[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M71JJC31G16PP9ED8AXDCHM3":3,"skill-same:01M71JJC31G16PP9ED8AXDCHM3":79},{"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":39,"linkOnly":40,"markdownContent":41,"markdownBody":42,"frontMatter":43,"refName":44,"commitSha":45,"skillDirectory":5,"forksCount":46,"repositoryHtmlUrl":47,"repositoryHomepageUrl":48,"ownerHtmlUrl":49,"ownerBlogUrl":50,"ownerDescription":51,"updatedAt":52,"install":53,"resources":66},"01M71JJC31G16PP9ED8AXDCHM3","kaggle-grind",null,"Autonomous, disciplined Kaggle competition workflow — the loop that took Playground S6E6 to 19th\u002F2742 (top 0.7%). Use when working a Kaggle competition (Playground\u002Ftabular especially): building a model zoo, stacking, deciding what to submit, choosing final submissions, or reacting to a public leaderboard that looks suspiciously high. Encodes the hard-won rules: trust CV not a flip-farmable public LB, hunt and stack genuine public OOF corpora, gate every candidate on the statistical noise floor, hedge two genuine decorrelated finals (never a flip), verify finals every pass, and run a resume-surviving scheduled loop. Triggers: \"kaggle\", \"leaderboard\", \"OOF\", \"stack\", \"which submission\", \"final selection\", \"the LB shuffled\", \"public score\", \"should I submit\".","caezium\u002Fskills","caezium","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F113233555?v=4","01M71JB77ZPBAWEK009Y2W8BP7","Henry Zhang","01M71JB792NVXCWSYD27HQ0964",3,"MIT","https:\u002F\u002Fgithub.com\u002Fcaezium\u002Fskills\u002Fblob\u002Fmain\u002Fkaggle-grind\u002FSKILL.md",[],[19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38],"kaggle grind","kaggle","lb","trust","playground","competition","run","top","stack","candidate","workflow","when","scheduled","loop","autonomous","selection","submit","model","never","which","2026-11-10 10:12:30",false,"---\nname: kaggle-grind\ndescription: >-\n  Autonomous, disciplined Kaggle competition workflow — the loop that took Playground S6E6 to\n  19th\u002F2742 (top 0.7%). Use when working a Kaggle competition (Playground\u002Ftabular especially):\n  building a model zoo, stacking, deciding what to submit, choosing final submissions, or reacting\n  to a public leaderboard that looks suspiciously high. Encodes the hard-won rules: trust CV not a\n  flip-farmable public LB, hunt and stack genuine public OOF corpora, gate every candidate on the\n  statistical noise floor, hedge two genuine decorrelated finals (never a flip), verify finals every\n  pass, and run a resume-surviving scheduled loop. Triggers: \"kaggle\", \"leaderboard\", \"OOF\", \"stack\",\n  \"which submission\", \"final selection\", \"the LB shuffled\", \"public score\", \"should I submit\".\n---\n\n# kaggle-grind — the autonomous competition loop\n\nThis skill encodes a **process**, not a model. It reliably converts \"disciplined grind\" into medal-zone\nfinishes by refusing the traps that sink most teams. It was validated on Playground S6E6: **19th \u002F 2742\n(top 0.7%), ▲202 places in the private shakeup, above a Grandmaster** — by doing exactly the below.\n\n## The one idea that matters most\n**Public LB ≠ the target.** Many comps (especially Playground) score the public LB on a small sample\n(e.g. 20% of test). Under fragile metrics (balanced accuracy, small-sample AUC) the public top fills\nwith **public-LB-probing \u002F \"flip\"** notebooks that reverse-engineer *public* test-row labels. Those are a\n**no-op on the private set** and collapse in the shakeup. **Optimize CV. Treat the public LB as one noisy\nsample, never the objective.** (A Grandmaster confirmed this post-S6E6, verbatim.)\n\n## The loop (run each of these as a pass; a cron can drive it — see below)\n\n### 0. Day-1 recon (do this FIRST, not on day 20)\n- Read the LB description: **what % is public?