[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M77M0GEMCX3CFFKTMSN705W9":3,"skill-same:01M77M0GEMCX3CFFKTMSN705W9":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},"01M77M0GEMCX3CFFKTMSN705W9","checkpoint-promotion",null,"Gate fine-tuned checkpoints with drift budgets, paired comparison, and forgetting checks before promotion. Use after a training run produces a checkpoint, when deciding whether a tuned model ships, or when a promoted model needs re-gating against updated goldens.","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\u002Fcheckpoint-promotion\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: checkpoint-promotion\ndescription: Gate fine-tuned checkpoints with drift budgets, paired comparison, and forgetting checks before promotion. Use after a training run produces a checkpoint, when deciding whether a tuned model ships, or when a promoted model needs re-gating against updated goldens.\n---\n\n# Checkpoint Promotion\n\nThe Phase 5 gate for the whole\nplugin: a checkpoint that trains\ncleanly and beats its task metric\nstill doesn't ship without\nclearing all four stages below.\n`eval-harness-first` built the\nsuite re-run here — this skill is\nwhere that suite's baseline\ndecides something.\n\n**Input:** a trained checkpoint,\n`eval\u002Fbaseline-\u003Cmodel>.json` from\n`eval-harness-first`, and the\nfrozen `eval\u002Fdrift-suite.yaml`.\n**Output format:**\n`promotion-report.md` — the\nfour-stage evidence plus a\nterminal `PROMOTE` or `REJECT`\nverdict that `\u002Ffinetune` Phase 5\nand `\u002Fpromote-checkpoint` consume\ndirectly.\n\n## The Four-Stage Gate\n\nEach stage gates the next — a\nfailure at stage 2 means stage 3\ndoesn't run. Stages 2 and 3 share\none expensive inference pass, so\nrunning them concurrently and\napplying gate order at verdict\ntime is licensed on a\n**deterministic** arena (nothing\nsaved by serializing); a\njudge-based arena should still\nwait for stage 2 first — that's\nwhere the real savings are.\n\n1. **Data-quality gate.** Before\n   any eval touches the\n   checkpoint: dedup the training\n   set, check for eval-goldens\n   leakage (the exact failure\n   `trace-to-training-data`'s\n   Hygiene section exists to\n   prevent), and scan for label\n   noise. A checkpoint trained on\n   leaked goldens invalidates\n   every later stage.\n2. **Held-out + frozen\n   capability-drift suite.**\n   Re-run `eval-harness-first`'s\n   `eval\u002Fdrift-suite.yaml` —\n   MMLU\u002FGSM8K\u002FIFEval plus 200–500\n   domain-adjacent items — against\n   the checkpoint and diff against\n   `baseline-\u003Cmodel>.json` per\n   benchmark against the Drift\n   Budget table below.\n3. **Paired arena vs. base.**\n   Position-randomized judge,\n   checkpoint vs. base model, same\n   prompts — or the deterministic\n   paired-comparison variant in\n   `references\u002Fgate-templates.md`\n   when every grader in the\n   harness is deterministic (no\n   LLM-judge; position\n   randomization N\u002FA there).\n   **A holdout win that\n   loses the live arena does not\n   ship** — stage-2 numbers and\n   stage-3 judgments must agree; a\n   win on frozen goldens and a\n   loss in paired comparison is a\n   real signal, not a discrepancy\n   to explain away.\n4. **Canary.** 5–10% stratified\n   rollout with auto-rollback for\n   any checkpoint reaching\n   production traffic. **Local-only\n   users stop at stage 3** —\n   skipping stage 4 for a local\n   deployment is the correct\n   stopping point, not a shortcut.