[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6CR4DF5GM8VJSWD8YJGKEXD":3,"skill-same:01M6CR4DF5GM8VJSWD8YJGKEXD":102},{"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":35,"linkOnly":36,"markdownContent":37,"markdownBody":38,"frontMatter":39,"refName":40,"commitSha":41,"skillDirectory":42,"forksCount":14,"repositoryHtmlUrl":43,"repositoryHomepageUrl":6,"ownerHtmlUrl":44,"ownerBlogUrl":45,"ownerDescription":46,"updatedAt":47,"install":48,"resources":61},"01M6CR4DF5GM8VJSWD8YJGKEXD","hitl",null,"A human is looking at the work right now and talking while they look. Re-read the tree before answering, treat every observation as a finding to be captured rather than a task to start, and do not touch anything until they say go. Use for \u002Fhitl, and whenever the user is walking through recent changes, reviewing a diff, checking a running service, or firing off several reactions in a row — 'what is this', 'why did that change', 'that looks wrong', 'hold on'. Ends by proposing one plan for approval, never by having quietly acted.","maxjbarfuss\u002Fmax_llm","maxjbarfuss","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F13053317?v=4","01M6CR3EGHV8C2C90PGAQEXZ27","Max Barfuss","01M6CR3EN0FXE8RZXAXCT7DWJG",0,"MIT","https:\u002F\u002Fgithub.com\u002Fmaxjbarfuss\u002Fmax_llm\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fhitl\u002FSKILL.md",[],[5,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34],"approval","task","do","look","say","never","diff","use","go","user","read","work","plan","why","start","change","2026-11-10 10:12:30",false,"---\nname: hitl\ndescription: \"A human is looking at the work right now and talking while they look. Re-read the tree before answering, treat every observation as a finding to be captured rather than a task to start, and do not touch anything until they say go. Use for \u002Fhitl, and whenever the user is walking through recent changes, reviewing a diff, checking a running service, or firing off several reactions in a row — 'what is this', 'why did that change', 'that looks wrong', 'hold on'. Ends by proposing one plan for approval, never by having quietly acted.\"\nlicense: MIT\n---\n\n# \u002Fhitl — a person is in the loop, and they are still talking\n\n*Vendor-neutral and machine-wide: this lives in `~\u002F.agents\u002Fskills\u002F`, the\ncross-tool location Codex, Grok and Kilo read themselves; Claude Code gets a\none-line pointer to it. It is written to be followed by any coding\nagent — Codex, Grok, Claude, Kilo, Cursor, Gemini, Copilot — or by a person. A\nproject may append its own section below the end marker: the commands that\nre-read its state, or the place its deferred findings are expected to land.*\n\nThe user is reviewing. They are looking at a diff, a running service, a screen\nof output, or a pile of recent commits, and they are reacting **as they go**.\nThat is the situation this skill is for, and it has two failure modes that are\nboth expensive.\n\n**Failure one: you act on the first thing they say.** They point at something\nodd, you start fixing it, and by the time they have finished looking you have\nedited three files they had not got to yet, abandoned the thing they actually\ncalled you in for, and turned a five-minute review into a rabbit hole they now\nhave to supervise.\n\n**Failure two: you let their observations evaporate.** They notice four real\nproblems while scrolling, you answer each one conversationally, and then the\nsession moves on. The findings were the valuable part of the review and none of\nthem survived it.\n\n**So: capture everything, act on nothing, until they say go.**\n\n## Triggers\n\n`\u002Fhitl`, and the shape of the conversation itself: the user is walking through\nrecent work, pasting output, asking \"what is this\", \"why did that change\",\n\"that looks wrong\", \"hold on\", \"wait\" — several reactions in a row, without a\nclear instruction to change anything.\n\nAmbiguity resolves toward capture. If you cannot tell whether something is an\ninstruction or an observation, it is an observation. Ask at the end, not now.\n\n## 1. Re-read before you answer anything\n\nYou are probably stale. Other sessions commit, the user edits in their editor,\nservices restart. Answering a review question from what you remember is how you\nconfidently describe code that no longer exists.