[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7DDCBY9RVBWS97SA3ZT2JT3":3,"skill-same:01M7DDCBY9RVBWS97SA3ZT2JT3":48},{"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":31,"ownerDescription":32,"updatedAt":33,"install":34,"resources":47},"01M7DDCBY9RVBWS97SA3ZT2JT3","ambient-qa",null,"QA of bise's product flows without the user's hands and without his accounts - throwaway hubs (scripted or with his model) on fake accounts (mail, Slack, Linear, calendar, GitHub, HN, Reddit, X, Bluesky, Kit), pages driven in a throwaway headless Chrome, asks to main measured, scripted product runs with a one-screen report, and bise desktop's QA world (a throwaway home with 2-3 seeded projects) that apps\u002Fdesktop\u002Ftests drive. Load it when you (ambient, ambient-pm, ambient-lead, the desktop QA agents) check a flow, a page, main's answers or the desktop app's world.","gvergnaud\u002Fbise","gvergnaud","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F9265418?v=4","01M7DDBF4P9FC0JR6BJDMX1Q2J","Gabriel Vergnaud","01M7DDBF5FKMW0JNE71GV5SKBN",51,"Apache-2.0","https:\u002F\u002Fgithub.com\u002Fgvergnaud\u002Fbise\u002Fblob\u002Fmain\u002Fdev\u002Fplugins\u002Fambient-qa\u002Fskills\u002Fambient-qa\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: ambient-qa\ndescription: QA of bise's product flows without the user's hands and without his accounts - throwaway hubs (scripted or with his model) on fake accounts (mail, Slack, Linear, calendar, GitHub, HN, Reddit, X, Bluesky, Kit), pages driven in a throwaway headless Chrome, asks to main measured, scripted product runs with a one-screen report, and bise desktop's QA world (a throwaway home with 2-3 seeded projects) that apps\u002Fdesktop\u002Ftests drive. Load it when you (ambient, ambient-pm, ambient-lead, the desktop QA agents) check a flow, a page, main's answers or the desktop app's world.\n---\n\n# QA on throwaway hubs\n\nEverything here runs on throwaway hubs with their own files under\n`\u003Cbise dev dir>\u002Fqa\u002F\u003Cyour agent name>\u002F` (one QA hub per agent: you never\ndrive another agent's), or, for the world, under `~\u002F.bise\u002Fgate`. Nothing\nhere needs the user. The Swift app this plugin once drove (QA.swift's\nsocket, scenes, fn, keys, snapshots) is gone; bise desktop's own checks\nlive in `apps\u002Fdesktop\u002Ftests` (its harness launches the Electron app, and\n`checks\u002Fworld.ts` \u002F `checks\u002Freal.sh` call `qa.py world up|down|state|say`).\n\n**The user's rule: no test, QA or demo touches his real accounts** (Gmail,\nOutlook, Slack, Linear, calendar, any connector, read or write, and his\nbrowser through computer use), unless he asks for that one run himself.\nEvery QA hub has fake accounts instead: `tools.gmail.*`, `tools.slack.*`,\n`tools.linear.*`, `tools.calendar.*`, `tools.github.*`, `tools.hn.*` over\ncanned items (Camille, ops at acme: 20 mails of her day with a few to\nanswer, newsletters and a thread where the offsite venue gets decided;\nSlack #team and #launch with the launch numbers; 5 Linear issues; today's\nmeetings; the repo's stars and traffic; a Show HN at #7), dated on the\nhub's today (FAKE_ACCOUNTS_TODAY pins it), whose writes (drafts, sends,\nposts, issues, events, comments) only land in the hub's outbox\n(`outbox()`). Real connectors and the computer plugin are refused\nunless `real_accounts_approved_by_user=\u003Cthe user's own message id>`: never\nset it on your own.\n\nNever on your real work: since docs\u002Fissues\u002F16 a hub takes a client's\nhello only from a process the user started, never from an agent's, so\nthe real workspace is refused (\"the real workspace needs a client the\nuser started\"). QA runs on throwaway hubs (fake, model, or a world).\n\n## Tools (`tools.ambient_qa.