[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7B2VR3X3JXG5B3NNXRGBXR5":3,"skill-same:01M7B2VR3X3JXG5B3NNXRGBXR5":49},{"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":26,"commitSha":27,"skillDirectory":28,"forksCount":29,"repositoryHtmlUrl":30,"repositoryHomepageUrl":6,"ownerHtmlUrl":31,"ownerBlogUrl":32,"ownerDescription":33,"updatedAt":34,"install":35,"resources":48},"01M7B2VR3X3JXG5B3NNXRGBXR5","eighty-twenty",null,"Get most of the value for a small fraction of the time, deliberately and with the blind spot named. Use before any expensive repeated activity — slow validation or build cycles, broad searches, large agent fan-outs, exhaustive captures, big sampling runs, reading a whole codebase — or whenever someone says this is too slow, speed it up, we need to go faster, or just give me the gist.","gethamster\u002Fskills","gethamster","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F215697643?v=4","01M7B2R6PGB3SDZZ71E0WE2S74","Hamster","01M7B2R6Q21HWJRXZZWB78BNDG",4,"MIT","https:\u002F\u002Fgithub.com\u002Fgethamster\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002F.experimental\u002Feighty-twenty\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: eighty-twenty\ndescription: \"Get most of the value for a small fraction of the time, deliberately and with the blind spot named. Use before any expensive repeated activity — slow validation or build cycles, broad searches, large agent fan-outs, exhaustive captures, big sampling runs, reading a whole codebase — or whenever someone says this is too slow, speed it up, we need to go faster, or just give me the gist.\"\nversion: \"1.0.0\"\ncategory: \"Experience\"\n---\n\n# eighty-twenty\n\nMost of the value in an activity usually sits in a small part of it, and most of the cost usually\nsits somewhere else. When those two facts are true at once, you can buy most of the outcome for a\nfraction of the effort. When they are not, forcing it makes you fast and wrong.\n\nThis skill is the discipline for telling those cases apart, and for making the shortcut **safe by**\n**naming what it stops seeing.**\n\n*Project-agnostic. Copy this directory into any project's skills folder; it assumes no particular*\n*language, tool, runner or repo layout.*\n\n## Why it matters beyond speed\n\nA slow cycle does not just cost time — it changes behaviour. At seven minutes a cycle you batch\nfive changes together, and when it fails you no longer know which one did it. At forty seconds you\ncheck each one. **Cycle time is a correctness tool**, and that is the real argument for compressing\nit, not impatience.\n\n## Quickstart — applying this to a loop you have never seen\n\n1. Name the **unit of value** this activity produces (assertions, findings, frames, decisions).\n2. Get **per-item cost** — a `--json`\u002F`--verbose` report, `time` around each step, or a timestamped log.\n3. Sort by cost, print **cumulative %**, and add a **value-per-cost** column.\n4. Cut items that are expensive **and** low-yield. Keep expensive-but-unique ones.\n5. Apply the three levers below to what remains.\n6. Make the fast path **print what it skipped**, and prove it can still fail.\n\n## The move, in four steps\n\n### 1. Define value before you rank anything\n\n\"80% of the value\" is meaningless until you say what value *is* here: assertions, bugs found,\ncoverage, frames, decisions unblocked, words the reader will act on. Write it down first. Ranking\nagainst the wrong value is how people optimise a loop into uselessness.\n\n### 2. Measure the distribution — never estimate it\n\nIntuition about where cost and value sit is reliably wrong. Get real per-item numbers, sort, and\nprint the **cumulative percentage**. You are looking for a small head carrying most of the total.\n\nIf you cannot measure it cheaply, that is itself information: a distribution you cannot see is one\nyou should not cut blind.\n\n### 3. Rank by value-per-cost, not by cost\n\nThis is where most attempts go wrong. Expensive is not the same as wasteful. Cut what is\n**expensive AND low-yield**; an expensive step that is the only thing proving something important\nstays.\n\n> **Worked example** — one project's validation suite, 540 s per run. Value unit: assertions.