[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7BDNRHHCP44ZJCSZS693FQ7":3,"skill-same:01M7BDNRHHCP44ZJCSZS693FQ7":52},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":24,"commitSha":25,"skillDirectory":26,"forksCount":27,"repositoryHtmlUrl":28,"repositoryHomepageUrl":29,"ownerHtmlUrl":30,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":31,"install":32,"resources":45},"01M7BDNRHHCP44ZJCSZS693FQ7","check-on-real-data",null,"Prove a change to how threads are stored, synced, sized or read against copies of the real threads on this machine, not only made-up test data. Use for storage, sync, size-limit and performance changes. Read-only on `~\u002F.mousse`.","mousse-agent\u002Fmousse","mousse-agent","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F302444486?v=4","01M78EZ8A6G8RR1Q2X8QNH3V5S","Mousse","01M78EZ8B6B27D7XND6F4HFW32",6,"MIT","https:\u002F\u002Fgithub.com\u002Fmousse-agent\u002Fmousse\u002Fblob\u002Fmaster\u002F.agents\u002Fskills\u002Fcheck-on-real-data\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: check-on-real-data\ndescription: Prove a change to how threads are stored, synced, sized or read against copies of the real threads on this machine, not only made-up test data. Use for storage, sync, size-limit and performance changes. Read-only on `~\u002F.mousse`.\n---\n\n# Check on real data\n\nTests build small, tidy threads. Real ones have hundreds of tool calls, big outputs, images, odd rows from older versions and years of history, and they break limits and queries that tests never reach. A thread of 485 messages could not open on a phone because its snapshot passed a 16,384-node bound that no test thread came near; 4 of the 107 threads on one machine had the same problem (#1045). Run this check before the PR of any change that touches how threads are stored, synced, sized or read, and put its numbers in the PR's Verification section.\n\n## Where the data is\n\n- Each thread: `~\u002F.mousse\u002Fprofiles\u002F\u003Cprofile>\u002Fthread-data\u002F**\u002F\u003Cthread id>\u002Fthread.db`. Messages are rows of the `items` table with `coll = 'm'`. Images and text over 64 KiB are files in the thread's `blobs\u002F` folder, and their place in a row is an empty string.\n- The Net database: `~\u002F.mousse\u002Fprofiles\u002F\u003Cprofile>\u002Fnet\u002Fnet.db`. It holds the request ledger (`net_executions`), display streams (`bridge_thread_streams`, `net_records`) and everything else Net keeps. It is the one database that grows with every device and thread.\n- A user's thread you are asked about can also be read there (see `docs\u002Fthread-storage.md`).\n\nNever write to `~\u002F.mousse`. Read copies: [`real-threads.ts`](real-threads.ts) copies each `thread.db` (with its write-ahead log) to a temporary folder and yields its messages.\n\n## 1. Same result, every thread\n\nWrite a throwaway script outside the repository that runs every real thread through the old code and the new code, and counts the differences. Run it with `npx tsx` from the task's worktree. Import the new code from the worktree and the old code from a detached worktree at the default branch, or a constant you compare against.\n\n```ts\nconst { realThreads } = require('\u003Cworktree>\u002F.agents\u002Fskills\u002Fcheck-on-real-data\u002Freal-threads')\nconst fresh = require('\u003Cworktree>\u002Fsrc\u002Fmms\u002F...')\nconst before = require('\u003Cdefault-branch worktree>\u002Fsrc\u002Fmms\u002F...')\nlet changed = 0, failedBefore = 0, failedNow = 0\nfor (const thread of realThreads()) { \u002F* run both on thread.messages, compare, count *\u002F }\nconsole.log({ changed, failedBefore, failedNow })\n```\n\nReport the counts: threads checked, ones that failed before, ones that fail now, and ones whose result changed. A change that is meant to keep the result must show zero changes. One that is meant to fix a failure shows how many threads it fixes.\n\n## 2. Time and memory\n\n- Time the old and new code on the largest real threads. Report the median of several runs, warm and cold. For a cold read, copy the database to a new file and drop it from the page cache first (`dd if=\u003Ccopy> iflag=nocache count=0`).\n- Note the process's peak memory (`\u002Fusr\u002Fbin\u002Ftime -v`) when the change builds whole threads or snapshots in memory.\n- A migration that adds an index or rewrites rows runs at startup on every user's database: time it on the largest real copy you have, and say how long the first start takes.