*`) and the same as one-line commands\n\nThe CLI is `python3 dev\u002Fplugins\u002Fambient-qa\u002Fqa.py \u003Ccmd>` from the repo root\n(this repo's own dev plugin: never in the release's `plugins\u002F`).\nEvery tool returns JSON.\n\n| tool | CLI | what |\n|---|---|---|\n| `launch({hub})` | `qa.py launch [--hub fake\\|model]` | a throwaway hub, its page server and the fake accounts. fake (default): tests\u002Fe2e.py's scripted provider, main answers `ack: \u003Cyour words>`. model: the same with the user's configured model and keys (costs credits; main's real behavior on the fake accounts). Returns `pages_url` and the outbox file |\n| `stop()` | `qa.py stop` | stop the throwaway hub (its own pids only) |\n| `outbox()` | `qa.py outbox` | what the QA hub's main wrote to the fake accounts (drafts, sends, posts, issues, events) |\n| `page_publish({file, id, edit, done, answers})` | `qa.py page publish [--file f.html] [--id x]` | the throwaway hub's main publishes a page (`sb page publish`): your file or amb-kit's weekly-update example; returns its URL |\n| `page_notes({id, notes, send})` | `qa.py page notes --id x --notes '[...]' --send` | draft notes through the page's own API (token, origin), as notes.js does; `send` sends them to the page's agent |\n| `page_meta({id})` | `qa.py page meta --id x` | versions, state, the notes and their status |\n| `page_drive({id, steps, schemes})` | `qa.py page drive --id x --steps f.json` | the page in a throwaway headless Chrome (no logins, local only): your steps, 2x shots in light and dark |\n| `page_fixture()` | `qa.py page fixture` | weekly-update at v1, four notes sent (note, voice, edit, unsure), then v2 marking one done, answering one, and moving the unsure one's block |\n| `ask({texts, capsule})` | `qa.py ask \"what's my next meeting\" \"plan the offsite\" [--file asks.txt] [--no-capsule]` | asks to main on the running throwaway hub (`launch --hub model` first), one after the other, through a `bise ambient-core` child: main's words, their length, the time until all idle, the pages made or updated, the cards; asks-\u003Ctime>.json |\n| `tick()` | `qa.py tick [--n 1]` | the fake launch's next batch of comments comes in now |\n| `run({scenario})` | `qa.py run weekly-update\\|inbox\\|offsite\\|... [--keep]` | a scripted product run on a model hub with the fake accounts (costs credits); a one-screen report (see \"Scripted product runs\") |\n| `world({do: \"up\"\\|\"state\"\\|\"say\"\\|\"feature\"\\|\"down\"})` | `qa.py world up [--projects 3] [--bise \u003Cpath>]`, `world state`, `world say \u003Cproject\\|bise> \"words\"`, `world feature \u003Cproject> \u003Cname>`, `world down` | bise desktop's QA world (below) |\n\n## The world (bise desktop)\n\n`world up` makes one isolated HOME under `~\u002F.bise\u002Fgate` with bise's home\nhub and 2-3 projects (telemetry, shop, docs), each its own fake-provider\nhub, seeded by its fake main: agents in worktrees, a committed diff ready\nto merge, an agent at work in a dirty worktree, cards with options, a\nmerge question, a blocked agent, an artifact, a message between two\nagents, a question main answered for him (with why), a tool approval he\nrefused with a note in auto mode, receipts. `world state` lists each\nhub's agents, commits, diffs, cards and socket; `world say` sends words\nto that project's fake main (`[[bash: ...]]` scripts it); `world feature`\nmakes a real approval card. `world computer \u003Cproject> \u003Cagent>\ndriving|paused|stopped|none ['Chrome · amazon.fr']` sets that agent's\ncomputer-use row in a fake computer-use world under the world's run dir\n(fake_cu.py: state.json in the broker's shape and a fake broker on the\ncommands' socket answering pause\u002Fresume\u002Fstop\u002Fpeek with a canned still;\nno browser, never his computer use), so the app's real core, the window\nand the TUI show it; `world computer-log` lists the calls that fake\nbroker got ({act, agent, key, args, ok}). Always `world down` when done: it stops\nevery hub by pid. A world outside `~\u002F.bise\u002Fgate`, or a value reaching his\nhubs, is refused.