\n> \n> ```\n> 135.3s   25% cum   step A    31 assertions     0.23 \u002Fsec\n>     76.7s   39% cum   step B    (whole-tree parse)\n>     73.8s   53% cum   step C    10 assertions     0.14 \u002Fsec\n>     68.4s   66% cum   step D    (perf measurement only)\n>     62.5s   77% cum   step E    15 assertions     0.24 \u002Fsec\n>     ...\n>      2.9s   97% cum   step R   177 assertions    61.0  \u002Fsec\n> ```\n> \n> Five steps were 77% of the clock and 56 assertions between them; one 2.9 s step carried 177.\n> A **400× spread in yield per second** — and it was invisible until the ratio column existed.\n> Sorting by total cost alone would have suggested optimising step A; the ratio says *delete*\n> step C and keep step R untouched. Result: **540 s → 47 s (91% cut), keeping 91% of assertions.**\n\n### 4. Name the blind spot in the output itself\n\nEvery shortcut creates one. An unnamed shortcut becomes a false claim the moment someone quotes the\nresult as proof of something it never checked. Put the disclosure where the **result** is read, not\nin documentation nobody opens:\n\n```\n\u003CPASS BANNER>  (FAST PATH — \u003Cwhat was skipped> not run;\n                run \u003Cfull command> before \u003Cthe gate this cannot clear>)\n```\n\nThat sentence is the safety mechanism, not decoration. In the worked example the skipped steps were\nthe only ones exercising behaviour over time — so the fast path could prove a *rule* was intact and\ncould not prove the *experience* was.\n\n## The three levers, in order of payoff\n\n1. **Do less work** — process only what changed. Parsing 88 files when 3 changed is 96% waste; that\n  single change took 77 s → ~0 s. Applies to linting, type-checking, compiling, capturing,\n  re-indexing. *Every incremental scheme has a soundness hole* — a changed interface breaks a file\n  that did not change. Accept it deliberately, state it, and keep the full pass honest.\n2. **Do it concurrently** — independent subprocesses are usually IO-bound from the driver's view, so\n  a small worker pool helps. 16 jobs went 117 s serial → 41 s wall at 3 workers. Only parallelise\n  things that genuinely do not interact: a shared **read-only** cache is fine, a writable one is\n  not. Stay conservative on shared or underpowered machines.\n3. **Do it later** — push slow-but-valuable work onto a gate that already exists (pre-merge, CI,\n  release, review) instead of paying it every iteration.\n\n## Two lanes, never one blurred one\n\nKeep the thorough version intact and add the fast one beside it. **Both must apply identical**\n**rules** — same failure classification, thresholds, minimums — or they will disagree and you will\nlearn to distrust both. Share the code that decides pass\u002Ffail rather than reimplementing it.\n\nThen verify, both directions:\n\n- the two lanes **agree on a clean tree**, and\n- the fast lane **goes red when you break something on purpose**.\n\nA fast lane that cannot fail is worse than none, because it is trusted.\n\n## Where this applies beyond test suites\n\n- **Searching \u002F reading:** read the 3 files that define the contract, not the 30 that use it.\n- **Agent fan-out:** 4 well-scoped agents usually beat 12 overlapping ones; the marginal agent\n  mostly re-finds what the others found.\n- **Capture \u002F screenshots \u002F recordings:** shoot the 2 moments that would show the defect, not 15\n  for symmetry.\n- **Simulation and sampling:** 4 seeds show a trend; 40 refine a number you are not ready to defend.\n- **Balance, tuning, and any parameter search:** change one dial, measure, repeat — a fast loop\n  beats a clever first guess.\n- **Builds and deploys:** stage the artifact that changed, not the world around it.\n- **Writing:** the reader acts on a few sentences; the rest is context they already have.\n\n## What 80\u002F20 must NOT touch\n\nThis section is the reason the skill is safe. Compress the loop **around** the work, never the\nevidence the work turns on.\n\n- **The thing you are actually trying to learn.** If the question is \"did this change how it\n  feels\u002Fperforms\u002Freads\", the long run *is* the deliverable. Sampling it answers a different\n  question quickly.\n- **The tail, when the tail is the point.