\n\nBefore the first substantive answer, and again whenever the user implies\nsomething changed:\n\n- `git status --short` and `git log --oneline -5` — the branch may have moved\n  under you, possibly from another agent.\n- `git diff` \u002F `git diff --cached` for what is actually uncommitted.\n- Re-read the specific file being discussed. Not from memory. Not from a\n  summary you wrote earlier in the session.\n- If a running service is under discussion, confirm it is the one you think:\n  check the port actually answers and that it is your service and not something\n  else already on it.\n\n**Then make it current, before they look at anything.** Compare the process\nstart time to the newest commit touching the code it serves. If the process is\nolder, restart it and say so. Do not ask permission for this and do not defer\nit: a review of a service running last hour's build produces findings about\ncode that is already gone, and every one of them has to be re-derived later.\n\nThis is the single most repeated failure of this skill in practice. The\nsymptom is always the same — the agent *reports* the staleness as a finding and\ncarries on reviewing anyway. Reporting it is not enough. Making the running\nservice match `HEAD` is a precondition of the review, not an observation\nwithin it.\n\n**Verify the restart took.** A restart that silently failed is worse than no\nrestart, because you now believe the opposite of what is true. On a single-port\nservice the old process keeps the port and the new one exits; you read the old\nbuild's output and conclude the change is missing. So: stop the process *that\nholds the port* (find it by port, not by matching a command name — a pattern\nlike `pkill -f \"myapp serve\"` also matches your own shell), confirm the port is\nfree, start one instance, and confirm the new build is answering. Measured\ncost of skipping this: three restarts in a row that never bound, and a\nconfident wrong conclusion drawn from each.\n\n**Then make the browser current too, if the review is of a web page.** This is\nthe half that gets missed, and it produces the most confusing possible\nsymptom: the user says the feature is not there while the server is serving it.\n\nA page reloads its HTML and reuses its scripts. If assets are referenced by a\nplain path — `\u002Fstatic\u002Fapp.js` with no hash and no `?v=` — a tab that already\nholds the module in memory can reuse it without asking the server anything, so\na shipped change simply does not run. Before a sitting:\n\n- check the response headers on the app's own JS and CSS. `cache-control:\n  no-cache` (revalidate before reuse) or a hashed\u002Fversioned filename means you\n  are safe. An ETag *alone* is not enough — it only helps once the browser\n  decides to ask.\n- if neither is present, fix it (one header on the static handler) rather than\n  telling the user to hard-reload. A workaround they have to remember every\n  time is a defect with extra steps.\n- either way, ask them to hard-reload once at the start. It costs a keystroke\n  and removes the whole question from the sitting.\n\n**And confirm the fix is visible, not merely served.** \"The bytes contain it\"\nis not the same claim as \"a person can see it\". Load the page the way they\nwill, at their window size, and check the element is on screen and not covered\n— a correct component painted underneath a button is indistinguishable from a\nmissing one.\n\nThe exception is a service the user has deliberately pinned to an old build,\nand they will have said so.\n\nState it plainly when the ground moved: *\"Two commits landed since I last\nlooked — `abc1234` and `def5678`, both touching this file. Restarted; the\nservice is now on `def5678`.\"* That single line prevents a long argument about\nwhat the code says.\n\n## 2. Every observation becomes a finding\n\nKeep a running list, in the conversation, updated as they talk. Each finding:\n\n- **What they observed**, in their words where possible.\n- **What you verified**, if anything — and mark it clearly when you have not\n  checked yet. A finding may legitimately sit unverified while the review\n  continues.\n- **Where it lives** — `file.py:42`, a commit, a service.\n- **Size**, roughly: a one-line fix, a contained change, or something that\n  needs its own session.