** Note the metric and whether it's shakeup-prone.\n- Pull the #1 public notebook (`kaggle kernels list -c \u003Ccomp> --sort-by scoreDescending`,\n  `kaggle kernels pull`). **Classify it: genuine model vs public-LB-probe.** Fingerprints of a flip\u002Fprobe:\n  many teams at the *identical* 5-decimal score; 1–2-submission teams at the top; code that edits specific\n  `test_row_id -> label` using known public scores; hardcoded \"bad id\" lists.\n- Compute the **noise floor**: for balanced accuracy on n rows, SE ≈ sqrt(sum_k r_k(1-r_k)\u002Fn_k)\u002FK ≈ ~0.0003\n  at n~500k. **Any OOF\u002FLB Δ below ~2·SE (~0.0006) is noise.** Write it down; gate everything on it.\n\n### 1. Build a shared-fold, leak-free OOF zoo\n- ONE fold split across every model: `StratifiedKFold(k, shuffle=True, random_state=42)`. Non-negotiable —\n  it's what makes OOFs stackable. Save each model's OOF `(n_train,C)` + test `(n_test,C)` as .npy.\n- Diverse families: LGBM\u002FXGB\u002FCatBoost\u002FHistGB\u002FExtraTrees + a couple NN families + logreg\u002FkNN. But expect\n  trees to dominate tabular; NNs are often soft. Diversity for the *stack*, not for standalone strength.\n- Leakage paranoia: fit every y-dependent transform (target encoding, scaler, pseudo-labels) on train[~fold]\n  ONLY. Verify: a shuffled-label control must give ~chance.\n\n### 2. ★ Hunt and stack GENUINE public OOF corpora (the highest-EV lever)\n- This is usually what breaks your own-zoo ceiling. Search **early**:\n  `kaggle datasets list -s \"\u003Ccomp> oof\"`, `-s \"\u003Ccomp> stacker\"`, and scan top notebooks for shared\n  OOF\u002Ftest-pred datasets. Grandmasters publish these (e.g. cdeotte-style OOF+test dumps).\n- **RIGOR before trusting any external OOF**: id-align it to your train\u002Ftest order, then verify its OOF\n  argmax balanced-accuracy on your y lands in the sane model range (e.g. 0.95–0.97), NOT ~chance. A file\n  that scores chance is misaligned — reject it. (A naive misaligned blend cost us a submission once.)\n- Stacking public **OOFs** (train-side, out-of-fold, id-aligned) is legitimate and **private-valid** —\n  categorically different from copying a flip submission.\n\n### 3. Meta-stack + decision rule\n- Multinomial **Logistic Regression on log-prob (logit) features** of all OOFs. Regularized LR generalizes\n  where Caruana hill-climbing overfits. Sweep C.\n- **Correlation-prune** first: keep a model only if its argmax labels disagree with every selected model on\n  >~0.8% of rows. Diversity, not count, drives the gain.\n- **Decision rule** (metric-specific, often the biggest single lever): for balanced accuracy, tune a per-class\n  **log-offset** `[0,a,b]` added to log-probs before argmax — argmax under-predicts minorities.\n\n### 4. Gate, probe, accept\n- Accept a candidate only if OOF Δ **exceeds the noise floor** AND (ideally) an LB probe confirms. Many\n  \"higher OOF\" candidates LOSE on LB — trust LB-held gains, distrust 4th-decimal OOF nudges.\n- BUT on the **final day with free submission budget, PROBE every distinct genuine candidate anyway** — the\n  noise floor bounds confidence, not the value of a cheap real data point. (Our final #1 was \"within noise\"\n  on OOF yet +0.0011 on LB.)\n\n### 5. Finals (the medal decision — rule #13)\n- Pick **two GENUINE, decorrelated** submissions. **NEVER a flip** as a final (vanity on public, no-op\u002Frisk\n  on private). Submit strong candidates EARLY (dense plateaus tiebreak by submission time).\n- **Verify the selected finals EVERY pass** once submissions exist — assert both are genuine. (A flip had\n  auto-slotted into final #1; caught on deadline day. That check alone can save the medal.)\n- Manually pin — don't let Kaggle auto-pick two correlated public-bests.