\n\n### Drift Budget\n\n| Drift (pts) | Verdict |\n|---|---|\n| ≤1 | Noise — proceed |\n| 2–5 | Rerun with seed variation before deciding |\n| >5 | **HARD FAIL** — no exception for task gains |\n\nThe >5pt row governs regardless\nof the others: a checkpoint that\ngained 8 points on the target\ntask and lost 6 points of general\ncapability still fails here —\ntask improvement never buys back\na drift-budget breach.\n\n**Item count derives from the\nbudget, not convenience:** the\nstrict n for a half-width under\nhalf the 5pt hard-fail threshold\nis ~1,300 at typical accuracy\n(p≈0.7); n=200 is a pragmatic\nfloor (±6pt half-width at that\nsame p, n=50 ±13pt) — report the\nhalf-width with every verdict,\nand treat a margin smaller than\nit as `REJECT (uncertain)`, not\nPASS\u002FHARD FAIL. Full math and a\n5-run cautionary example:\n`references\u002Fgate-templates.md`.\n\n**RERUN is not a verdict.** A\n2–5pt drift only ever produces a\n`PROMOTE` or `REJECT` after the\nseed-variation rerun completes —\n`PROMOTE` requires landing back\nat ≤1pt (noise); any rerun still\n>1pt — 2–5pt band or >5pt breach\nalike — resolves stage 2 to a\nhard `REJECT`. No report may\nreach the Verdict section with\nstage 2 still showing `RERUN`.\n\n## Catastrophic Forgetting\n\nUnmanaged LoRA fine-tuning loses\nreal general capability, and\nstage 2 is what catches it:\n\n- **~43% knowledge loss\n  unmanaged** — no replay, no\n  regularization.\n- **~10% with basic management**\n  — some replay or a conservative\n  LR.\n- **~3% with replay + EWC** — the\n  disciplined case.\n- **10–30% general-data replay\n  mix is the standard\n  mitigation** — blend general-\n  domain data into training\n  rather than target-task data\n  alone.\n\nIf a checkpoint hits the >5pt\nhard fail in stage 2, work this\nescalation ladder in order — the\none canonical order this skill\nand `references\u002Fgate-templates.md`\nboth point to:\n\n1. **Adjust the replay-mix\n   fraction — swap rows, don't\n   add them** (adding confounds\n   fraction with total optimizer\n   steps). Dose is not monotonic\n   at small-run scale (\u003C~100\n   steps) — re-check drift after\n   any swap.\n2. **Lower the learning rate.**\n3. **Fewer epochs.**\n4. **A smaller LoRA rank** — the\n   same rank\u002FLR levers\n   `lora-qlora-recipes` and\n   `preference-optimization` tune\n   for the training run, applied\n   here in reverse.\n\nThis order is a default, not a\nlaw: **remediation guidance from\na single before\u002Fafter run pair\nis a hypothesis** — label it\nlow-confidence once any lever\nproduces a reversal, and prefer\na seed-variation repeat over\ntrusting the next rung blindly.\nA lever that clears the drift\nbreach but drops a\nsuccess-criterion metric below\ntarget is a two-sided tradeoff\nfor a human, not a reason to\nkeep descending the ladder. Full\nreasoning and the 5-run\ntrajectory behind both caveats:\n`references\u002Fgate-templates.md`.\n\n**Disclose drift-suite\ninstruction reuse.** A replay row\ncopying the drift harness's exact\ninstruction phrasing (not just\ndisjoint source items) makes that\nbenchmark's post-replay score an\nupper bound — flag it\ninstruction-familiar, or re-probe\nwith a paraphrase, before\ntreating a near-budget pass as\nclean.\n\n## The Verdict\n\n`promotion-report.md` covers all\nfour stages as sections and\n**must end with a terminal\nverdict: `PROMOTE` or `REJECT`**,\nthe evidence that produced it,\nand exactly one top remediation\nwhen the verdict is `REJECT`.\nTemplate: `references\u002Fgate-templates.md`.\nThe terminal contract other\nskills parse:\n\n```\n## Verdict\n\nREJECT\n\nEvidence: domain-adjacent drift\nsuite dropped 6.2pt (threshold:\n>5pt hard fail) despite +8pt on\nthe target task.