[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M77XFWRYH97KGTKCF8DF7PQR":3,"skill-same:01M77XFWRYH97KGTKCF8DF7PQR":53},{"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":31,"ownerHtmlUrl":32,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":33,"install":34,"resources":46},"01M77XFWRYH97KGTKCF8DF7PQR","tune",null,"Find and fix real defects in a skill, AGENTS.md, or CLAUDE.md, then prove the description still routes. Runs the bundled token-spending runner, so it starts only on the human's explicit instruction.","StartupBros-com\u002Fhov-stack","StartupBros-com","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F262958016?v=4","01M77WY8AH50FY939DTGZYHVE9","StartupBros.com","01M77WY8B5Q2D7VSS0P6Y0JX6E",1,"MIT","https:\u002F\u002Fgithub.com\u002FStartupBros-com\u002Fhov-stack\u002Fblob\u002Fmain\u002Fplugins\u002Fskill-tuner\u002Fskills\u002Ftune\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: tune\ndescription: Find and fix real defects in a skill, AGENTS.md, or CLAUDE.md, then prove the description still routes. Runs the bundled token-spending runner, so it starts only on the human's explicit instruction.\nargument-hint: \u003Cpath to SKILL.md \u002F AGENTS.md \u002F CLAUDE.md>\ndisable-model-invocation: true\n---\n\nTune the document at `$ARGUMENTS`.\n\nYou are running the loop the human just asked for, so the runner may spend\ntokens. Report the estimate before the first call and the actual after.\n\n`${CLAUDE_PLUGIN_ROOT}` below is Claude Code's expansion for the plugin's\ninstall root. In a client that does not expand it, resolve it yourself: the\nplugin root is the directory three levels up from this SKILL.md — the one\nthat contains `skills\u002F`. The command blocks below use `\u003Ctarget-path>` as a\nslot, not a variable: replace it with the document path the human gave,\nkeeping the surrounding double quotes so a path with spaces stays one\nargument. Never pass an unexpanded placeholder through to a shell.\n\n## 1. Read it\n\nRead `$ARGUMENTS` in full, and the bundled doctrine at\n`${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Fskill-tuner\u002FSKILL.md`. You need both: the probe\ntells you *what* is wrong, the doctrine tells you *what good looks like* when\nyou fix it.\n\n## 2. Probe it\n\n```\npython3 ${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Fskill-tuner\u002Fscripts\u002Ftune.py probe \\\n  --target \"\u003Ctarget-path>\" --run-id tune-$(date +%s) --yes --budget-usd 3 --verify-trials 3\n```\n\nEvery finding it reports has already survived three independent skeptics in\nfresh contexts, plus a code-level check that the quoted text genuinely appears\nin the document. **Treat confirmed findings as real and act on them.** They are\nnot suggestions to evaluate; the evaluating already happened.\n\nRead the run's `report.md` for each finding's quote and proposed fix.\n\nRefuted findings are listed too, with the mode that killed them\n(`verifier` = the rule did not apply; `quote_not_found` = the probe invented\nits evidence). Treat refuted findings as closed: their one remaining job is\nthe report's refuted line, where the human sees what the probe got wrong.\n\n## 3. Fix what it found\n\nApply each confirmed finding. The doctrine's rules are the target shape —\npositive phrasing over prohibition, one trigger per branch, checkable\ncompletion bounds, one meaning in one place.\n\nFix the defect, not the sentence containing it. A finding about a vague\ncompletion bound is asking for a bound that can be checked, not for the word\n\"clearly\" to be deleted.\n\nBefore each edit, one question: would a competent agent with the current\ninstructions still be expected to fail this way? If yes, there is a gap. If\nno, defer — the text already required the behavior. When nothing clears that\nbar, opening no change is the correct outcome, and the report says so per\nfinding; a speculative change is worse than none.\n\nOne check before acting: the skeptic panel verifies text against text — it\ncannot see the filesystem. A finding that asserts an environment fact (a\npath resolves, a file exists, a config value holds) needs that fact checked\non disk first. When the claim turns out false, the ambiguity that made the\ntext read as wrong is usually still real: fix that, not the false claim.\n\n## 4. Re-probe\n\nRun step 2 again against the edited file.\n\nDone when **either** the re-probe confirms nothing, **or** it confirms only\nfindings disjoint from every earlier pass. A finding that overlaps an earlier\nfix means that fix has not landed — address it and re-probe again; a fix that\ntrades one confirmed defect for another has not landed either.