\n\n## Pages in a throwaway Chrome\n\n1. `launch()`, then `page_fixture()` (or `page_publish(...)`): the page\n   on the throwaway hub's page server (127.0.0.1).\n2. `page_drive({id, steps, schemes: [\"light\", \"dark\"]})` (CLI: `qa.py page\n   drive --id weekly-update --steps steps.json`): a throwaway headless\n   Chrome (its own profile, no logins, deleted after; local URLs only,\n   never the user's browser) runs your steps: `{eval}` (read the DOM, the\n   notes' state), `{mouse: [x, y], click}`, `{key}`, `{type}` (select,\n   note, edit, send like the user), `{wait}`, `{shot: \"name\"}` (2x PNGs,\n   once per color scheme). `\u002Fp\u002F\u003Cid>\u002Fv\u002F1` and `\u002Fv\u002F2` give each version.\n3. Check the result with `page_meta(id)` (the notes' status: open, sent,\n   done, answered, moved).\n4. `stop()` when done.\n\nThe `computer` plugin (the user's own Chrome, with his logins) is not for\nQA: only when the user asks for that run himself.\n\n## Scripted product runs\n\n`qa.py run \u003Cscenario>` measures speed and the notes loop the same way\nevery time, on a model hub (your model and keys, throwaway state, costs\ncredits) with the fake accounts only:\n\n| scenario | the ask |\n|---|---|\n| `weekly-update` | draft my weekly update from Slack and Linear |\n| `inbox` | go through my mail, draft replies, don't send |\n| `offsite` | plan the offsite: the venue we decided, the budget, and an agenda for the two days, from my mail and Slack |\n| `launch-watch` | watch the launch until tomorrow, tell me what needs an answer (no notes: after v1 it ticks once, the page must republish with the new comments that need an answer; any write or a card not about such a comment fails) |\n| `morning` | make today's morning page (roadmap E: first a scripted hub seeds the night with no credits: fix-csv done with PR #241, a cookie-banner question card, the inbox review with unsent replies, the domain plan with his steps, the launch watch and its 30 min timer; then the model hub reuses that state with the `today` set: meetings 15, 60 and 150 min after the hub's start, Granola notes, people.md. Checks: id `morning-\u003Ctoday>`, no review, question or sources, 5 headings at most, every line outside the meetings links to `\u002Fp\u002F\u003Cpage>#\u003Citem>` or https, each meeting's brief 3 lines at most, all three meetings and the night's five items there, outbox empty) |\n| `meetings` | brief me before my meetings today, and do the follow-ups after (real time, about 75 min: the `today` set puts the Northwind renewal 15 min after the hub's start, its Granola notes at 47, the team sync at 60; the hub's own timers fire, nothing is ticked. Checks: a brief page `meeting-\u003Cslug>-\u003Cdate>` for each before its start, with a checklist of what is open, 3 questions in a prose `\u003Col>` and a link to last time; the renewal's v2 within 30 min of its end with his rows `data-who=\"yours\"`, an email or message draft and Granola's decisions (3-year, SSO, 45 seats); no card about a meeting; outbox empty. run.json lists the `sb every` timers main set) |\n\nRoadmap A's inputs: a fake `tools.kit.*` (subscriber_count 1204,\nlist_broadcasts, broadcast_stats, create_draft → a draft url, written to\nthe outbox as kind `kit-draft`; no send tool), `tools.github.list_releases`\n(v2.5.0's notes for the newsletter), Lélio's Slack thread in #ops asking\nfor a written summary to HR by mail, and Benjamin's DM (`@benjamin.roy`)\nasking how to install the CLI. Every outbox line has a `kind`\n(`gmail-draft`, `kit-draft`, `slack-post_message`...). Roadmap C: #acme-feedback\nholds Benjamin's three reports. Slack messages carry their permalink.