\n\nTop remediation: swap the\nreplay-mix fraction from 10%\ntoward 20%, holding step count\nconstant.\n```\n\n- **REJECT is a result, not an\n  error.** A checkpoint that\n  fails stage 2's drift budget or\n  stage 3's arena comparison did\n  its job. Don't treat a REJECT\n  as a failed run needing a rerun\n  of this skill; it's the correct\n  output of a working gate.\n- **One remediation, not a\n  menu.** Evidence sections may\n  list everything observed; the\n  verdict section names the\n  single highest-leverage fix per\n  the escalation ladder above. A\n  report that hedges across three\n  possible fixes hasn't done the\n  prioritization this skill\n  exists to do.\n- **No auto-retraining.** This\n  skill produces a verdict and a\n  report, not a re-triggered\n  training run. A `REJECT` hands\n  the remediation back to a human\n  decision at\n  `finetuning-method-selection` or\n  the relevant training skill.\n\n## Related Skills\n\n- `eval-harness-first` — owns the\n  drift suite and baseline this\n  skill re-runs and diffs\n  against; no `baseline-\u003Cmodel>.json`\n  means nothing to gate against.\n- `quantized-export` — the only\n  valid next step after a\n  `PROMOTE` verdict.\n- `preference-optimization` and\n  `lora-qlora-recipes` — own the\n  LR and rank levers in the\n  Catastrophic Forgetting\n  escalation path; this skill\n  diagnoses the breach, those\n  skills own the config that\n  caused it.\n- `dataset-curation` — owns the\n  replay-mix construction recipe\n  the escalation ladder's first\n  rung applies.\n\nComplete `promotion-report.md`\ntemplate with all four stages,\nthe drift-suite scoring table,\nthe paired-arena protocol (item\ncount, position randomization,\nwin-rate threshold), and a\nreplay-mix configuration example:\n`references\u002Fgate-templates.md`.\n","# Checkpoint Promotion\n\nThe Phase 5 gate for the whole\nplugin: a checkpoint that trains\ncleanly and beats its task metric\nstill doesn't ship without\nclearing all four stages below.\n`eval-harness-first` built the\nsuite re-run here — this skill is\nwhere that suite's baseline\ndecides something.\n\n**Input:** a trained checkpoint,\n`eval\u002Fbaseline-\u003Cmodel>.json` from\n`eval-harness-first`, and the\nfrozen `eval\u002Fdrift-suite.yaml`.\n**Output format:**\n`promotion-report.md` — the\nfour-stage evidence plus a\nterminal `PROMOTE` or `REJECT`\nverdict that `\u002Ffinetune` Phase 5\nand `\u002Fpromote-checkpoint` consume\ndirectly.\n\n## The Four-Stage Gate\n\nEach stage gates the next — a\nfailure at stage 2 means stage 3\ndoesn't run. Stages 2 and 3 share\none expensive inference pass, so\nrunning them concurrently and\napplying gate order at verdict\ntime is licensed on a\n**deterministic** arena (nothing\nsaved by serializing); a\njudge-based arena should still\nwait for stage 2 first — that's\nwhere the real savings are.\n\n1. **Data-quality gate.** Before\n   any eval touches the\n   checkpoint: dedup the training\n   set, check for eval-goldens\n   leakage (the exact failure\n   `trace-to-training-data`'s\n   Hygiene section exists to\n   prevent), and scan for label\n   noise. A checkpoint trained on\n   leaked goldens invalidates\n   every later stage.\n2. **Held-out + frozen\n   capability-drift suite.**\n   Re-run `eval-harness-first`'s\n   `eval\u002Fdrift-suite.yaml` —\n   MMLU\u002FGSM8K\u002FIFEval plus 200–500\n   domain-adjacent items — against\n   the checkpoint and diff against\n   `baseline-\u003Cmodel>.json` per\n   benchmark against the Drift\n   Budget table below.\n3. **Paired arena vs. base.