\n\n## Anti-patterns (proven to waste time \u002F lose)\n- Treating a flip-farmed public LB as the ceiling. (We wasted ~10 days on this once.)\n- Writing an **unqualified \"ceiling\"** conclusion. Scope it: \"ceiling of OUR models\" + always ask \"what would\n  a stronger external model add?\" Our \"genuine ceiling\" was wrong twice — public OOFs broke it.\n- Chasing 4th-decimal OOF gains, more tree-FE, or more correlated GBDTs. Cats that are binned color-indices,\n  external data with high adversarial AUC, iterative pseudo — usually ≤+0.0002 or negative.\n- Local-eval \"improvements\" that aren't verified to generalize (they overfit your exact local set).\n\n## Running it as a durable loop (optional)\n- Use the host harness's recurring-monitor or scheduled-task mechanism to fire a self-contained pass prompt.\n  In Claude Code this may be a cron integration; in Codex\u002FT3 use the product's recurring monitoring flow.\n  Guard with a lock file\n  (`\u002Ftmp\u002F\u003Ccomp>_loop.lock`, epoch + TTL) so passes don't overlap. `\u002Ftmp` is wiped on reboot — **re-download\n  public OOF datasets each pass.** Use `nohup python -u ...` for long jobs so they survive session resumes.\n- **Convergence downgrade**: if N consecutive passes find no candidate above the noise floor AND no new public\n  OOF source, drop cron frequency (hourly → 6h) — don't burn tokens re-confirming a ceiling.\n- **CLI\u002Fsubmit caveat**: Kaggle's `competitions.*`\u002F`kernels.*` API can 401 for hours while `datasets.*` works.\n  Confirm your real submit path EARLY (CLI vs browser-upload vs user-drag). The browser file-upload tool is\n  sandboxed to session-shared paths — it can't upload arbitrary repo files, so a human may need to drag-drop.\n\n## Reference\nSee `references\u002Fs6e6_retrospective.md` for the full worked example (every lever, the exact numbers, the\n7 workflow improvements) and `references\u002Frepro_stack.py` for the winning stack recipe.\n","# kaggle-grind — the autonomous competition loop\n\nThis skill encodes a **process**, not a model. It reliably converts \"disciplined grind\" into medal-zone\nfinishes by refusing the traps that sink most teams. It was validated on Playground S6E6: **19th \u002F 2742\n(top 0.7%), ▲202 places in the private shakeup, above a Grandmaster** — by doing exactly the below.\n\n## The one idea that matters most\n**Public LB ≠ the target.** Many comps (especially Playground) score the public LB on a small sample\n(e.g. 20% of test). Under fragile metrics (balanced accuracy, small-sample AUC) the public top fills\nwith **public-LB-probing \u002F \"flip\"** notebooks that reverse-engineer *public* test-row labels. Those are a\n**no-op on the private set** and collapse in the shakeup. **Optimize CV. Treat the public LB as one noisy\nsample, never the objective.** (A Grandmaster confirmed this post-S6E6, verbatim.)\n\n## The loop (run each of these as a pass; a cron can drive it — see below)\n\n### 0. Day-1 recon (do this FIRST, not on day 20)\n- Read the LB description: **what % is public?** Note the metric and whether it's shakeup-prone.\n- Pull the #1 public notebook (`kaggle kernels list -c \u003Ccomp> --sort-by scoreDescending`,\n  `kaggle kernels pull`). **Classify it: genuine model vs public-LB-probe.** Fingerprints of a flip\u002Fprobe:\n  many teams at the *identical* 5-decimal score; 1–2-submission teams at the top; code that edits specific\n  `test_row_id -> label` using known public scores; hardcoded \"bad id\" lists.\n- Compute the **noise floor**: for balanced accuracy on n rows, SE ≈ sqrt(sum_k r_k(1-r_k)\u002Fn_k)\u002FK ≈ ~0.0003\n  at n~500k. **Any OOF\u002FLB Δ below ~2·SE (~0.0006) is noise.** Write it down; gate everything on it.\n\n### 1. Build a shared-fold, leak-free OOF zoo\n- ONE fold split across every model: `StratifiedKFold(k, shuffle=True, random_state=42)`. Non-negotiable —\n  it's what makes OOFs stackable. Save each model's OOF `(n_train,C)` + test `(n_test,C)` as .npy.