\n\nThe launch (roadmap B) lives in the fake `tools.hn.*`, `tools.reddit.*`,\n`tools.x.*`, `tools.bluesky.*` (the real plugins' tool names, read-only):\nacme's Show HN, r\u002FSaaS post, X and Bluesky posts, 19 comments in three\nbatches (questions, bugs, praise, spam, one \"does it work with Claude\nCode?\"). Batch 0 is there at the hub's start; `tick()` brings the next\none now (else every 30 min of hub time, AMBIENT_QA_TICK_MIN), and points,\nrank and likes move with each batch. Links are the real places'\n(news.ycombinator.com\u002Fitem?id=, reddit \u002Fr\u002F\u003Csub>\u002Fcomments\u002F\u003Cid>,\nx.com\u002F\u003Cuser>\u002Fstatus\u002F\u003Cid>, bsky.app\u002Fprofile\u002F...).\n\nThe ask goes through a `bise ambient-core` child (text only: no mic, no\nvoice). It records the time to the page's first appearance (`sb page\nstart`) and to v1, v1's blocks, then posts 3 scripted notes over the page\nAPI (make this shorter \u002F drop this part \u002F where does this number come\nfrom), sends, and records the time to v2 and each note's outcome; the\noutbox (a send on `inbox` fails the run; drafts are fine); main's line\nand its length (over 120 characters fails); cards that showed. Output in\n`\u003Cqa dir>\u002Fout\u002Fruns\u002F\u003Cscenario>-\u003Ctime>\u002F`: `run.json`, `report.txt` (one\nscreen), the page bodies. It stops the hub it started unless `--keep`.\nThe report names the build: a tree behind ambient-app prints a WARNING\n(a fail may be the stale build); `--require \u003Csha>,\u003Csha>` refuses a tree\nthat lacks those commits. One run at a time: it uses your QA dir's hub.\n\n## The heavy lock\n\n`qa.py world up` takes `~\u002F.bise\u002Frun\u002Fheavy.lock` itself only around a\nhelpers build that will compile (scripts\u002Fbins.sh, stale helpers); with\nbuilt helpers and an existing bise it never waits on the lock (main\nm_10008: the lock is for cargo builds and gates only). It skips the lock\nwhen it already runs under a `lockf ... heavy.lock` of its own ancestry,\nso don't wrap `world up` in the lock yourself (a `lockf -k heavy.lock\nqa.py world up` once waited on itself and held the lock for everyone,\nmain m_9310).\n\n## Artifacts\n\nOne artifact per gate or check, never one per image: register its index\npage (a bise page or an index.html next to the shots) or a contact sheet\nof the shots with `sb artifact add`, once. Never `sb artifact add` each\nPNG: the user's \u002Fartifacts filled with 47 QA screenshots (art-archive,\nm_16871). Old ones go away with `sb artifact archive`.\n\n## Limits\n\n- One throwaway hub per agent; a `model` hub costs credits.\n- The core's mic, listener and voice are not exercised (amb-core's tests\n  cover those); asks reach main as typed text.\n- No app here: the desktop app's windows, keys and shots are\n  apps\u002Fdesktop\u002Ftests' (its harness, on fake cores or on this world).\n","# QA on throwaway hubs\n\nEverything here runs on throwaway hubs with their own files under\n`\u003Cbise dev dir>\u002Fqa\u002F\u003Cyour agent name>\u002F` (one QA hub per agent: you never\ndrive another agent's), or, for the world, under `~\u002F.bise\u002Fgate`. Nothing\nhere needs the user. The Swift app this plugin once drove (QA.swift's\nsocket, scenes, fn, keys, snapshots) is gone; bise desktop's own checks\nlive in `apps\u002Fdesktop\u002Ftests` (its harness launches the Electron app, and\n`checks\u002Fworld.ts` \u002F `checks\u002Freal.sh` call `qa.py world up|down|state|say`).\n\n**The user's rule: no test, QA or demo touches his real accounts** (Gmail,\nOutlook, Slack, Linear, calendar, any connector, read or write, and his\nbrowser through computer use), unless he asks for that one run himself.\nEvery QA hub has fake accounts instead: `tools.gmail.*`, `tools.slack.*`,\n`tools.linear.*`, `tools.calendar.