@aria-framework/ask 0.0.0-stage → 0.1.1

package/CHANGELOG.md

@@ -0,0 +1,77 @@
+# @aria-framework/ask
+
+## 0.1.1 — 2026-11-03
+
+Fix: the answer check refused every multi-sentence answer that cites before a full stop — the shape the answer rules ask
+for. A cite or placeholder was replaced by a space, so "Waiting on us [r1]. It" read "us  . It"; the spaced-dot rule
+(0.1.0's re-review) joined that into "us.It", a domain, and the answer was withheld as `link` (the desk's bench: 41 such
+failures, payload-free twins included; real tokens `us.It`, `Logistics.This`, `agents.Thandi`).
+
+- A run of cites/placeholders directly before `. , ; : ! ?` is now removed WITH the whitespace before it ("us [r1]."
+  reads "us."); a run directly after one of those and directly before a non-space is removed too ("evil.[r1]Com" reads
+  "evil.Com"). Elsewhere a mark is still a space. Both only join text onto punctuation, so `evil [r1].com`,
+  `evil [r1]. com`, `evil [r1] .com`, `evil{{r1}}.com` and the NBSP / newline / fullwidth / ideographic variants stay
+  `link`. `SPACED_DOT` and `DOMAINISH` are unchanged. "evil [r1]. Com" reads as the prose "evil. Com" and gets its verdict.
+- Both new rules are in the check's `RULE_SOURCES`.
+- **The scorer hash moves** (the check's fingerprint is in it): every app's stored bench record stops being `usable()`,
+  so every app re-benches once after pinning 0.1.1. The prompt hash and the catalogue hash do not move.
+
+## 0.1.0 — 2026-11-02
+
+First release: extracted from the hub's (0.97.2) and soc's (0.89.5) copies of Ask anything, taking the
+stricter of each function (the hub's `.claude/plans/ask-package.md` §2.2; D-0169).
+
+- From soc, now every app's: all-of `permissions` (`permission` normalised); attacker-path-aware sayable hosts; a pass
+  bound to the scorer (`scorerHash`, `usable`); stuck `working` rows closed by retention (`closeStale`); the context
+  resolved only when its tool is offered (enforced in the router); the unoffered step described in its own call's words;
+  quotes drawn apart.
+- From the hub: `finish` survives a failed write on the success path (now three attempts: full, full error, outcome
+  alone).
+- New: the check removes format characters (zero-width, soft hyphen, BOM) and reads U+3002/U+FF61 as dots before its
+  word rules; `persist` merges only into a record of the same prompt hash AND scorer; the scorer hash covers the app's
+  samples, sampleArgs and control; a tool's result without a result shape is an error; `retention.purge` runs at the
+  house's reach itself; a partner token limit of 1,000,000,000 parses (the apps' `\d{1,9}` refused the top of its own
+  range); a stored `enabled` that is neither 'true' nor 'false' is reported invalid.
+- The reference table design: `REFERENCE`, `referenceMigration`, `bin/ask-migration.js`, `diffMigration`,
+  `diffLiveTables`. Both apps' existing tables satisfy it (their only difference is the context kinds).
+- One canonical settings set; `storageKey` for backward compatibility only.
+- The `ask/1` stream, `principalOf`, the declared-but-unused remote tool.
+- Views (drawer, admin, review), `ask.js`, `ask.css` on theme tokens.
+
+Review round (PR #9, before publishing):
+- The check removes every default-ignorable code point and combining mark (`\p{Cf}`, `\p{Default_Ignorable_Code_Point}`,
+  `\p{M}`) before its word rules, reads bracketed, spelled-out, spaced (`evil . com`, `evil. example`) and middle dots —
+  and the dot-shaped letters ꞏ ᐧ ˑ ˌ — as dots, and finally ALLOW-LISTS the model's text after NFC: letters, digits, space,
+  newline, `. , ; : ! ? ' " ‘ ’ “ ” ( ) - – — _` (new refusal code `disallowed_char`, a format code). Action-claim, sign-in,
+  other-tenant and instruction phrasings widened.
+- The scorer hash covers the check's rule sources and config (the app's own id test included) and the plan contract; the
+  gate re-runs every control on each question and refuses every endpoint if one fails, and the admin page says so.