\n\nRe-state the list when it grows past about three items, or when the user pauses.\nThat is the artifact of the review, and it is what makes it safe to say \"not\nnow\" to any individual item.\n\nFindings that turn out to be non-issues stay on the list, marked as checked and\nclear. The user asked about them for a reason and will otherwise ask again.\n\n## 3. Answer, but do not start\n\nWhile the review is running you may: read files, run read-only commands, check\nlogs, look at a database, run existing tests to confirm a claim, and explain\nwhat you find.\n\nYou may not: edit files, write new ones, commit, install anything, restart a\nservice for any reason but the staleness check in section 1, or start the fix for something they just pointed at — however small\nand however obvious. **\"That is a one-line fix\" is not permission to make it.**\n\nTwo exceptions, and they are narrow. Something actively destructive in progress\n(a bad command still running, data being overwritten) — say so immediately and\nloudly. And an explicit, unambiguous instruction to do a specific thing now\n(\"fix that typo, then keep going\") — do exactly that thing and return to\nreviewing.\n\nIf they interrupt the work you were doing before the review, that work is\npaused, not cancelled. Say what state you left it in, in one line, so nobody\nhas to reconstruct it later.\n\n## 4. Do not follow the interesting thread\n\nReviews generate tangents, and the tangents are often genuinely more\ninteresting than the thing being reviewed. That is exactly why they are\ndangerous: you will have a good reason each time.\n\nWhen something turns out to be deeper than it looked, it becomes a finding with\n\"needs its own session\" on it. You do not explore it now, even to \"just check\none thing\" — checking one thing is how the review ends.\n\n## 5. Converge — one plan, and wait\n\nWhen the user signals they are done looking (they stop reacting, or say\nsomething like \"okay\", \"so\", \"what do you think\"), propose **one** plan. Not a\nmenu of options, not a list of everything possible.\n\n- **Findings**, ordered by what you would do first, with the size of each.\n- **What you propose doing now** — usually a small subset, and often just one.\n- **What you propose deferring**, and where it will live so it is not lost: a\n  TODO, an issue, a note in the project's own docs. Say which.\n- **What you are not going to do**, if you think something they raised is not\n  worth acting on. Say why, once, without arguing.\n\nThen stop and wait for approval. The whole point of the skill is that the human\ndecides what happens next.\n\nIf several rounds of feedback contradict each other — and they will, because\nthey were reacting in real time — surface the conflict rather than silently\npicking: *\"Earlier you wanted X; the last comment implies Y. Which?\"*\n\n## 6. After approval\n\nOnce they approve, the skill is over — do the work normally. Two carry-overs:\n\n- Deferred findings must actually land somewhere durable before the session\n  ends. A finding that exists only in scrollback did not survive the review.\n- If the approved work is significant, `\u002Fhandoff` afterwards to sync docs and\n  commit.\n\n## What this is not\n\nNot a code review — you are not hunting for defects on your own initiative.\nNot an approval gate for routine work. And not a reason to be timid once\npermission is given: the caution is about *when*, not about *how much*.\n","# \u002Fhitl — a person is in the loop, and they are still talking\n\n*Vendor-neutral and machine-wide: this lives in `~\u002F.agents\u002Fskills\u002F`, the\ncross-tool location Codex, Grok and Kilo read themselves; Claude Code gets a\none-line pointer to it. It is written to be followed by any coding\nagent — Codex, Grok, Claude, Kilo, Cursor, Gemini, Copilot — or by a person. A\nproject may append its own section below the end marker: the commands that\nre-read its state, or the place its deferred findings are expected to land.*\n\nThe user is reviewing. They are looking at a diff, a running service, a screen\nof output, or a pile of recent commits, and they are reacting **as they go**.\nThat is the situation this skill is for, and it has two failure modes that are\nboth expensive.\n\n**Failure one: you act on the first thing they say.