** Bugs, security holes and edge cases live in the tail by\n  definition. A 20% sample of edge cases finds 20% of edge-case bugs — that is not 80\u002F20, it is\n  just less testing.\n- **Correctness gates and irreversible actions.** Merging, releasing, deleting, publishing,\n  migrating — anything with a blast radius. Speed is for the iteration, not for the commit.\n- **Verification of a claim you are about to make.** If you will assert it, check it properly.\n- **The last mile of a deliverable.** 80% of a shipped thing is not 80% shipped.\n\n## Knowing when to stop\n\nOptimising has its own diminishing return. Stop when the loop is no longer the thing slowing you\ndown — if the cycle is 47 s and you spend 10 minutes thinking between runs, further speed buys\nnothing. Spend the win on more iterations, not on more optimisation.\n\n## Exit\n\n- What \"value\" meant here, stated.\n- Before\u002Fafter numbers recorded — not impressions.\n- The fast path applies the same rules as the thorough one.\n- Its blind spot is named where the result is read.\n- Verified both ways: agrees when clean, fails when genuinely broken.\n- The thorough path still exists and is still what closes work.\n","# eighty-twenty\n\nMost of the value in an activity usually sits in a small part of it, and most of the cost usually\nsits somewhere else. When those two facts are true at once, you can buy most of the outcome for a\nfraction of the effort. When they are not, forcing it makes you fast and wrong.\n\nThis skill is the discipline for telling those cases apart, and for making the shortcut **safe by**\n**naming what it stops seeing.**\n\n*Project-agnostic. Copy this directory into any project's skills folder; it assumes no particular*\n*language, tool, runner or repo layout.*\n\n## Why it matters beyond speed\n\nA slow cycle does not just cost time — it changes behaviour. At seven minutes a cycle you batch\nfive changes together, and when it fails you no longer know which one did it. At forty seconds you\ncheck each one. **Cycle time is a correctness tool**, and that is the real argument for compressing\nit, not impatience.\n\n## Quickstart — applying this to a loop you have never seen\n\n1. Name the **unit of value** this activity produces (assertions, findings, frames, decisions).\n2. Get **per-item cost** — a `--json`\u002F`--verbose` report, `time` around each step, or a timestamped log.\n3. Sort by cost, print **cumulative %**, and add a **value-per-cost** column.\n4. Cut items that are expensive **and** low-yield. Keep expensive-but-unique ones.\n5. Apply the three levers below to what remains.\n6. Make the fast path **print what it skipped**, and prove it can still fail.\n\n## The move, in four steps\n\n### 1. Define value before you rank anything\n\n\"80% of the value\" is meaningless until you say what value *is* here: assertions, bugs found,\ncoverage, frames, decisions unblocked, words the reader will act on. Write it down first. Ranking\nagainst the wrong value is how people optimise a loop into uselessness.\n\n### 2. Measure the distribution — never estimate it\n\nIntuition about where cost and value sit is reliably wrong. Get real per-item numbers, sort, and\nprint the **cumulative percentage**. You are looking for a small head carrying most of the total.\n\nIf you cannot measure it cheaply, that is itself information: a distribution you cannot see is one\nyou should not cut blind.\n\n### 3. Rank by value-per-cost, not by cost\n\nThis is where most attempts go wrong. Expensive is not the same as wasteful. Cut what is\n**expensive AND low-yield**; an expensive step that is the only thing proving something important\nstays.\n\n> **Worked example** — one project's validation suite, 540 s per run. Value unit: assertions.\n> \n> ```\n> 135.3s   25% cum   step A    31 assertions     0.23 \u002Fsec\n>     76.7s   39% cum   step B    (whole-tree parse)\n>     73.8s   53% cum   step C    10 assertions     0.14 \u002Fsec\n>     68.4s   66% cum   step D    (perf measurement only)\n>     62.5s   77% cum   step E    15 assertions     0.24 \u002Fsec\n>     ...\n>      2.9s   97% cum   step R   177 assertions    61.0  \u002Fsec\n> ```\n> \n> Five steps were 77% of the clock and 56 assertions between them; one 2.9 s step carried 177.