+- `diffLiveTables` on Postgres with `tenancy: true` checks the whole generated tenancy: the composite key (ON DELETE
+  CASCADE, never ON UPDATE SET …), UNIQUE (id, partner_id), the session default EXACTLY, one FOR ALL policy TO public whose
+  USING and WITH CHECK are EXACTLY the generated expression (casts, brackets and spacing aside), the key to partners, btree
+  indexes — and refuses extra policies, indexes, triggers, checks, keys and uniques.
+- `tools_off` fails closed: an unknown id drops only itself; an unparseable list switches every tool off; both are
+  reported.
+- Only `whyCode` reaches the person; the forms post to the router's mount; provider kinds and the Jobs link are derived.
+
+Known limit: the allow-list admits every letter (`\p{L}`), so names in any script pass; look-alike letters in a domain or a
+ref are caught by the NFKC, dot and link rules, not by the allow-list (decided with the coordinator, PR #9).
+
+Known limits of the check, which is a floor and not a proof (recorded at the re-review of PR #9):
+- Paraphrases that obey a payload in words no rule names (the bench's held-out set) still pass the check; the bench
+  measures them, and Ask cannot act whatever it says.
+- Number words are English only: "cinq", "fünf" or "vyf" are not refused (digits in any script are).
+- A letter with a precomposed accent inside a number word ("tẃo") is a different word to the rule and passes; NFC keeps
+  it composed and the allow-list admits it as a letter.
+- The spaced-dot reading withholds ordinary text such as "e.g. the record", and the spelled-out one "the dot in the
+  name" — the safe side, accepted.
+
+Adoption notes:
+- The hub's check split result strings on `:` when collecting sayable hosts; the package (as soc) does not, so a
+  `host:port`-like token is no longer sayable whole. Kept on purpose (stricter).
+- Curly quotes (“ ” ‘ ’) are allowed. Answers with a slash, `&` or `%` are now withheld as `disallowed_char`; the
+  apps' honest corpora contain none.
+
+**Schema: no change** from either app's tables. **Hashes:** `promptHash` and `catalogueHash` are byte-identical to both
+apps' for the same inputs; `scorerHash` is new for the hub and moves for Soc (its function sources changed), so each
+app re-benches once at adoption (D-0169 item 3).

package/README.md

@@ -1,3 +1,145 @@
-# Temporary Holding Version
+# @aria-framework/ask
 
-This version is a temporary placeholder for this package. An operational version to replace this has been submitted for review and is awaiting a staged release.
+"Ask anything": a read-only question box over an app's own records, in the asker's own request.
+
+```
+plan    the model sees the question, the page context as a ref (or "@context") and the tools this person may use — no data
+run     the app's tools, inside the asker's request: its tenant context, its own `can`, its page permissions asked again
+answer  the model sees the results as ONE fenced document (@aria-framework/ai `fenced`) and writes {answer, cites}
+check   every figure is a {{rN}} placeholder the app fills; no typed number, no link, no unreturned id, and only an
+        allow-listed set of characters; else WITHHELD (the person sees fixed words and the refusal's code, never the
+        model's words)
+audit   `ask.asked`, whatever happened
+```
+
+The package owns the mechanism. **The app owns its tools, its words, its readings, its context, its permissions, its
+help pages, its question bank, its tables and its migrations.** It names no product and no app tool.