*`, `tools.github.*`, `tools.hn.*` over\ncanned items (Camille, ops at acme: 20 mails of her day with a few to\nanswer, newsletters and a thread where the offsite venue gets decided;\nSlack #team and #launch with the launch numbers; 5 Linear issues; today's\nmeetings; the repo's stars and traffic; a Show HN at #7), dated on the\nhub's today (FAKE_ACCOUNTS_TODAY pins it), whose writes (drafts, sends,\nposts, issues, events, comments) only land in the hub's outbox\n(`outbox()`). Real connectors and the computer plugin are refused\nunless `real_accounts_approved_by_user=\u003Cthe user's own message id>`: never\nset it on your own.\n\nNever on your real work: since docs\u002Fissues\u002F16 a hub takes a client's\nhello only from a process the user started, never from an agent's, so\nthe real workspace is refused (\"the real workspace needs a client the\nuser started\"). QA runs on throwaway hubs (fake, model, or a world).\n\n## Tools (`tools.ambient_qa.*`) and the same as one-line commands\n\nThe CLI is `python3 dev\u002Fplugins\u002Fambient-qa\u002Fqa.py \u003Ccmd>` from the repo root\n(this repo's own dev plugin: never in the release's `plugins\u002F`).\nEvery tool returns JSON.\n\n| tool | CLI | what |\n|---|---|---|\n| `launch({hub})` | `qa.py launch [--hub fake\\|model]` | a throwaway hub, its page server and the fake accounts. fake (default): tests\u002Fe2e.py's scripted provider, main answers `ack: \u003Cyour words>`. model: the same with the user's configured model and keys (costs credits; main's real behavior on the fake accounts). Returns `pages_url` and the outbox file |\n| `stop()` | `qa.py stop` | stop the throwaway hub (its own pids only) |\n| `outbox()` | `qa.py outbox` | what the QA hub's main wrote to the fake accounts (drafts, sends, posts, issues, events) |\n| `page_publish({file, id, edit, done, answers})` | `qa.py page publish [--file f.html] [--id x]` | the throwaway hub's main publishes a page (`sb page publish`): your file or amb-kit's weekly-update example; returns its URL |\n| `page_notes({id, notes, send})` | `qa.py page notes --id x --notes '[...]' --send` | draft notes through the page's own API (token, origin), as notes.js does; `send` sends them to the page's agent |\n| `page_meta({id})` | `qa.py page meta --id x` | versions, state, the notes and their status |\n| `page_drive({id, steps, schemes})` | `qa.py page drive --id x --steps f.json` | the page in a throwaway headless Chrome (no logins, local only): your steps, 2x shots in light and dark |\n| `page_fixture()` | `qa.py page fixture` | weekly-update at v1, four notes sent (note, voice, edit, unsure), then v2 marking one done, answering one, and moving the unsure one's block |\n| `ask({texts, capsule})` | `qa.py ask \"what's my next meeting\" \"plan the offsite\" [--file asks.txt] [--no-capsule]` | asks to main on the running throwaway hub (`launch --hub model` first), one after the other, through a `bise ambient-core` child: main's words, their length, the time until all idle, the pages made or updated, the cards; asks-\u003Ctime>.json |\n| `tick()` | `qa.py tick [--n 1]` | the fake launch's next batch of comments comes in now |\n| `run({scenario})` | `qa.py run weekly-update\\|inbox\\|offsite\\|... [--keep]` | a scripted product run on a model hub with the fake accounts (costs credits); a one-screen report (see \"Scripted product runs\") |\n| `world({do: \"up\"\\|\"state\"\\|\"say\"\\|\"feature\"\\|\"down\"})` | `qa.py world up [--projects 3] [--bise \u003Cpath>]`, `world state`, `world say \u003Cproject\\|bise> \"words\"`, `world feature \u003Cproject> \u003Cname>`, `world down` | bise desktop's QA world (below) |\n\n## The world (bise desktop)\n\n`world up` makes one isolated HOME under `~\u002F.bise\u002Fgate` with bise's home\nhub and 2-3 projects (telemetry, shop, docs), each its own fake-provider\nhub, seeded by its fake main: agents in worktrees, a committed diff ready\nto merge, an agent at work in a dirty worktree, cards with options, a\nmerge question, a blocked agent, an artifact, a message between two\nagents, a question main answered for him (with why), a tool approval he\nrefused with a note in auto mode, receipts. `world state` lists each\nhub's agents, commits, diffs, cards and socket; `world say` sends words\nto that project's fake main (`[[bash: ...]]