribe models by size class only (for example,\n\"8B-class LoRA,\" not a model name).\n\nThe catalog is dated on purpose — model rankings\nturn over quarterly. It carries a \"last verified\"\ndate and a refresh checklist. Before trusting a\nrow, check that date; if stale, work the refresh\nchecklist in the catalog before recommending a\nmodel from it.\n\n**Precedence when the catalog and a method skill\ndisagree:** the catalog's per-row Notes column\nstates hardware\u002Fsize-class *feasibility*, not a\nmethod recommendation — `lora-qlora-recipes`'s\nLoRA vs QLoRA vs Full FT table (routed by task\nshape) governs the actual method choice.\n\n## Memory Feasibility\n\nBefore committing to a method, size it: total\nmemory ≈ **params × dtype bytes + optimizer\nstate + gradients + activations**. Work each\nterm for the chosen dtype and method (full\nfine-tune, LoRA, or QLoRA) — worked worksheets\nand size-class examples live in\n`references\u002Fmemory-math.md`.\n\nOn DGX Spark specifically, unified-memory\nbehavior breaks the naive estimate (transient\nload peaks, `nvidia-smi` underreporting, thermal\nthrottling on long runs). Once the\n`dgx-spark-ops` plugin is installed, defer\nSpark-specific feasibility calls to its\n`spark-memory-thermal-ops` skill rather than\nre-deriving them here.\n\n## Related Skills\n\nOnce this skill has picked a method, hand off to\nthe skill that executes it:\n\n- `lora-qlora-recipes` — SFT via LoRA\u002FQLoRA\n- `preference-optimization` — DPO, ORPO, KTO\n- `grpo-rlvr-training` — GRPO with verifiable\n  rewards\n\nNo method is selected before the eval harness\nexists — see `eval-harness-first`.\n",{"name":5,"description":7},"main","58843b159d7a4050d6b785a98b908cce505c0b92",".agents\u002Fskills\u002Ffinetuning-method-selection",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 finetuning-method-selection",{"tool":38,"command":39},"bunx","bunx skills add Reihantt6\u002Fsns-myagent --skill finetuning-method-selection",{"tool":41,"command":42},"pnpm","pnpm dlx skills add Reihantt6\u002Fsns-myagent --skill finetuning-method-selection","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Ffinetuning-method-selection\u002FSKILL.md",true,[46,52],{"path":47,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":50,"sourceHtmlUrl":51},"references\u002Fmemory-math.md","REFERENCE","参考文档",5405,"https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Ffinetuning-method-selection\u002Freferences\u002Fmemory-math.md",{"path":53,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":54,"sourceHtmlUrl":55},"references\u002Fmodel-catalog.md",2724,"https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Ffinetuning-method-selection\u002Freferences\u002Fmodel-catalog.md",[57,64,71,78,85,92],{"skillNo":58,"name":59,"nameZh":6,"description":60,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":61,"categories":62,"keywords":63,"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":65,"name":66,"nameZh":6,"description":67,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":68,"categories":69,"keywords":70,"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":72,"name":73,"nameZh":6,"description":74,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":75,"categories":76,"keywords":77,"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":79,"name":80,"nameZh":6,"description":81,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":82,"categories":83,"keywords":84,"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":86,"name":87,"nameZh":6,"description":88,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":89,"categories":90,"keywords":91,"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":93,"name":94,"nameZh":6,"description":95,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":96,"categories":97,"keywords":98,"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",[],[]]