\n\n## 3. Query plans\n\nA new index can make SQLite pick a worse plan for a query that already existed. For a query on `net.db`, or one that runs on every open or send, add a test that reads `EXPLAIN QUERY PLAN` and fails when the expected index is not used (as `tests\u002Fnet\u002Fsync\u002FledgerPruning.test.ts` does).\n\n## What not to do\n\n- Do not commit real data, thread ids, transcripts or check scripts. Report numbers only.\n- Do not use this in place of tests. Turn each failure it finds into a small test with made-up data that has the same shape.\n","# Check on real data\n\nTests build small, tidy threads. Real ones have hundreds of tool calls, big outputs, images, odd rows from older versions and years of history, and they break limits and queries that tests never reach. A thread of 485 messages could not open on a phone because its snapshot passed a 16,384-node bound that no test thread came near; 4 of the 107 threads on one machine had the same problem (#1045). Run this check before the PR of any change that touches how threads are stored, synced, sized or read, and put its numbers in the PR's Verification section.\n\n## Where the data is\n\n- Each thread: `~\u002F.mousse\u002Fprofiles\u002F\u003Cprofile>\u002Fthread-data\u002F**\u002F\u003Cthread id>\u002Fthread.db`. Messages are rows of the `items` table with `coll = 'm'`. Images and text over 64 KiB are files in the thread's `blobs\u002F` folder, and their place in a row is an empty string.\n- The Net database: `~\u002F.mousse\u002Fprofiles\u002F\u003Cprofile>\u002Fnet\u002Fnet.db`. It holds the request ledger (`net_executions`), display streams (`bridge_thread_streams`, `net_records`) and everything else Net keeps. It is the one database that grows with every device and thread.\n- A user's thread you are asked about can also be read there (see `docs\u002Fthread-storage.md`).\n\nNever write to `~\u002F.mousse`. Read copies: [`real-threads.ts`](real-threads.ts) copies each `thread.db` (with its write-ahead log) to a temporary folder and yields its messages.\n\n## 1. Same result, every thread\n\nWrite a throwaway script outside the repository that runs every real thread through the old code and the new code, and counts the differences. Run it with `npx tsx` from the task's worktree. Import the new code from the worktree and the old code from a detached worktree at the default branch, or a constant you compare against.\n\n```ts\nconst { realThreads } = require('\u003Cworktree>\u002F.agents\u002Fskills\u002Fcheck-on-real-data\u002Freal-threads')\nconst fresh = require('\u003Cworktree>\u002Fsrc\u002Fmms\u002F...')\nconst before = require('\u003Cdefault-branch worktree>\u002Fsrc\u002Fmms\u002F...')\nlet changed = 0, failedBefore = 0, failedNow = 0\nfor (const thread of realThreads()) { \u002F* run both on thread.messages, compare, count *\u002F }\nconsole.log({ changed, failedBefore, failedNow })\n```\n\nReport the counts: threads checked, ones that failed before, ones that fail now, and ones whose result changed. A change that is meant to keep the result must show zero changes. One that is meant to fix a failure shows how many threads it fixes.\n\n## 2. Time and memory\n\n- Time the old and new code on the largest real threads. Report the median of several runs, warm and cold. For a cold read, copy the database to a new file and drop it from the page cache first (`dd if=\u003Ccopy> iflag=nocache count=0`).\n- Note the process's peak memory (`\u002Fusr\u002Fbin\u002Ftime -v`) when the change builds whole threads or snapshots in memory.\n- A migration that adds an index or rewrites rows runs at startup on every user's database: time it on the largest real copy you have, and say how long the first start takes.\n\n## 3. Query plans\n\nA new index can make SQLite pick a worse plan for a query that already existed. For a query on `net.db`, or one that runs on every open or send, add a test that reads `EXPLAIN QUERY PLAN` and fails when the expected index is not used (as `tests\u002Fnet\u002Fsync\u002FledgerPruning.test.ts` does).\n\n## What not to do\n\n- Do not commit real data, thread ids, transcripts or check scripts. Report numbers only.\n- Do not use this in place of tests. Turn each failure it finds into a small test with made-up data that has the same shape.