` scripts it); `world feature`\nmakes a real approval card. `world computer \u003Cproject> \u003Cagent>\ndriving|paused|stopped|none ['Chrome · amazon.fr']` sets that agent's\ncomputer-use row in a fake computer-use world under the world's run dir\n(fake_cu.py: state.json in the broker's shape and a fake broker on the\ncommands' socket answering pause\u002Fresume\u002Fstop\u002Fpeek with a canned still;\nno browser, never his computer use), so the app's real core, the window\nand the TUI show it; `world computer-log` lists the calls that fake\nbroker got ({act, agent, key, args, ok}). Always `world down` when done: it stops\nevery hub by pid. A world outside `~\u002F.bise\u002Fgate`, or a value reaching his\nhubs, is refused.\n\n## Pages in a throwaway Chrome\n\n1. `launch()`, then `page_fixture()` (or `page_publish(...)`): the page\n   on the throwaway hub's page server (127.0.0.1).\n2. `page_drive({id, steps, schemes: [\"light\", \"dark\"]})` (CLI: `qa.py page\n   drive --id weekly-update --steps steps.json`): a throwaway headless\n   Chrome (its own profile, no logins, deleted after; local URLs only,\n   never the user's browser) runs your steps: `{eval}` (read the DOM, the\n   notes' state), `{mouse: [x, y], click}`, `{key}`, `{type}` (select,\n   note, edit, send like the user), `{wait}`, `{shot: \"name\"}` (2x PNGs,\n   once per color scheme). `\u002Fp\u002F\u003Cid>\u002Fv\u002F1` and `\u002Fv\u002F2` give each version.\n3. Check the result with `page_meta(id)` (the notes' status: open, sent,\n   done, answered, moved).\n4. `stop()` when done.\n\nThe `computer` plugin (the user's own Chrome, with his logins) is not for\nQA: only when the user asks for that run himself.\n\n## Scripted product runs\n\n`qa.py run \u003Cscenario>` measures speed and the notes loop the same way\nevery time, on a model hub (your model and keys, throwaway state, costs\ncredits) with the fake accounts only:\n\n| scenario | the ask |\n|---|---|\n| `weekly-update` | draft my weekly update from Slack and Linear |\n| `inbox` | go through my mail, draft replies, don't send |\n| `offsite` | plan the offsite: the venue we decided, the budget, and an agenda for the two days, from my mail and Slack |\n| `launch-watch` | watch the launch until tomorrow, tell me what needs an answer (no notes: after v1 it ticks once, the page must republish with the new comments that need an answer; any write or a card not about such a comment fails) |\n| `morning` | make today's morning page (roadmap E: first a scripted hub seeds the night with no credits: fix-csv done with PR #241, a cookie-banner question card, the inbox review with unsent replies, the domain plan with his steps, the launch watch and its 30 min timer; then the model hub reuses that state with the `today` set: meetings 15, 60 and 150 min after the hub's start, Granola notes, people.md. Checks: id `morning-\u003Ctoday>`, no review, question or sources, 5 headings at most, every line outside the meetings links to `\u002Fp\u002F\u003Cpage>#\u003Citem>` or https, each meeting's brief 3 lines at most, all three meetings and the night's five items there, outbox empty) |\n| `meetings` | brief me before my meetings today, and do the follow-ups after (real time, about 75 min: the `today` set puts the Northwind renewal 15 min after the hub's start, its Granola notes at 47, the team sync at 60; the hub's own timers fire, nothing is ticked. Checks: a brief page `meeting-\u003Cslug>-\u003Cdate>` for each before its start, with a checklist of what is open, 3 questions in a prose `\u003Col>` and a link to last time; the renewal's v2 within 30 min of its end with his rows `data-who=\"yours\"`, an email or message draft and Granola's decisions (3-year, SSO, 45 seats); no card about a meeting; outbox empty. run.json lists the `sb every` timers main set) |\n\nRoadmap A's inputs: a fake `tools.kit.