\n> A **400× spread in yield per second** — and it was invisible until the ratio column existed.\n> Sorting by total cost alone would have suggested optimising step A; the ratio says *delete*\n> step C and keep step R untouched. Result: **540 s → 47 s (91% cut), keeping 91% of assertions.**\n\n### 4. Name the blind spot in the output itself\n\nEvery shortcut creates one. An unnamed shortcut becomes a false claim the moment someone quotes the\nresult as proof of something it never checked. Put the disclosure where the **result** is read, not\nin documentation nobody opens:\n\n```\n\u003CPASS BANNER>  (FAST PATH — \u003Cwhat was skipped> not run;\n                run \u003Cfull command> before \u003Cthe gate this cannot clear>)\n```\n\nThat sentence is the safety mechanism, not decoration. In the worked example the skipped steps were\nthe only ones exercising behaviour over time — so the fast path could prove a *rule* was intact and\ncould not prove the *experience* was.\n\n## The three levers, in order of payoff\n\n1. **Do less work** — process only what changed. Parsing 88 files when 3 changed is 96% waste; that\n  single change took 77 s → ~0 s. Applies to linting, type-checking, compiling, capturing,\n  re-indexing. *Every incremental scheme has a soundness hole* — a changed interface breaks a file\n  that did not change. Accept it deliberately, state it, and keep the full pass honest.\n2. **Do it concurrently** — independent subprocesses are usually IO-bound from the driver's view, so\n  a small worker pool helps. 16 jobs went 117 s serial → 41 s wall at 3 workers. Only parallelise\n  things that genuinely do not interact: a shared **read-only** cache is fine, a writable one is\n  not. Stay conservative on shared or underpowered machines.\n3. **Do it later** — push slow-but-valuable work onto a gate that already exists (pre-merge, CI,\n  release, review) instead of paying it every iteration.\n\n## Two lanes, never one blurred one\n\nKeep the thorough version intact and add the fast one beside it. **Both must apply identical**\n**rules** — same failure classification, thresholds, minimums — or they will disagree and you will\nlearn to distrust both. Share the code that decides pass\u002Ffail rather than reimplementing it.\n\nThen verify, both directions:\n\n- the two lanes **agree on a clean tree**, and\n- the fast lane **goes red when you break something on purpose**.\n\nA fast lane that cannot fail is worse than none, because it is trusted.\n\n## Where this applies beyond test suites\n\n- **Searching \u002F reading:** read the 3 files that define the contract, not the 30 that use it.\n- **Agent fan-out:** 4 well-scoped agents usually beat 12 overlapping ones; the marginal agent\n  mostly re-finds what the others found.\n- **Capture \u002F screenshots \u002F recordings:** shoot the 2 moments that would show the defect, not 15\n  for symmetry.\n- **Simulation and sampling:** 4 seeds show a trend; 40 refine a number you are not ready to defend.\n- **Balance, tuning, and any parameter search:** change one dial, measure, repeat — a fast loop\n  beats a clever first guess.\n- **Builds and deploys:** stage the artifact that changed, not the world around it.\n- **Writing:** the reader acts on a few sentences; the rest is context they already have.\n\n## What 80\u002F20 must NOT touch\n\nThis section is the reason the skill is safe. Compress the loop **around** the work, never the\nevidence the work turns on.\n\n- **The thing you are actually trying to learn.** If the question is \"did this change how it\n  feels\u002Fperforms\u002Freads\", the long run *is* the deliverable. Sampling it answers a different\n  question quickly.\n- **The tail, when the tail is the point.** Bugs, security holes and edge cases live in the tail by\n  definition. A 20% sample of edge cases finds 20% of edge-case bugs — that is not 80\u002F20, it is\n  just less testing.\n- **Correctness gates and irreversible actions.** Merging, releasing, deleting, publishing,\n  migrating — anything with a blast radius. Speed is for the iteration, not for the commit.\n- **Verification of a claim you are about to make.** If you will assert it, check it properly.\n- **The last mile of a deliverable.** 80% of a shipped thing is not 80% shipped.