** They point at something\nodd, you start fixing it, and by the time they have finished looking you have\nedited three files they had not got to yet, abandoned the thing they actually\ncalled you in for, and turned a five-minute review into a rabbit hole they now\nhave to supervise.\n\n**Failure two: you let their observations evaporate.** They notice four real\nproblems while scrolling, you answer each one conversationally, and then the\nsession moves on. The findings were the valuable part of the review and none of\nthem survived it.\n\n**So: capture everything, act on nothing, until they say go.**\n\n## Triggers\n\n`\u002Fhitl`, and the shape of the conversation itself: the user is walking through\nrecent work, pasting output, asking \"what is this\", \"why did that change\",\n\"that looks wrong\", \"hold on\", \"wait\" — several reactions in a row, without a\nclear instruction to change anything.\n\nAmbiguity resolves toward capture. If you cannot tell whether something is an\ninstruction or an observation, it is an observation. Ask at the end, not now.\n\n## 1. Re-read before you answer anything\n\nYou are probably stale. Other sessions commit, the user edits in their editor,\nservices restart. Answering a review question from what you remember is how you\nconfidently describe code that no longer exists.\n\nBefore the first substantive answer, and again whenever the user implies\nsomething changed:\n\n- `git status --short` and `git log --oneline -5` — the branch may have moved\n  under you, possibly from another agent.\n- `git diff` \u002F `git diff --cached` for what is actually uncommitted.\n- Re-read the specific file being discussed. Not from memory. Not from a\n  summary you wrote earlier in the session.\n- If a running service is under discussion, confirm it is the one you think:\n  check the port actually answers and that it is your service and not something\n  else already on it.\n\n**Then make it current, before they look at anything.** Compare the process\nstart time to the newest commit touching the code it serves. If the process is\nolder, restart it and say so. Do not ask permission for this and do not defer\nit: a review of a service running last hour's build produces findings about\ncode that is already gone, and every one of them has to be re-derived later.\n\nThis is the single most repeated failure of this skill in practice. The\nsymptom is always the same — the agent *reports* the staleness as a finding and\ncarries on reviewing anyway. Reporting it is not enough. Making the running\nservice match `HEAD` is a precondition of the review, not an observation\nwithin it.\n\n**Verify the restart took.** A restart that silently failed is worse than no\nrestart, because you now believe the opposite of what is true. On a single-port\nservice the old process keeps the port and the new one exits; you read the old\nbuild's output and conclude the change is missing. So: stop the process *that\nholds the port* (find it by port, not by matching a command name — a pattern\nlike `pkill -f \"myapp serve\"` also matches your own shell), confirm the port is\nfree, start one instance, and confirm the new build is answering. Measured\ncost of skipping this: three restarts in a row that never bound, and a\nconfident wrong conclusion drawn from each.\n\n**Then make the browser current too, if the review is of a web page.** This is\nthe half that gets missed, and it produces the most confusing possible\nsymptom: the user says the feature is not there while the server is serving it.\n\nA page reloads its HTML and reuses its scripts. If assets are referenced by a\nplain path — `\u002Fstatic\u002Fapp.js` with no hash and no `?v=` — a tab that already\nholds the module in memory can reuse it without asking the server anything, so\na shipped change simply does not run. Before a sitting:\n\n- check the response headers on the app's own JS and CSS. `cache-control:\n  no-cache` (revalidate before reuse) or a hashed\u002Fversioned filename means you\n  are safe. An ETag *alone* is not enough — it only helps once the browser\n  decides to ask.\n- if neither is present, fix it (one header on the static handler) rather than\n  telling the user to hard-reload. A workaround they have to remember every\n  time is a defect with extra steps.\n- either way, ask them to hard-reload once at the start. It costs a keystroke\n  and removes the whole question from the sitting.