*` (subscriber_count 1204,\nlist_broadcasts, broadcast_stats, create_draft → a draft url, written to\nthe outbox as kind `kit-draft`; no send tool), `tools.github.list_releases`\n(v2.5.0's notes for the newsletter), Lélio's Slack thread in #ops asking\nfor a written summary to HR by mail, and Benjamin's DM (`@benjamin.roy`)\nasking how to install the CLI. Every outbox line has a `kind`\n(`gmail-draft`, `kit-draft`, `slack-post_message`...). Roadmap C: #acme-feedback\nholds Benjamin's three reports. Slack messages carry their permalink.\n\nThe launch (roadmap B) lives in the fake `tools.hn.*`, `tools.reddit.*`,\n`tools.x.*`, `tools.bluesky.*` (the real plugins' tool names, read-only):\nacme's Show HN, r\u002FSaaS post, X and Bluesky posts, 19 comments in three\nbatches (questions, bugs, praise, spam, one \"does it work with Claude\nCode?\"). Batch 0 is there at the hub's start; `tick()` brings the next\none now (else every 30 min of hub time, AMBIENT_QA_TICK_MIN), and points,\nrank and likes move with each batch. Links are the real places'\n(news.ycombinator.com\u002Fitem?id=, reddit \u002Fr\u002F\u003Csub>\u002Fcomments\u002F\u003Cid>,\nx.com\u002F\u003Cuser>\u002Fstatus\u002F\u003Cid>, bsky.app\u002Fprofile\u002F...).\n\nThe ask goes through a `bise ambient-core` child (text only: no mic, no\nvoice). It records the time to the page's first appearance (`sb page\nstart`) and to v1, v1's blocks, then posts 3 scripted notes over the page\nAPI (make this shorter \u002F drop this part \u002F where does this number come\nfrom), sends, and records the time to v2 and each note's outcome; the\noutbox (a send on `inbox` fails the run; drafts are fine); main's line\nand its length (over 120 characters fails); cards that showed. Output in\n`\u003Cqa dir>\u002Fout\u002Fruns\u002F\u003Cscenario>-\u003Ctime>\u002F`: `run.json`, `report.txt` (one\nscreen), the page bodies. It stops the hub it started unless `--keep`.\nThe report names the build: a tree behind ambient-app prints a WARNING\n(a fail may be the stale build); `--require \u003Csha>,\u003Csha>` refuses a tree\nthat lacks those commits. One run at a time: it uses your QA dir's hub.\n\n## The heavy lock\n\n`qa.py world up` takes `~\u002F.bise\u002Frun\u002Fheavy.lock` itself only around a\nhelpers build that will compile (scripts\u002Fbins.sh, stale helpers); with\nbuilt helpers and an existing bise it never waits on the lock (main\nm_10008: the lock is for cargo builds and gates only). It skips the lock\nwhen it already runs under a `lockf ... heavy.lock` of its own ancestry,\nso don't wrap `world up` in the lock yourself (a `lockf -k heavy.lock\nqa.py world up` once waited on itself and held the lock for everyone,\nmain m_9310).\n\n## Artifacts\n\nOne artifact per gate or check, never one per image: register its index\npage (a bise page or an index.html next to the shots) or a contact sheet\nof the shots with `sb artifact add`, once. Never `sb artifact add` each\nPNG: the user's \u002Fartifacts filled with 47 QA screenshots (art-archive,\nm_16871). Old ones go away with `sb artifact archive`.\n\n## Limits\n\n- One throwaway hub per agent; a `model` hub costs credits.\n- The core's mic, listener and voice are not exercised (amb-core's tests\n  cover those); asks reach main as typed text.