+
+## createAsk(config)
+
+```js
+const ask = require('@aria-framework/ask');
+const docs = ask.createDocs({ helpDir, permissions: ALL_PERMISSIONS, embed: (t) => ai().embed(t, { route: 'ask-docs' }), asHouse, logger });
+const A = ask.createAsk({
+  invoke: db.invoke,                        // the app's db-client; Ask calls 'AskQuestion' (modelName), 'AiProvider.byId', 'AiRoute.forRoute'
+  logger, tenant: { asPlatform, asHouse }, scopeOf: (req, can) => ({ acting, partnerId, … }),
+  settingsStore: { get: (k) => db.invoke('Settings', 'getInstallation', k), set: (k, v) => db.invoke('Settings', 'setInstallation', k, v) },
+  ai: () => require('./ai'),                // complete(), configFor(id) — lazy
+  auditDays: () => auditRetention.currentDays(),
+  isPublicId,                               // this app's own id test
+  tools: [require('./tools/workList'), …, ask.docsTool(docs, { gate: { permission: 'ai.ask' }, queryType: 'string' })],
+  contextKinds: ['customer', 'work'],       // = the context_kind CHECK of the app's migration
+  contextTools: { 'work.item': 'work' },    // which tool "@context" goes back into, by kind
+  contextRef: (c) => c.remoteRef,           // optional: the id put back (default c.ref)
+  openLabel: (kind) => …,                   // optional: the refused-write link's words
+  words: { not_found, refused_write, cannot, cannot_plan, ambiguous, withheld, off, busy, error, nothing_to_ask },
+  prompts: { planRules, answerRules, example, contextSentence: (ctx) => '\nPage context: …', fenceWhat, hashApp },
+  check: { structured: [...], numericStructured: [], sayableIdFields: { facts: [...], rows: [...] }, namedTokens: [[/\bP[1-4]\b/g, 'band']],
+           sayIps: false, sayWindows: false, rowChipLabel: (row) => … },
+  bench: { samples, sampleArgs: (tool) => ({…}), control: { tool, args, ownId, host, unreturnedRef }, benchCustomer },
+  storageKey: (name) => `ask_${name}`,     // ONLY for an app whose stored keys predate the package (see Settings)
+  principalOf: (req, can) => ({ id, name, acting, scope })   // optional; default: the session user + scopeOf
+});
+```
+
+Every missing piece throws a sentence at boot. Then:
+
+```js
+app.use('/ask', requireAuth, A.routes.ask({ express, requirePermission, audit, readingOf, contextOf }));
+const admin = A.routes.admin({ express, requirePermission, audit, routePolicy, ai, partners, endpoints, onBenchChange, appName });
+app.use('/admin/ai/ask', admin.settingsRouter);
+app.use('/admin/ai/ask-review', admin.reviewRouter);
+scheduler.daily(() => A.retention.purge());      // runs at the house's reach itself; closes stuck rows first
+```
+
+- **The admin routers are the platform's.** Both gate on `ai_config.view` (to see) and `ai_config.edit` (to change), and
+  the app MUST declare those permissions platform-only in its catalogue: the review list reads every partner's questions
+  and the model refuses it at any scope but platform, and the settings are installation-wide. Mount them only behind
+  that gate. `jobsHref` (optional) links the empty state to the app's page for assigning endpoints; the provider kinds
+  shown come from `routePolicy('ask')`; the forms post to the mount the router was given.
+- `contextOf(body, can, req, { offered })` resolves the page's context under the request's own scope and page
+  permissions. **The package calls it only when the kind's context tool is offered** (tools-off applied) and drops a
+  result of another kind or with a ref over 64 characters.
+- `readingOf(req, can)` is the drawer's "Reading:" line.
+- Views: append `ask.viewsDir` to the app's view roots (app views first). `ask/admin` and `ask/review` include the
+  app's `partials/staff-head`, `partials/admin-tabs`, `partials/staff-foot` (override with the `askLayout` local) and use
+  the app's `can`, `csrfToken`, `exactTag`, `whenTag` locals. Include `ask/drawer` in staff-foot with `askWords`
+  (`heading`, `intro`, `base`, `historyNote`) and `askSuggestions` (`{ <kind>: [{ title, items }], '*': [...] }`, `{name}` in a
+  title is the context's name).
+- Assets: mount `ask.assetsDir` at `/js/ask` and load `ask.css` and `ask.js`. The CSS is on Bootstrap 5.3 theme tokens;
+  delete the app's own `.ask-*` block.
+
+## The question model
+
+```js
+// models/AskQuestion.js — the app's worker registry
+module.exports = require('@aria-framework/ask').createAskQuestionModel({
+  driver: () => database.driver(),   // a FUNCTION: read per call, never captured at require time
+  sql,                               // the app's: andTenant() (required), returningId(), now()
+  publicId: () => publicId('ask_question'),
+  currentTenant                      // @aria-framework/db-worker
+});
+```
+
+Platform-scope guards as before: the reads keyed on the asker refuse any other scope and are bounded by `asked_by`.
+`closeStale(cut)` closes `working` rows the loop could not finish.