\n\n## Knowing when to stop\n\nOptimising has its own diminishing return. Stop when the loop is no longer the thing slowing you\ndown — if the cycle is 47 s and you spend 10 minutes thinking between runs, further speed buys\nnothing. Spend the win on more iterations, not on more optimisation.\n\n## Exit\n\n- What \"value\" meant here, stated.\n- Before\u002Fafter numbers recorded — not impressions.\n- The fast path applies the same rules as the thorough one.\n- Its blind spot is named where the result is read.\n- Verified both ways: agrees when clean, fails when genuinely broken.\n- The thorough path still exists and is still what closes work.\n",{"name":5,"version":24,"category":25,"description":7},"1.0.0","Experience","main","fe3bb1476c3ec23d8a366eac0fa8cd9f0852360a","skills\u002F.experimental\u002Feighty-twenty",0,"https:\u002F\u002Fgithub.com\u002Fgethamster\u002Fskills","https:\u002F\u002Fgithub.com\u002Fgethamster","https:\u002F\u002Ftryhamster.com","We're building the AI-native future of work, unlocking the promised acceleration for your whole team, from idea to impact.","2026-11-05 11:33:33",{"source":8,"skillName":5,"commands":36,"prompt":46,"zipAvailable":47},[37,40,43],{"tool":38,"command":39},"npx","npx skills add gethamster\u002Fskills --skill eighty-twenty",{"tool":41,"command":42},"bunx","bunx skills add gethamster\u002Fskills --skill eighty-twenty",{"tool":44,"command":45},"pnpm","pnpm dlx skills add gethamster\u002Fskills --skill eighty-twenty","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fgethamster\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002F.experimental\u002Feighty-twenty\u002FSKILL.md",true,[],[50,57,64,71,78,85],{"skillNo":51,"name":52,"nameZh":6,"description":53,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":54,"categories":55,"keywords":56,"publishedAt":19},"01M7B5MWYW0KE251276S56KQDZ","writing-waterfall-requirements-documents","Write a waterfall requirements document: elicit, structure and review testable, traceable requirements, then baseline the specification before design.","https:\u002F\u002Fgithub.com\u002Fgethamster\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fwriting-waterfall-requirements-documents\u002FSKILL.md",[],[],{"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},"01M7B5MRCXWWPKREPB1CHQ7H90","writing-vision-and-values-statements","Write a V2MOM vision a reader can test and a ranked list of values that settles real tradeoffs, so the rest of the document has a clear anchor.","https:\u002F\u002Fgithub.com\u002Fgethamster\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fwriting-vision-and-values-statements\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},"01M7B5MKH2CH22YZAHQW3080VQ","writing-system-prompts-for-claude","Write Claude prompts and CLAUDE.md files that supply exact inputs, scope the task, and define done so the first attempt can be checked.","https:\u002F\u002Fgithub.com\u002Fgethamster\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fwriting-system-prompts-for-claude\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},"01M7B5ME1Y3X9RNRKP5HKHKACZ","writing-page-specific-website-copy","Website copywriting by page type: write homepage, landing page, pricing page and feature page copy that fits each page's visitor and single job.","https:\u002F\u002Fgithub.com\u002Fgethamster\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fwriting-page-specific-website-copy\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},"01M7B5M8MQNM8MN8PYH6BVWJ2B","writing-internal-press-releases","How to write an internal press release: the one-page working backwards press release that opens a PR\u002FFAQ document before anything is built.","https:\u002F\u002Fgithub.com\u002Fgethamster\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fwriting-internal-press-releases\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},"01M7B5M3P3YH1XF37R8PWXXSBK","writing-email-sequences-that-sell","Email sequence copywriting for nurture and sales: give each email one job, write clear subject lines, and close with one specific call to action.","https:\u002F\u002Fgithub.com\u002Fgethamster\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fwriting-email-sequences-that-sell\u002FSKILL.md",[],[]]