\n- No app here: the desktop app's windows, keys and shots are\n  apps\u002Fdesktop\u002Ftests' (its harness, on fake cores or on this world).\n",{"name":5,"description":7},"main","7c09ce329eff2caff1dc3e4d44dfd6c236a282ce","dev\u002Fplugins\u002Fambient-qa\u002Fskills\u002Fambient-qa",9,"https:\u002F\u002Fgithub.com\u002Fgvergnaud\u002Fbise","https:\u002F\u002Fbise.dev","https:\u002F\u002Fgithub.com\u002Fgvergnaud","https:\u002F\u002Ftype-level-typescript.com","Building agents at Mistral — Author of bise, type-level-typescript.com, TS-Pattern, HOTScript...","2026-11-15 04:07:17",{"source":8,"skillName":5,"commands":35,"prompt":45,"zipAvailable":46},[36,39,42],{"tool":37,"command":38},"npx","npx skills add gvergnaud\u002Fbise --skill ambient-qa",{"tool":40,"command":41},"bunx","bunx skills add gvergnaud\u002Fbise --skill ambient-qa",{"tool":43,"command":44},"pnpm","pnpm dlx skills add gvergnaud\u002Fbise --skill ambient-qa","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fgvergnaud\u002Fbise\u002Fblob\u002Fmain\u002Fdev\u002Fplugins\u002Fambient-qa\u002Fskills\u002Fambient-qa\u002FSKILL.md",true,[],[49,56,63,70,77,83],{"skillNo":50,"name":51,"nameZh":6,"description":52,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":53,"categories":54,"keywords":55,"publishedAt":19},"01M7DDF53D28NXA17HZP08094A","greet","Greet someone the hello-plugin way. Use when asked for a hello-plugin greeting.","https:\u002F\u002Fgithub.com\u002Fgvergnaud\u002Fbise\u002Fblob\u002Fmain\u002Frust\u002Fplugins\u002Ftests\u002Ffixtures\u002Fhello-plugin\u002Fskills\u002Fgreet\u002FSKILL.md",[],[],{"skillNo":57,"name":58,"nameZh":6,"description":59,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":60,"categories":61,"keywords":62,"publishedAt":19},"01M7DDF3V0J5P7VEWAS7NM458P","triage","triage Linear issues","https:\u002F\u002Fgithub.com\u002Fgvergnaud\u002Fbise\u002Fblob\u002Fmain\u002Frust\u002Fplugins\u002Ftests\u002Ffixtures\u002Fadd\u002Fcodex-linear\u002Fin\u002Fskills\u002Ftriage\u002FSKILL.md",[],[],{"skillNo":64,"name":65,"nameZh":6,"description":66,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":67,"categories":68,"keywords":69,"publishedAt":19},"01M7DDF1EPY3GMFG4WG368B3Q8","jot","jot a note","https:\u002F\u002Fgithub.com\u002Fgvergnaud\u002Fbise\u002Fblob\u002Fmain\u002Frust\u002Fplugins\u002Ftests\u002Ffixtures\u002Fadd\u002Fclaude-notes\u002Fin\u002Fskills\u002Fjot\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":6,"sourceHtmlUrl":74,"categories":75,"keywords":76,"publishedAt":19},"01M7DDETP44AE35MSDENY1THS9","bise-demo","Run bise's built-in demo, a short role-play where a small product team (a pm, a designer, a dev) starts, talks like colleagues, asks the user one question and one agent is dropped, with no file or git change. Use when the user asks to see what bise or you can do, for a demo, a tour or your features (show me what you can do, give me a demo, what can you do?, demo all your features), in any language.","https:\u002F\u002Fgithub.com\u002Fgvergnaud\u002Fbise\u002Fblob\u002Fmain\u002Fprompts\u002Fskills\u002Fbise-demo\u002FSKILL.md",[],[],{"skillNo":78,"name":79,"nameZh":6,"description":6,"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},"01M7DDERQ5JJHVR314X762FD5K","bise-pages","https:\u002F\u002Fgithub.com\u002Fgvergnaud\u002Fbise\u002Fblob\u002Fmain\u002Fprompts\u002Fskills-all\u002Fbise-pages\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},"01M7DDEMV9FHB1HTNWY9W5RXAZ","slides","Make, read or edit a PowerPoint deck (.pptx) - a review, a pitch, a status deck, slides with tables and charts - with a TypeScript script run by `bise office run`. Use when the user asks for slides, a deck or a pptx, sends one, or leaves notes on a pptx artifact.","https:\u002F\u002Fgithub.com\u002Fgvergnaud\u002Fbise\u002Fblob\u002Fmain\u002Fplugins\u002Fslides\u002Fskills\u002Fslides\u002FSKILL.md",[],[]]