+
+## The reference table design (your migration owns the tables)
+
+A package cannot ship a migration that runs (db-worker reads one directory and ledgers by filename). So the app writes
+its own, **from the reference**, and its test holds it there:
+
+```bash
+npx ask-migration --number 140 --kinds customer,ticket --out migrations --tenancy 141_ask_tenancy
+```
+
+writes `140_ask.sql` and `140_ask.postgres.sql`: `ask_questions` and `ask_tool_calls`, the tables only. Then:
+claim both numbers in the app's MIGRATIONS.md first; **generate** the tenancy pair with the app's tenancy generator
+(never typed); pin the files in the commit that adds them. In the app's tests:
+
+```js
+assert.deepStrictEqual(ask.diffMigration(fs.readFileSync('migrations/140_ask.sql', 'utf8'), { dialect: 'sqlite', contextKinds }), []);
+assert.deepStrictEqual(await ask.diffLiveTables(driver, { contextKinds, tenancy: true }), []);   // SQLite or Postgres
+```
+
+`diffMigration` compares every statement (comments and COMMENT ON aside). `diffLiveTables` compares on SQLite the stored
+CREATE text and on Postgres the catalog: types, nullability, defaults, identity, each CHECK's values and bounds, the
+cascading key, the indexes, and with `tenancy: true` partner_id, RLS enabled AND forced and a policy. `SCHEMA` and
+`REFERENCE` export the column list. A later package version that needs a column says so in CHANGELOG.md: the app adds a
+NEW migration; `diffLiveTables` makes an app that skipped it fail its tests, not production.
+
+## Settings — one canonical set
+
+| Name (canonical key `ask.<name>`) | Label | Default |
+|---|---|---|
+| `enabled` | Answer questions here | off |
+| `limit_hour` | Questions per person, per hour | 30 (1–10,000) |
+| `limit_day` | Questions per person, per day | 200 (1–100,000) |
+| `partner_tokens_day` | Tokens per partner, per day | 500,000 (0 = none) |
+| `retention` | Retention | 30 days (30, 90, audit) |
+| `tools_off` | Tools | none off |
+| `bench` | (the bench record) | — |
+
+Labels, help text, defaults and ranges are the package's (`ask.SETTINGS`) and the same in every app. New apps use the
+canonical keys and pass nothing. `storageKey` exists only so an app with stored rows keeps reading them.
+
+## Hashes and pinning
+
+`promptHash` is `sha256(JSON.stringify({v:1, app, plan, answer, answerSchema, tools}))` — the apps' own document, so an app
+that passes its rule texts verbatim keeps its hash. `catalogueHash` serialises each tool as it declared its gate
+(`permission` or `permissions`). The bench record is bound to the prompt hash AND `scorerHash()` (which covers the
+package's scoring code and the app's samples, sampleArgs and control). Pin this package EXACTLY and add a
+`test-ask-hashes.js` holding `A.hashes()` as constants: any upgrade that moves one switches Ask off on stage until the
+bench passes again, and the PR must say so.
+
+## The remote-Ask seam (interface only)
+
+- `principalOf(req, can)` is the loop's only source of identity.
+- `ask/1` (lib/stream.js, `STREAM_VERSION`, `stream.parseStream`) is the answer stream. Hrefs are app-relative.
+- A tool may be declared `{ id, kind: 'remote', service, … }`: validated, hashed, shown on the admin page, **never
+  offered**. No routes, tokens or fan-out exist yet.
+
+## Tests
+
+`npm test` runs every suite in `test/` (the runner fails on a suite it does not list). PGlite and better-sqlite3 are
+dev dependencies: a missing engine is a failure, not a skip.