\n\nDisjoint fresh findings at that point are the instrument's floor, not your\nfailure: the probe surfaces ~2 genuine defects per pass on almost any\ndocument, including already-audited ones (its measured marginal-value\nproperty), so a rich document never re-probes to zero — one loop here ran\neight passes before that was accepted. Apply the fresh ones that are cheap\nand clearly right, and report the rest as residuals.\n\n## 5. If you changed the description, prove it still routes\n\nA reworded description is a routing change, and routing is the one thing here\nthat can be measured properly on a single document. Build a battery config\nfollowing `${CLAUDE_PLUGIN_ROOT}\u002Fconfigs\u002Freceipts-routing-001-neutral.json`, with the original\ndescription as `original` and yours as `pruned`, and keep its `\"neutral_ids\": true`: the\nbattery lists each skill's id beside its description, and an id routes on its own, so with\nthe id visible a parity verdict can be carried by the name rather than by the description\nyou are testing (the doctrine's routing-parity rule records the measurement). Then:\n\n```\npython3 ${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Fskill-tuner\u002Fscripts\u002Ftune.py routing-parity \\\n  --config \u003Cyour-config> --run-id route-$(date +%s) --yes --budget-usd 3\n```\n\nKeep the new description only on a `land` verdict. On `refuse`, restore the\noriginal and say which prompts it lost.\n\n## What you may not claim\n\nYou probed one document. **A before\u002Fafter count on one document is not\nevidence of improvement** — this project has the receipts for that, having\nrecorded LOST, WON and WORSE for one question at n=1, 6 and 16 respectively.\n\nSo: report the defects you fixed and the routing verdict, both of which are\nreal. Do not report that the document is \"better\" as a measured fact, and do\nnot compare finding counts across the two probe runs as if the difference\nmeant something.\n\nMeasuring an actual improvement takes a paired comparison across many\ndocuments — `tune.py compare`, which needs at least 3 and realistically many\nmore. Reach for it when changing something that affects a whole suite, such as\na shared convention or a doctrine, and never to grade a single edit.\n\n## Report\n\n- Confirmed defects found, and what you changed for each\n- Anything refuted, so the human sees what the probe got wrong\n- The routing verdict, if the description moved\n- Actual spend, from the run output\n","Tune the document at `$ARGUMENTS`.\n\nYou are running the loop the human just asked for, so the runner may spend\ntokens. Report the estimate before the first call and the actual after.\n\n`${CLAUDE_PLUGIN_ROOT}` below is Claude Code's expansion for the plugin's\ninstall root. In a client that does not expand it, resolve it yourself: the\nplugin root is the directory three levels up from this SKILL.md — the one\nthat contains `skills\u002F`. The command blocks below use `\u003Ctarget-path>` as a\nslot, not a variable: replace it with the document path the human gave,\nkeeping the surrounding double quotes so a path with spaces stays one\nargument. Never pass an unexpanded placeholder through to a shell.\n\n## 1. Read it\n\nRead `$ARGUMENTS` in full, and the bundled doctrine at\n`${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Fskill-tuner\u002FSKILL.md`. You need both: the probe\ntells you *what* is wrong, the doctrine tells you *what good looks like* when\nyou fix it.\n\n## 2. Probe it\n\n```\npython3 ${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Fskill-tuner\u002Fscripts\u002Ftune.py probe \\\n  --target \"\u003Ctarget-path>\" --run-id tune-$(date +%s) --yes --budget-usd 3 --verify-trials 3\n```\n\nEvery finding it reports has already survived three independent skeptics in\nfresh contexts, plus a code-level check that the quoted text genuinely appears\nin the document. **Treat confirmed findings as real and act on them.** They are\nnot suggestions to evaluate; the evaluating already happened.\n\nRead the run's `report.md` for each finding's quote and proposed fix.\n\nRefuted findings are listed too, with the mode that killed them\n(`verifier` = the rule did not apply; `quote_not_found` = the probe invented\nits evidence). Treat refuted findings as closed: their one remaining job is\nthe report's refuted line, where the human sees what the probe got wrong.\n\n## 3. Fix what it found\n\nApply each confirmed finding. The doctrine's rules are the target shape —\npositive phrasing over prohibition, one trigger per branch, checkable\ncompletion bounds, one meaning in one place.