\n- Diverse families: LGBM\u002FXGB\u002FCatBoost\u002FHistGB\u002FExtraTrees + a couple NN families + logreg\u002FkNN. But expect\n  trees to dominate tabular; NNs are often soft. Diversity for the *stack*, not for standalone strength.\n- Leakage paranoia: fit every y-dependent transform (target encoding, scaler, pseudo-labels) on train[~fold]\n  ONLY. Verify: a shuffled-label control must give ~chance.\n\n### 2. ★ Hunt and stack GENUINE public OOF corpora (the highest-EV lever)\n- This is usually what breaks your own-zoo ceiling. Search **early**:\n  `kaggle datasets list -s \"\u003Ccomp> oof\"`, `-s \"\u003Ccomp> stacker\"`, and scan top notebooks for shared\n  OOF\u002Ftest-pred datasets. Grandmasters publish these (e.g. cdeotte-style OOF+test dumps).\n- **RIGOR before trusting any external OOF**: id-align it to your train\u002Ftest order, then verify its OOF\n  argmax balanced-accuracy on your y lands in the sane model range (e.g. 0.95–0.97), NOT ~chance. A file\n  that scores chance is misaligned — reject it. (A naive misaligned blend cost us a submission once.)\n- Stacking public **OOFs** (train-side, out-of-fold, id-aligned) is legitimate and **private-valid** —\n  categorically different from copying a flip submission.\n\n### 3. Meta-stack + decision rule\n- Multinomial **Logistic Regression on log-prob (logit) features** of all OOFs. Regularized LR generalizes\n  where Caruana hill-climbing overfits. Sweep C.\n- **Correlation-prune** first: keep a model only if its argmax labels disagree with every selected model on\n  >~0.8% of rows. Diversity, not count, drives the gain.\n- **Decision rule** (metric-specific, often the biggest single lever): for balanced accuracy, tune a per-class\n  **log-offset** `[0,a,b]` added to log-probs before argmax — argmax under-predicts minorities.\n\n### 4. Gate, probe, accept\n- Accept a candidate only if OOF Δ **exceeds the noise floor** AND (ideally) an LB probe confirms. Many\n  \"higher OOF\" candidates LOSE on LB — trust LB-held gains, distrust 4th-decimal OOF nudges.\n- BUT on the **final day with free submission budget, PROBE every distinct genuine candidate anyway** — the\n  noise floor bounds confidence, not the value of a cheap real data point. (Our final #1 was \"within noise\"\n  on OOF yet +0.0011 on LB.)\n\n### 5. Finals (the medal decision — rule #13)\n- Pick **two GENUINE, decorrelated** submissions. **NEVER a flip** as a final (vanity on public, no-op\u002Frisk\n  on private). Submit strong candidates EARLY (dense plateaus tiebreak by submission time).\n- **Verify the selected finals EVERY pass** once submissions exist — assert both are genuine. (A flip had\n  auto-slotted into final #1; caught on deadline day. That check alone can save the medal.)\n- Manually pin — don't let Kaggle auto-pick two correlated public-bests.\n\n## Anti-patterns (proven to waste time \u002F lose)\n- Treating a flip-farmed public LB as the ceiling. (We wasted ~10 days on this once.)\n- Writing an **unqualified \"ceiling\"** conclusion. Scope it: \"ceiling of OUR models\" + always ask \"what would\n  a stronger external model add?\" Our \"genuine ceiling\" was wrong twice — public OOFs broke it.\n- Chasing 4th-decimal OOF gains, more tree-FE, or more correlated GBDTs. Cats that are binned color-indices,\n  external data with high adversarial AUC, iterative pseudo — usually ≤+0.0002 or negative.\n- Local-eval \"improvements\" that aren't verified to generalize (they overfit your exact local set).\n\n## Running it as a durable loop (optional)\n- Use the host harness's recurring-monitor or scheduled-task mechanism to fire a self-contained pass prompt.\n  In Claude Code this may be a cron integration; in Codex\u002FT3 use the product's recurring monitoring flow.