\n",{"name":5,"description":7},"master","1763bf831068028dbde4418c6c23c437eb27f3d7",".agents\u002Fskills\u002Fcheck-on-real-data",0,"https:\u002F\u002Fgithub.com\u002Fmousse-agent\u002Fmousse","https:\u002F\u002Fmousse.work","https:\u002F\u002Fgithub.com\u002Fmousse-agent","2026-11-15 03:51:13",{"source":8,"skillName":5,"commands":33,"prompt":43,"zipAvailable":44},[34,37,40],{"tool":35,"command":36},"npx","npx skills add mousse-agent\u002Fmousse --skill check-on-real-data",{"tool":38,"command":39},"bunx","bunx skills add mousse-agent\u002Fmousse --skill check-on-real-data",{"tool":41,"command":42},"pnpm","pnpm dlx skills add mousse-agent\u002Fmousse --skill check-on-real-data","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fmousse-agent\u002Fmousse\u002Fblob\u002Fmaster\u002F.agents\u002Fskills\u002Fcheck-on-real-data\u002FSKILL.md",true,[46],{"path":47,"resourceType":48,"resourceTypeDesc":49,"sizeBytes":50,"sourceHtmlUrl":51},"real-threads.ts","SCRIPT","脚本",2660,"https:\u002F\u002Fgithub.com\u002Fmousse-agent\u002Fmousse\u002Fblob\u002Fmaster\u002F.agents\u002Fskills\u002Fcheck-on-real-data\u002Freal-threads.ts",[53,60,67,74,81,88],{"skillNo":54,"name":55,"nameZh":6,"description":56,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":57,"categories":58,"keywords":59,"publishedAt":19},"01M7BDNWK4FPAK58R3V5ZRNVXC","test-the-claim","Decide what to test for a change, at which level, and what not to write. Use before writing or changing tests for a feature or fix, and when reviewing a PR's tests. Complements docs\u002Ftesting.md, which covers commands, fixtures and what must stay.","https:\u002F\u002Fgithub.com\u002Fmousse-agent\u002Fmousse\u002Fblob\u002Fmaster\u002F.agents\u002Fskills\u002Ftest-the-claim\u002FSKILL.md",[],[],{"skillNo":61,"name":62,"nameZh":6,"description":63,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":64,"categories":65,"keywords":66,"publishedAt":19},"01M78EZWBMSDNV3HFHZYDM7G2N","debug-journal","Use when the user says \"fixed\" or \"fixed it\" after a long debugging session. Appends one line about the root cause and the faster path to journal.md. Only for hard-won debugging, never features or trivial fixes.","https:\u002F\u002Fgithub.com\u002Fmousse-agent\u002Fmousse\u002Fblob\u002Fmaster\u002F.opencode\u002Fskills\u002Fdebug-journal\u002FSKILL.md",[],[],{"skillNo":68,"name":69,"nameZh":6,"description":70,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":71,"categories":72,"keywords":73,"publishedAt":19},"01M78EZQZB3C7GWF39SNKT9KB3","work","Start or resume authorized implementation or local testing in this repository. Keep experiments local and coordinate meaningful changes intended for the shared codebase. Excludes discussion, planning, review, and diagnosis without a request to implement.","https:\u002F\u002Fgithub.com\u002Fmousse-agent\u002Fmousse\u002Fblob\u002Fmaster\u002F.agents\u002Fskills\u002Fwork\u002FSKILL.md",[],[],{"skillNo":75,"name":76,"nameZh":6,"description":77,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":78,"categories":79,"keywords":80,"publishedAt":19},"01M78EZMRTK1HREHEES96R6V7T","nightly-build-release","Build and publish a nightly prerelease of master on request, report each job as it finishes, and fix whatever blocks it, as many times as needed, until a nightly is published. Optionally install it on the user's machine when asked.","https:\u002F\u002Fgithub.com\u002Fmousse-agent\u002Fmousse\u002Fblob\u002Fmaster\u002F.agents\u002Fskills\u002Fnightly-build-release\u002FSKILL.md",[],[],{"skillNo":82,"name":83,"nameZh":6,"description":84,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":85,"categories":86,"keywords":87,"publishedAt":19},"01M78EZHYVFPN79C71D93V96DK","mousse-ui","Mousse's UI\u002FUX principles for designing or reviewing any renderer surface (settings pages, dialogs, menus, lists, pickers). Use before building or restyling UI in src\u002Frenderer, and when judging whether a screen is \"slop\".","https:\u002F\u002Fgithub.com\u002Fmousse-agent\u002Fmousse\u002Fblob\u002Fmaster\u002F.agents\u002Fskills\u002Fmousse-ui\u002FSKILL.md",[],[],{"skillNo":89,"name":90,"nameZh":6,"description":91,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":92,"categories":93,"keywords":94,"publishedAt":19},"01M78EZEHBXBXBBDY2WFYS5TMJ","end-session","Wrap up repository work only when the user explicitly asks to end the session. Preserve local experiments or publish the selected shared-work outcome and record a durable handoff. Ordinary implementation, commit\u002Fpush requests, task completion, and casual goodbyes do not trigger this skill.","https:\u002F\u002Fgithub.com\u002Fmousse-agent\u002Fmousse\u002Fblob\u002Fmaster\u002F.agents\u002Fskills\u002Fend-session\u002FSKILL.md",[],[]]