\n\nFix the defect, not the sentence containing it. A finding about a vague\ncompletion bound is asking for a bound that can be checked, not for the word\n\"clearly\" to be deleted.\n\nBefore each edit, one question: would a competent agent with the current\ninstructions still be expected to fail this way? If yes, there is a gap. If\nno, defer — the text already required the behavior. When nothing clears that\nbar, opening no change is the correct outcome, and the report says so per\nfinding; a speculative change is worse than none.\n\nOne check before acting: the skeptic panel verifies text against text — it\ncannot see the filesystem. A finding that asserts an environment fact (a\npath resolves, a file exists, a config value holds) needs that fact checked\non disk first. When the claim turns out false, the ambiguity that made the\ntext read as wrong is usually still real: fix that, not the false claim.\n\n## 4. Re-probe\n\nRun step 2 again against the edited file.\n\nDone when **either** the re-probe confirms nothing, **or** it confirms only\nfindings disjoint from every earlier pass. A finding that overlaps an earlier\nfix means that fix has not landed — address it and re-probe again; a fix that\ntrades one confirmed defect for another has not landed either.\n\nDisjoint fresh findings at that point are the instrument's floor, not your\nfailure: the probe surfaces ~2 genuine defects per pass on almost any\ndocument, including already-audited ones (its measured marginal-value\nproperty), so a rich document never re-probes to zero — one loop here ran\neight passes before that was accepted. Apply the fresh ones that are cheap\nand clearly right, and report the rest as residuals.\n\n## 5. If you changed the description, prove it still routes\n\nA reworded description is a routing change, and routing is the one thing here\nthat can be measured properly on a single document. Build a battery config\nfollowing `${CLAUDE_PLUGIN_ROOT}\u002Fconfigs\u002Freceipts-routing-001-neutral.json`, with the original\ndescription as `original` and yours as `pruned`, and keep its `\"neutral_ids\": true`: the\nbattery lists each skill's id beside its description, and an id routes on its own, so with\nthe id visible a parity verdict can be carried by the name rather than by the description\nyou are testing (the doctrine's routing-parity rule records the measurement). Then:\n\n```\npython3 ${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Fskill-tuner\u002Fscripts\u002Ftune.py routing-parity \\\n  --config \u003Cyour-config> --run-id route-$(date +%s) --yes --budget-usd 3\n```\n\nKeep the new description only on a `land` verdict. On `refuse`, restore the\noriginal and say which prompts it lost.\n\n## What you may not claim\n\nYou probed one document. **A before\u002Fafter count on one document is not\nevidence of improvement** — this project has the receipts for that, having\nrecorded LOST, WON and WORSE for one question at n=1, 6 and 16 respectively.\n\nSo: report the defects you fixed and the routing verdict, both of which are\nreal. Do not report that the document is \"better\" as a measured fact, and do\nnot compare finding counts across the two probe runs as if the difference\nmeant something.\n\nMeasuring an actual improvement takes a paired comparison across many\ndocuments — `tune.py compare`, which needs at least 3 and realistically many\nmore. Reach for it when changing something that affects a whole suite, such as\na shared convention or a doctrine, and never to grade a single edit.\n\n## Report\n\n- Confirmed defects found, and what you changed for each\n- Anything refuted, so the human sees what the probe got wrong\n- The routing verdict, if the description moved\n- Actual spend, from the run output\n",{"name":5,"description":7,"argument-hint":24,"disable-model-invocation":25},"\u003Cpath to SKILL.md \u002F AGENTS.md \u002F CLAUDE.md>",true,"main","5889e24531cb4e925523afbc48cbe1f42b5e4cd7","plugins\u002Fskill-tuner\u002Fskills\u002Ftune",0,"https:\u002F\u002Fgithub.com\u002FStartupBros-com\u002Fhov-stack","https:\u002F\u002Fhouseofvibe.ai","https:\u002F\u002Fgithub.com\u002FStartupBros-com","2026-11-15 05:18:00",{"source":8,"skillName":5,"commands":35,"prompt":45,"zipAvailable":25},[36,39,42],{"tool":37,"command":38},"npx","npx skills add StartupBros-com\u002Fhov-stack --skill tune",{"tool":40,"command":41},"bunx","bunx skills add StartupBros-com\u002Fhov-stack --skill tune",{"tool":43,"command":44},"pnpm","pnpm dlx skills add StartupBros-com\u002Fhov-stack --skill