\n  Guard with a lock file\n  (`\u002Ftmp\u002F\u003Ccomp>_loop.lock`, epoch + TTL) so passes don't overlap. `\u002Ftmp` is wiped on reboot — **re-download\n  public OOF datasets each pass.** Use `nohup python -u ...` for long jobs so they survive session resumes.\n- **Convergence downgrade**: if N consecutive passes find no candidate above the noise floor AND no new public\n  OOF source, drop cron frequency (hourly → 6h) — don't burn tokens re-confirming a ceiling.\n- **CLI\u002Fsubmit caveat**: Kaggle's `competitions.*`\u002F`kernels.*` API can 401 for hours while `datasets.*` works.\n  Confirm your real submit path EARLY (CLI vs browser-upload vs user-drag). The browser file-upload tool is\n  sandboxed to session-shared paths — it can't upload arbitrary repo files, so a human may need to drag-drop.\n\n## Reference\nSee `references\u002Fs6e6_retrospective.md` for the full worked example (every lever, the exact numbers, the\n7 workflow improvements) and `references\u002Frepro_stack.py` for the winning stack recipe.\n",{"name":5,"description":7},"main","6ef866230c1ba3459fd08fc65141dfd37fffd41a",0,"https:\u002F\u002Fgithub.com\u002Fcaezium\u002Fskills","https:\u002F\u002Fskills.henryzh.dev","https:\u002F\u002Fgithub.com\u002Fcaezium","henryzh.dev","quant, data science, full-stack, robotics.\r\nbuilding cool stuff","2026-08-24 19:21:44",{"source":8,"skillName":5,"commands":54,"prompt":64,"zipAvailable":65},[55,58,61],{"tool":56,"command":57},"npx","npx skills add caezium\u002Fskills --skill kaggle-grind",{"tool":59,"command":60},"bunx","bunx skills add caezium\u002Fskills --skill kaggle-grind",{"tool":62,"command":63},"pnpm","pnpm dlx skills add caezium\u002Fskills --skill kaggle-grind","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fcaezium\u002Fskills\u002Fblob\u002Fmain\u002Fkaggle-grind\u002FSKILL.md",true,[67,73],{"path":68,"resourceType":69,"resourceTypeDesc":70,"sizeBytes":71,"sourceHtmlUrl":72},"references\u002Frepro_stack.py","SCRIPT","脚本",7917,"https:\u002F\u002Fgithub.com\u002Fcaezium\u002Fskills\u002Fblob\u002Fmain\u002Fkaggle-grind\u002Freferences\u002Frepro_stack.py",{"path":74,"resourceType":75,"resourceTypeDesc":76,"sizeBytes":77,"sourceHtmlUrl":78},"references\u002Fs6e6_retrospective.md","REFERENCE","参考文档",5988,"https:\u002F\u002Fgithub.com\u002Fcaezium\u002Fskills\u002Fblob\u002Fmain\u002Fkaggle-grind\u002Freferences\u002Fs6e6_retrospective.md",[80,103,131,149,170,192],{"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":96,"publishedAt":39},"01M71JT4RYCAS18RF6003YYR1D","writing-for-agents","Writing documents for agents. Use when creating or editing skills, or modifying AGENTS.md or CLAUDE.md.","https:\u002F\u002Fgithub.com\u002Fcaezium\u002Fskills\u002Fblob\u002Fmain\u002Fwriting-for-agents\u002FSKILL.md",[86,91],{"categoryNo":87,"categoryName":88,"taxonomyType":89,"taxonomyTypeDesc":90},"01JBSEEDCAT000000000000018","写作与文案","SCENARIO","使用场景",{"categoryNo":92,"categoryName":93,"taxonomyType":94,"taxonomyTypeDesc":95},"01JBSEEDCAT000000000000036","内容创作者","PROFESSION","适用职业",[97,98,99,100,101,102,30],"writing for agents","for","writing","editing","use","documents",{"skillNo":104,"name":105,"nameZh":6,"description":106,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":107,"categories":108,"keywords":112,"publishedAt":39},"01M71JT0GES4AZ5ADH8EADNN1Z","wrangler","Cloudflare Workers CLI for deploying, developing, and managing Workers, KV, R2, D1, Vectorize, Hyperdrive, Workers AI, Containers, Queues, Workflows, Pipelines, and Secrets Store. Load before running wrangler commands to ensure correct syntax and best practices. Biases towards retrieval from Cloudflare docs over pre-trained