**\n   Position-randomized judge,\n   checkpoint vs. base model, same\n   prompts — or the deterministic\n   paired-comparison variant in\n   `references\u002Fgate-templates.md`\n   when every grader in the\n   harness is deterministic (no\n   LLM-judge; position\n   randomization N\u002FA there).\n   **A holdout win that\n   loses the live arena does not\n   ship** — stage-2 numbers and\n   stage-3 judgments must agree; a\n   win on frozen goldens and a\n   loss in paired comparison is a\n   real signal, not a discrepancy\n   to explain away.\n4. **Canary.** 5–10% stratified\n   rollout with auto-rollback for\n   any checkpoint reaching\n   production traffic. **Local-only\n   users stop at stage 3** —\n   skipping stage 4 for a local\n   deployment is the correct\n   stopping point, not a shortcut.\n\n### Drift Budget\n\n| Drift (pts) | Verdict |\n|---|---|\n| ≤1 | Noise — proceed |\n| 2–5 | Rerun with seed variation before deciding |\n| >5 | **HARD FAIL** — no exception for task gains |\n\nThe >5pt row governs regardless\nof the others: a checkpoint that\ngained 8 points on the target\ntask and lost 6 points of general\ncapability still fails here —\ntask improvement never buys back\na drift-budget breach.\n\n**Item count derives from the\nbudget, not convenience:** the\nstrict n for a half-width under\nhalf the 5pt hard-fail threshold\nis ~1,300 at typical accuracy\n(p≈0.7); n=200 is a pragmatic\nfloor (±6pt half-width at that\nsame p, n=50 ±13pt) — report the\nhalf-width with every verdict,\nand treat a margin smaller than\nit as `REJECT (uncertain)`, not\nPASS\u002FHARD FAIL. Full math and a\n5-run cautionary example:\n`references\u002Fgate-templates.md`.\n\n**RERUN is not a verdict.** A\n2–5pt drift only ever produces a\n`PROMOTE` or `REJECT` after the\nseed-variation rerun completes —\n`PROMOTE` requires landing back\nat ≤1pt (noise); any rerun still\n>1pt — 2–5pt band or >5pt breach\nalike — resolves stage 2 to a\nhard `REJECT`. No report may\nreach the Verdict section with\nstage 2 still showing `RERUN`.\n\n## Catastrophic Forgetting\n\nUnmanaged LoRA fine-tuning loses\nreal general capability, and\nstage 2 is what catches it:\n\n- **~43% knowledge loss\n  unmanaged** — no replay, no\n  regularization.\n- **~10% with basic management**\n  — some replay or a conservative\n  LR.\n- **~3% with replay + EWC** — the\n  disciplined case.\n- **10–30% general-data replay\n  mix is the standard\n  mitigation** — blend general-\n  domain data into training\n  rather than target-task data\n  alone.\n\nIf a checkpoint hits the >5pt\nhard fail in stage 2, work this\nescalation ladder in order — the\none canonical order this skill\nand `references\u002Fgate-templates.md`\nboth point to:\n\n1. **Adjust the replay-mix\n   fraction — swap rows, don't\n   add them** (adding confounds\n   fraction with total optimizer\n   steps). Dose is not monotonic\n   at small-run scale (\u003C~100\n   steps) — re-check drift after\n   any swap.\n2. **Lower the learning rate.**\n3. **Fewer epochs.**\n4. **A smaller LoRA rank** — the\n   same rank\u002FLR levers\n   `lora-qlora-recipes` and\n   `preference-optimization` tune\n   for the training run, applied\n   here in reverse.\n\nThis order is a default, not a\nlaw: **remediation guidance from\na single before\u002Fafter run pair\nis a hypothesis** — label it\nlow-confidence once any lever\nproduces a reversal, and prefer\na seed-variation repeat over\ntrusting the next rung blindly.\nA lever that clears the drift\nbreach but drops a\nsuccess-criterion metric below\ntarget is a two-sided tradeoff\nfor a human, not a reason to\nkeep descending the ladder. Full\nreasoning and the 5-run\ntrajectory behind both caveats:\n`references\u002Fgate-templates.md`.