tune","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FStartupBros-com\u002Fhov-stack\u002Fblob\u002Fmain\u002Fplugins\u002Fskill-tuner\u002Fskills\u002Ftune\u002FSKILL.md",[47],{"path":48,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":51,"sourceHtmlUrl":52},"agents\u002Fopenai.yaml","REFERENCE","参考文档",205,"https:\u002F\u002Fgithub.com\u002FStartupBros-com\u002Fhov-stack\u002Fblob\u002Fmain\u002Fplugins\u002Fskill-tuner\u002Fskills\u002Ftune\u002Fagents\u002Fopenai.yaml",[54,62,70,78,83,90],{"skillNo":55,"name":56,"nameZh":6,"description":57,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":58,"sourceHtmlUrl":59,"categories":60,"keywords":61,"publishedAt":19},"01M77XGSFPK8Q1XBHYTC1TF2MV","review-adversarial-first","Operator-invoked adversarial review: 3 refute-by-default lenses plus a deterministic judge against a PR, branch, or path, then a decision on whether \u002Fce-code-review's topic fan-out runs. Off the default merge path since the 2026-10-17 merge-and-end ruling; run only when asked for by name.","https:\u002F\u002Fgithub.com\u002FStartupBros-com\u002Fhov-stack\u002Fblob\u002Fmain\u002Fskills-stack\u002Freview-adversarial-first\u002FLICENSE","https:\u002F\u002Fgithub.com\u002FStartupBros-com\u002Fhov-stack\u002Fblob\u002Fmain\u002Fskills-stack\u002Freview-adversarial-first\u002FSKILL.md",[],[],{"skillNo":63,"name":64,"nameZh":6,"description":65,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":66,"sourceHtmlUrl":67,"categories":68,"keywords":69,"publishedAt":19},"01M77XGNSF0ARDTK9P3EF2S4E4","memory-mine","Run one deterministic, budgeted directed-capture batch for an explicit Claude Code project store. Use manual workspace mode only when cross-agent cass sessions are intentionally in scope.","https:\u002F\u002Fgithub.com\u002FStartupBros-com\u002Fhov-stack\u002Fblob\u002Fmain\u002Fskills-stack\u002Fmemory-mine\u002FLICENSE","https:\u002F\u002Fgithub.com\u002FStartupBros-com\u002Fhov-stack\u002Fblob\u002Fmain\u002Fskills-stack\u002Fmemory-mine\u002FSKILL.md",[],[],{"skillNo":71,"name":72,"nameZh":6,"description":73,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":74,"sourceHtmlUrl":75,"categories":76,"keywords":77,"publishedAt":19},"01M77XGHB8NDSNVSDWFCN1WJBY","papers","Search preprints from arXiv, bioRxiv, and medRxiv without API keys. Supports search, lookup, extraction, citations, enrichment, and paper downloads that bundle PDFs plus linked GitHub code.","https:\u002F\u002Fgithub.com\u002FStartupBros-com\u002Fhov-stack\u002Fblob\u002Fmain\u002Fskills-patches\u002Fpapers\u002FLICENSE","https:\u002F\u002Fgithub.com\u002FStartupBros-com\u002Fhov-stack\u002Fblob\u002Fmain\u002Fskills-patches\u002Fpapers\u002FSKILL.md",[],[],{"skillNo":79,"name":72,"nameZh":6,"description":73,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":80,"categories":81,"keywords":82,"publishedAt":19},"01M77XGBC08DP018FQY16FM0X8","https:\u002F\u002Fgithub.com\u002FStartupBros-com\u002Fhov-stack\u002Fblob\u002Fmain\u002Fskills-patches\u002Fpapers\u002F.base\u002FSKILL.md",[],[],{"skillNo":84,"name":85,"nameZh":6,"description":86,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":87,"categories":88,"keywords":89,"publishedAt":19},"01M77XG4P5EM9PM8NGRZSJV4CB","wsl-cdp","Drive the user's real, logged-in Windows browser (Chrome, Edge, or Brave) from a Claude Code session running inside WSL2, over the Chrome DevTools Protocol - read authenticated dashboards and consoles, take real screenshots, open and read pages with the user's sessions. Use when a task needs the user's existing web logins and the session runs in WSL2, where the Claude in Chrome extension cannot work. Not for native-Windows or macOS Claude Code (use the official Claude in Chrome extension), not for anonymous scraping or CI (use headless Playwright inside WSL), not for getting past login walls (those are handed to the human).","https:\u002F\u002Fgithub.com\u002FStartupBros-com\u002Fhov-stack\u002Fblob\u002Fmain\u002Fplugins\u002Fwsl-cdp\u002Fskills\u002Fwsl-cdp\u002FSKILL.md",[],[],{"skillNo":91,"name":92,"nameZh":6,"description":93,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":94,"categories":95,"keywords":96,"publishedAt":19},"01M77XFZE48M4SQ2Y7A3GFY4RD","token-eater","[BETA] Spend a subscription's idle, about-to-expire credits on safe, machine-verified cleanup. You pick the service (Grok \u002F Codex \u002F Claude) and a token-heavy maintenance skill (de-monolithize, de-slop, simplify, dead-code); that service runs the whole skill -> gate -> self-review -> draft-PR loop on its own credits, in an isolated worktree. token-eater verifies the gate and never merges. Manual invocation only during beta.","https:\u002F\u002Fgithub.com\u002FStartupBros-com\u002Fhov-stack\u002Fblob\u002Fmain\u002Fplugins\u002Ftoken-eater\u002Fskills\u002Ftoken-eater\u002FSKILL.md",[],[]]