\n\n**And confirm the fix is visible, not merely served.** \"The bytes contain it\"\nis not the same claim as \"a person can see it\". Load the page the way they\nwill, at their window size, and check the element is on screen and not covered\n— a correct component painted underneath a button is indistinguishable from a\nmissing one.\n\nThe exception is a service the user has deliberately pinned to an old build,\nand they will have said so.\n\nState it plainly when the ground moved: *\"Two commits landed since I last\nlooked — `abc1234` and `def5678`, both touching this file. Restarted; the\nservice is now on `def5678`.\"* That single line prevents a long argument about\nwhat the code says.\n\n## 2. Every observation becomes a finding\n\nKeep a running list, in the conversation, updated as they talk. Each finding:\n\n- **What they observed**, in their words where possible.\n- **What you verified**, if anything — and mark it clearly when you have not\n  checked yet. A finding may legitimately sit unverified while the review\n  continues.\n- **Where it lives** — `file.py:42`, a commit, a service.\n- **Size**, roughly: a one-line fix, a contained change, or something that\n  needs its own session.\n\nRe-state the list when it grows past about three items, or when the user pauses.\nThat is the artifact of the review, and it is what makes it safe to say \"not\nnow\" to any individual item.\n\nFindings that turn out to be non-issues stay on the list, marked as checked and\nclear. The user asked about them for a reason and will otherwise ask again.\n\n## 3. Answer, but do not start\n\nWhile the review is running you may: read files, run read-only commands, check\nlogs, look at a database, run existing tests to confirm a claim, and explain\nwhat you find.\n\nYou may not: edit files, write new ones, commit, install anything, restart a\nservice for any reason but the staleness check in section 1, or start the fix for something they just pointed at — however small\nand however obvious. **\"That is a one-line fix\" is not permission to make it.**\n\nTwo exceptions, and they are narrow. Something actively destructive in progress\n(a bad command still running, data being overwritten) — say so immediately and\nloudly. And an explicit, unambiguous instruction to do a specific thing now\n(\"fix that typo, then keep going\") — do exactly that thing and return to\nreviewing.\n\nIf they interrupt the work you were doing before the review, that work is\npaused, not cancelled. Say what state you left it in, in one line, so nobody\nhas to reconstruct it later.\n\n## 4. Do not follow the interesting thread\n\nReviews generate tangents, and the tangents are often genuinely more\ninteresting than the thing being reviewed. That is exactly why they are\ndangerous: you will have a good reason each time.\n\nWhen something turns out to be deeper than it looked, it becomes a finding with\n\"needs its own session\" on it. You do not explore it now, even to \"just check\none thing\" — checking one thing is how the review ends.\n\n## 5. Converge — one plan, and wait\n\nWhen the user signals they are done looking (they stop reacting, or say\nsomething like \"okay\", \"so\", \"what do you think\"), propose **one** plan. Not a\nmenu of options, not a list of everything possible.\n\n- **Findings**, ordered by what you would do first, with the size of each.\n- **What you propose doing now** — usually a small subset, and often just one.\n- **What you propose deferring**, and where it will live so it is not lost: a\n  TODO, an issue, a note in the project's own docs. Say which.\n- **What you are not going to do**, if you think something they raised is not\n  worth acting on. Say why, once, without arguing.\n\nThen stop and wait for approval. The whole point of the skill is that the human\ndecides what happens next.\n\nIf several rounds of feedback contradict each other — and they will, because\nthey were reacting in real time — surface the conflict rather than silently\npicking: *\"Earlier you wanted X; the last comment implies Y. Which?\"*\n\n## 6. After approval\n\nOnce they approve, the skill is over — do the work normally. Two carry-overs:\n\n- Deferred findings must actually land somewhere durable before the session\n  ends. A finding that exists only in scrollback did not survive the review.