knowledge.","https:\u002F\u002Fgithub.com\u002Fcaezium\u002Fskills\u002Fblob\u002Fmain\u002Fwrangler\u002FSKILL.md",[109],{"categoryNo":110,"categoryName":111,"taxonomyType":89,"taxonomyTypeDesc":90},"01JBSEEDCAT000000000000010","后端与接口",[105,113,114,115,116,117,118,119,120,121,122,123,124,98,125,126,127,128,129,130],"cloudflare","d1","knowledge","workflows","cli","secrets","commands","syntax","cloudflare workers","r2","queues","running","containers","correct","load","workers","best practices","store",{"skillNo":132,"name":133,"nameZh":6,"description":134,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":135,"categories":136,"keywords":137,"publishedAt":39},"01M71JSWX4R203M32K9A18AAW8","workers-best-practices","Reviews and authors Cloudflare Workers code against production best practices. Load when writing new Workers, reviewing Worker code, configuring wrangler.jsonc, or checking for common Workers anti-patterns (streaming, floating promises, global state, secrets, bindings, observability). Biases towards retrieval from Cloudflare docs over pre-trained knowledge.","https:\u002F\u002Fgithub.com\u002Fcaezium\u002Fskills\u002Fblob\u002Fmain\u002Fworkers-best-practices\u002FSKILL.md",[],[138,128,129,139,140,141,98,113,142,115,143,105,99,127,30,144,145,146,147,148],"workers best practices","observability","reviews","promises","worker","new","code","common","streaming","state","anti-patterns",{"skillNo":150,"name":151,"nameZh":6,"description":152,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":153,"categories":154,"keywords":155,"publishedAt":39},"01M71JST6P88MXGT2MWDTPJGEG","wizard","Generate an interactive bash wizard that walks a human through steps only they can perform. Use when provisioning infrastructure, setting up credentials or CI secrets, walking an unfamiliar third-party dashboard, or running a one-off migration or cutover. Don't invoke this for steps the agent can perform itself.","https:\u002F\u002Fgithub.com\u002Fcaezium\u002Fskills\u002Fblob\u002Fmain\u002Fwizard\u002FSKILL.md",[],[151,124,156,98,157,158,159,160,101,161,30,162,163,164,165,166,167,168,169,118],"setting","up","ci","provisioning infrastructure","only","invoke","Bash","one","interactive","can","credentials","dashboard","generate","human",{"skillNo":171,"name":172,"nameZh":6,"description":173,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":174,"sourceHtmlUrl":175,"categories":176,"keywords":177,"publishedAt":39},"01M71JSQ95S7BQ61DAPN95X2W4","why","Use for 'why does X work this way', 'why we picked Y', design rationale, regressions, postmortems, or data-backed thresholds. Discovers available evidence sources, searches each relevant category, and returns a cited read on decisions and tradeoffs. Use how for runtime behavior.","https:\u002F\u002Fgithub.com\u002Fcaezium\u002Fskills\u002Fblob\u002Fmain\u002Fwhy\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fcaezium\u002Fskills\u002Fblob\u002Fmain\u002Fwhy\u002FSKILL.md",[],[172,178,179,180,181,98,182,183,101,184,185,186,187,188,189,190,191],"sources","thresholds","evidence","relevant","runtime","design","decisions","work","how","behavior","read","data","rationale","returns",{"skillNo":193,"name":194,"nameZh":6,"description":195,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":196,"categories":197,"keywords":201,"publishedAt":39},"01M71JSMHFVSH8HK8GXN6YAX6J","week-prep","Generate a briefing with calendar, tasks, and attention flags. Use for \"week prep\", \"briefing\", or \"what's on this week\".","https:\u002F\u002Fgithub.com\u002Fcaezium\u002Fskills\u002Fblob\u002Fmain\u002Fweek-prep\u002FSKILL.md",[198],{"categoryNo":199,"categoryName":200,"taxonomyType":89,"taxonomyTypeDesc":90},"01JBSEEDCAT000000000000023","Skill与智能体开发",[202,203,204,205,206,98,207,208,209,210,101,162,211,212,213,168,214,188,215,216],"week prep","prep","deep","attention","get","search","messages","briefing","MCP","tasks","with","slack","calendar","flags","grep"]