\n\n**Disclose drift-suite\ninstruction reuse.** A replay row\ncopying the drift harness's exact\ninstruction phrasing (not just\ndisjoint source items) makes that\nbenchmark's post-replay score an\nupper bound — flag it\ninstruction-familiar, or re-probe\nwith a paraphrase, before\ntreating a near-budget pass as\nclean.\n\n## The Verdict\n\n`promotion-report.md` covers all\nfour stages as sections and\n**must end with a terminal\nverdict: `PROMOTE` or `REJECT`**,\nthe evidence that produced it,\nand exactly one top remediation\nwhen the verdict is `REJECT`.\nTemplate: `references\u002Fgate-templates.md`.\nThe terminal contract other\nskills parse:\n\n```\n## Verdict\n\nREJECT\n\nEvidence: domain-adjacent drift\nsuite dropped 6.2pt (threshold:\n>5pt hard fail) despite +8pt on\nthe target task.\n\nTop remediation: swap the\nreplay-mix fraction from 10%\ntoward 20%, holding step count\nconstant.\n```\n\n- **REJECT is a result, not an\n  error.** A checkpoint that\n  fails stage 2's drift budget or\n  stage 3's arena comparison did\n  its job. Don't treat a REJECT\n  as a failed run needing a rerun\n  of this skill; it's the correct\n  output of a working gate.\n- **One remediation, not a\n  menu.** Evidence sections may\n  list everything observed; the\n  verdict section names the\n  single highest-leverage fix per\n  the escalation ladder above. A\n  report that hedges across three\n  possible fixes hasn't done the\n  prioritization this skill\n  exists to do.\n- **No auto-retraining.** This\n  skill produces a verdict and a\n  report, not a re-triggered\n  training run. A `REJECT` hands\n  the remediation back to a human\n  decision at\n  `finetuning-method-selection` or\n  the relevant training skill.\n\n## Related Skills\n\n- `eval-harness-first` — owns the\n  drift suite and baseline this\n  skill re-runs and diffs\n  against; no `baseline-\u003Cmodel>.json`\n  means nothing to gate against.\n- `quantized-export` — the only\n  valid next step after a\n  `PROMOTE` verdict.\n- `preference-optimization` and\n  `lora-qlora-recipes` — own the\n  LR and rank levers in the\n  Catastrophic Forgetting\n  escalation path; this skill\n  diagnoses the breach, those\n  skills own the config that\n  caused it.\n- `dataset-curation` — owns the\n  replay-mix construction recipe\n  the escalation ladder's first\n  rung applies.\n\nComplete `promotion-report.md`\ntemplate with all four stages,\nthe drift-suite scoring table,\nthe paired-arena protocol (item\ncount, position randomization,\nwin-rate threshold), and a\nreplay-mix configuration example:\n`references\u002Fgate-templates.md`.\n",{"name":5,"description":7},"main","58843b159d7a4050d6b785a98b908cce505c0b92",".agents\u002Fskills\u002Fcheckpoint-promotion",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 checkpoint-promotion",{"tool":38,"command":39},"bunx","bunx skills add Reihantt6\u002Fsns-myagent --skill checkpoint-promotion",{"tool":41,"command":42},"pnpm","pnpm dlx skills add Reihantt6\u002Fsns-myagent --skill checkpoint-promotion","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fcheckpoint-promotion\u002FSKILL.md",true,[46],{"path":47,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":50,"sourceHtmlUrl":51},"references\u002Fgate-templates.md","REFERENCE","参考文档",13172,"https:\u002F\u002Fgithub.com\u002FReihantt6\u002Fsns-myagent\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fcheckpoint-promotion\u002Freferences\u002Fgate-templates.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",[],[]]