\n- If the approved work is significant, `\u002Fhandoff` afterwards to sync docs and\n  commit.\n\n## What this is not\n\nNot a code review — you are not hunting for defects on your own initiative.\nNot an approval gate for routine work. And not a reason to be timid once\npermission is given: the caution is about *when*, not about *how much*.\n",{"name":5,"license":15,"description":7},"main","16426a8647c4bf341798845ddf222a5b8c1ac861",".agents\u002Fskills\u002Fhitl","https:\u002F\u002Fgithub.com\u002Fmaxjbarfuss\u002Fmax_llm","https:\u002F\u002Fgithub.com\u002Fmaxjbarfuss","https:\u002F\u002Fwww.altavega.io","Software Engineer, AI, Robotics, Systems, Full-Stack","2026-09-29 21:59:21",{"source":8,"skillName":5,"commands":49,"prompt":59,"zipAvailable":60},[50,53,56],{"tool":51,"command":52},"npx","npx skills add maxjbarfuss\u002Fmax_llm --skill hitl",{"tool":54,"command":55},"bunx","bunx skills add maxjbarfuss\u002Fmax_llm --skill hitl",{"tool":57,"command":58},"pnpm","pnpm dlx skills add maxjbarfuss\u002Fmax_llm --skill hitl","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fmaxjbarfuss\u002Fmax_llm\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fhitl\u002FSKILL.md",true,[62,68,74,78,82,86,90,94,98],{"path":63,"resourceType":64,"resourceTypeDesc":65,"sizeBytes":66,"sourceHtmlUrl":67},"agents\u002Fobserver-loop.sh","SCRIPT","脚本",9876,"https:\u002F\u002Fgithub.com\u002Fendman100\u002Fskill-router-nexus\u002Fblob\u002Fmain\u002FKnowledge-and-Productivity\u002Fcontinuous-learning-v2\u002Fagents\u002Fobserver-loop.sh",{"path":69,"resourceType":70,"resourceTypeDesc":71,"sizeBytes":72,"sourceHtmlUrl":73},"agents\u002Fobserver.md","REFERENCE","参考文档",7396,"https:\u002F\u002Fgithub.com\u002Fendman100\u002Fskill-router-nexus\u002Fblob\u002Fmain\u002FKnowledge-and-Productivity\u002Fcontinuous-learning-v2\u002Fagents\u002Fobserver.md",{"path":75,"resourceType":64,"resourceTypeDesc":65,"sizeBytes":76,"sourceHtmlUrl":77},"agents\u002Fsession-guardian.sh",6392,"https:\u002F\u002Fgithub.com\u002Fendman100\u002Fskill-router-nexus\u002Fblob\u002Fmain\u002FKnowledge-and-Productivity\u002Fcontinuous-learning-v2\u002Fagents\u002Fsession-guardian.sh",{"path":79,"resourceType":64,"resourceTypeDesc":65,"sizeBytes":80,"sourceHtmlUrl":81},"agents\u002Fstart-observer.sh",7522,"https:\u002F\u002Fgithub.com\u002Fendman100\u002Fskill-router-nexus\u002Fblob\u002Fmain\u002FKnowledge-and-Productivity\u002Fcontinuous-learning-v2\u002Fagents\u002Fstart-observer.sh",{"path":83,"resourceType":70,"resourceTypeDesc":71,"sizeBytes":84,"sourceHtmlUrl":85},"config.json",135,"https:\u002F\u002Fgithub.com\u002Fendman100\u002Fskill-router-nexus\u002Fblob\u002Fmain\u002FKnowledge-and-Productivity\u002Fcontinuous-learning-v2\u002Fconfig.json",{"path":87,"resourceType":64,"resourceTypeDesc":65,"sizeBytes":88,"sourceHtmlUrl":89},"hooks\u002Fobserve.sh",16161,"https:\u002F\u002Fgithub.com\u002Fendman100\u002Fskill-router-nexus\u002Fblob\u002Fmain\u002FKnowledge-and-Productivity\u002Fcontinuous-learning-v2\u002Fhooks\u002Fobserve.sh",{"path":91,"resourceType":64,"resourceTypeDesc":65,"sizeBytes":92,"sourceHtmlUrl":93},"scripts\u002Fdetect-project.sh",7761,"https:\u002F\u002Fgithub.com\u002Fendman100\u002Fskill-router-nexus\u002Fblob\u002Fmain\u002FKnowledge-and-Productivity\u002Fcontinuous-learning-v2\u002Fscripts\u002Fdetect-project.sh",{"path":95,"resourceType":64,"resourceTypeDesc":65,"sizeBytes":96,"sourceHtmlUrl":97},"scripts\u002Finstinct-cli.py",57750,"https:\u002F\u002Fgithub.com\u002Fendman100\u002Fskill-router-nexus\u002Fblob\u002Fmain\u002FKnowledge-and-Productivity\u002Fcontinuous-learning-v2\u002Fscripts\u002Finstinct-cli.py",{"path":99,"resourceType":64,"resourceTypeDesc":65,"sizeBytes":100,"sourceHtmlUrl":101},"scripts\u002Ftest_parse_instinct.py",33106,"https:\u002F\u002Fgithub.com\u002Fendman100\u002Fskill-router-nexus\u002Fblob\u002Fmain\u002FKnowledge-and-Productivity\u002Fcontinuous-learning-v2\u002Fscripts\u002Ftest_parse_instinct.py",[103,119,142,156,170,185],{"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":107,"licenseFileUrl":6,"sourceHtmlUrl":108,"categories":109,"keywords":110,"publishedAt":35},"01M6CR6B20826Y0PPVXJQ2D0QJ","train","Launch a new max_llm training run on both GPUs (RTX 4090 + RTX 3090 Ti) through torchrun DDP, resolving the TOML config by name or path. Use only when the user explicitly asks to start, launch or begin a run — \"resume\", \"continue\" or \"keep training\" mean the existing run; takes an optional config name or path.","Apache-2.0","https:\u002F\u002Fgithub.com\u002Fmaxjbarfuss\u002Fmax_llm\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Ftrain\u002FSKILL.md",[],[105,111,112,113,114,26,28,115,33,116,117,118],"config","max","resume","only","continue","run","training","launch",{"skillNo":120,"name":121,"nameZh":6,"description":122,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":107,"licenseFileUrl":6,"sourceHtmlUrl":123,"categories":124,"keywords":125,"publishedAt":35},"01M6CR67YE4MW1R46KB8NV85ZT","serve-live","Before a person looks at a running service, make sure what is running IS the code in the checkout. Restart the services your change actually touched — not every service — and verify each is serving the current commit rather than merely answering; check the remaining dependencies are up without bouncing them. Use whenever you finish work a person will view in a browser or client, whenever you say a page or service is ready, after any commit that changes served code, and when asked to 'restart', 'refresh the services', 'is it live', or 'serve the latest'. Also covers waiting for lazy warm-up before calling a service ready.","https:\u002F\u002Fgithub.com\u002Fmaxjbarfuss\u002Fmax_llm\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fserve-live\u002FSKILL.md",[],[126,127,128,129,130,131,30,132,133,34,134,135,26,136,137,138,23,139,140,141],"serve live","services","verify","page","browser","code","check","commit","view","refresh","dependencies","calling","restart","finish","up","ready",{"skillNo":143,"name":144,"nameZh":6,"description":145,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":107,"licenseFileUrl":6,"sourceHtmlUrl":146,"categories":147,"keywords":148,"publishedAt":35},"01M6CR665ZD6GG1SQKD0XS4GJS","review","Review a local change, branch, pull request, or stack for correctness, regressions, security, tests, and unnecessary complexity. Use when asked to review rather than implement.","https:\u002F\u002Fgithub.com\u002Fmaxjbarfuss\u002Fmax_llm\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Freview\u002FSKILL.md",[],[144,149,34,26,150,151,152,153,154,155],"complexity","stack","pull","pull request","implement","security","branch",{"skillNo":157,"name":158,"nameZh":6,"description":159,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":107,"licenseFileUrl":6,"sourceHtmlUrl":160,"categories":161,"keywords":162,"publishedAt":35},"01M6CR64DYZJFD4JM1QCKMCGFN","regression-coverage","Reshape a slow or brittle test suite around observable user journeys and published contracts. Use when pruning tests, repairing weak smoke coverage, or addressing repeated feature regressions.","https:\u002F\u002Fgithub.com\u002Fmaxjbarfuss\u002Fmax_llm\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fregression-coverage\u002FSKILL.md",[],[163,164,165,166,167,168,26,28,169],"regression","coverage","regression coverage","smoke","contracts","test suite","test",{"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":107,"licenseFileUrl":6,"sourceHtmlUrl":174,"categories":175,"keywords":176,"publishedAt":35},"01M6CR62A9C4BB3YR6XZPVBKY3","pr-babysit","Monitor selected pull requests and report or safely act on review, CI, mergeability, and stack state.","https:\u002F\u002Fgithub.com\u002Fmaxjbarfuss\u002Fmax_llm\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fpr-babysit\u002FSKILL.md",[],[177,178,179,180,181,144,182,150,183,151,184],"pr babysit","babysit","pr","monitor","pull requests","report","ci","state",{"skillNo":186,"name":187,"nameZh":6,"description":188,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":107,"licenseFileUrl":6,"sourceHtmlUrl":189,"categories":190,"keywords":191,"publishedAt":35},"01M6CR5ZWFQFGJ6QY17DN0Q3NF","keel","Use a project's local keel code graph to assess change impact, find callers and reuse candidates, validate a plan, or check an edit. Applies only in a repository with a .keel\u002F directory; skip it for prose-only work.","https:\u002F\u002Fgithub.com\u002Fmaxjbarfuss\u002Fmax_llm\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fkeel\u002FSKILL.md",[],[187,192,193,114,26,194,132,30,31,195,196,131,197,34,198],"find","prose","edit","repository","graph","validate","directory"]