package/assets/ask.css

@@ -0,0 +1,81 @@
+/* ASK ANYTHING — the box and the drawer (@aria-framework/ask). Extracted from the hub's public/css/style.css (its .ask-*
+   block, hard-coded colours) onto Bootstrap 5.3 THEME TOKENS (the --bs-* variables @aria-framework/theme sets for light
+   and dark), so the drawer follows dark mode. The app links this file and deletes its own .ask-* block. 440 px on the
+   right over the page, not modal; full screen at phone width. Everything inside the conversation is text drawn by ask.js. */
+.askbox{ display:inline-flex; align-items:center; gap:8px; width:300px; height:38px; padding:0 8px 0 12px; border:1px solid var(--bs-border-color);
+  border-radius:10px; background:var(--bs-tertiary-bg); color:var(--bs-secondary-color); font-size:13px; cursor:text; text-align:left; }
+.askbox .ph{ flex:1; min-width:0; overflow:hidden; text-overflow:ellipsis; white-space:nowrap; }
+.askbox kbd{ font:600 11px Inter,system-ui,sans-serif; padding:2px 6px; border:1px solid var(--bs-border-color); border-radius:5px; background:var(--bs-body-bg); color:var(--bs-body-color); }
+.askbox[aria-expanded="true"]{ border-color:var(--bs-primary); box-shadow:0 0 0 3px rgba(var(--bs-primary-rgb),.15); background:var(--bs-body-bg); color:var(--bs-primary-text-emphasis); }
+.askbox:focus-visible{ outline:2px solid var(--bs-primary); outline-offset:1px; }
+.askicon{ color:var(--bs-primary-text-emphasis); }
+.ask-drawer{ position:fixed; top:0; right:0; bottom:0; width:min(440px,100%); z-index:1050; display:flex; flex-direction:column;
+  background:var(--bs-body-bg); color:var(--bs-body-color); border-left:1px solid var(--bs-border-color); box-shadow:-12px 0 32px rgba(0,0,0,.14); }
+.ask-drawer[hidden]{ display:none; }
+.ask-head{ display:flex; align-items:center; gap:8px; padding:14px 16px 6px; flex-wrap:wrap; }
+.ask-title{ font-size:15px; font-weight:600; margin:0; flex:1; min-width:0; }
+.ask-tools{ display:flex; align-items:center; gap:2px; }
+.ask-scope, .ask-acting{ padding:0 16px 8px; }
+.ask-acting{ margin:0 16px 8px; padding:6px 10px; border-radius:8px; background:var(--bs-warning-bg-subtle); color:var(--bs-warning-text-emphasis); }
+.ask-body{ flex:1; overflow-y:auto; padding:8px 16px; border-top:1px solid var(--bs-border-color); }
+.ask-sugg h3{ margin:12px 0 6px; font-size:11px; font-weight:600; text-transform:uppercase; letter-spacing:.06em; color:var(--bs-secondary-color); }
+.ask-sugg ul, .ask-conv, .hist, .steps{ list-style:none; margin:0; padding:0; }
+.ask-sugg li + li{ margin-top:6px; }
+.ask-sugg button{ width:100%; text-align:left; font-size:13.5px; color:var(--bs-body-color); background:var(--bs-body-bg); border:1px solid var(--bs-border-color); border-radius:8px; padding:9px 12px; }
+.ask-sugg button:hover{ border-color:var(--bs-primary-border-subtle); background:var(--bs-primary-bg-subtle); }
+.ask-turn{ margin:10px 0 16px; }
+.ask-q{ display:flex; gap:8px; justify-content:flex-end; align-items:baseline; }
+.ask-q > span{ background:var(--bs-primary-bg-subtle); color:var(--bs-body-color); padding:7px 11px; border-radius:10px; max-width:85%; overflow-wrap:anywhere; }
+.ask-time{ font-size:11.5px; color:var(--bs-secondary-color); }
+.steps{ display:flex; flex-direction:column; gap:6px; margin-top:10px; font-size:13px; color:var(--bs-secondary-color); }
+.steps li{ display:flex; gap:8px; align-items:center; }
+.steps .ic{ width:18px; text-align:center; flex:none; }
+.steps li.done .ic{ color:var(--bs-success-text-emphasis); }
+.steps li.run{ color:var(--bs-body-color); font-weight:500; }
+.ans{ background:var(--bs-body-bg); border:1px solid var(--bs-border-color); border-radius:10px; margin-top:10px; padding:10px 14px 12px; min-width:0; }
+.ans-head{ display:flex; align-items:center; gap:8px; margin-bottom:6px; }
+.ans-label{ font-size:11px; font-weight:600; text-transform:uppercase; letter-spacing:.06em; color:var(--bs-secondary-color); }
+.mw, .mw-tag{ display:inline-flex; align-items:center; font:600 10.5px Inter,system-ui,sans-serif; text-transform:uppercase; letter-spacing:.05em;
+  padding:2px 7px; border-radius:5px; background:var(--bs-info-bg-subtle); color:var(--bs-info-text-emphasis); }
+.mw.fixed{ background:var(--bs-secondary-bg); color:var(--bs-secondary-color); }
+.ans-head .ask-time{ margin-left:auto; }
+.ans-text{ margin:0 0 8px; overflow-wrap:anywhere; }
+.ask-stale{ margin:0 0 6px; color:var(--bs-warning-text-emphasis); }
+.fig{ font-weight:600; border-bottom:1.5px dotted var(--bs-primary); cursor:help; }
+.cite{ font-size:10.5px; margin-left:1px; }
+.cite span{ display:inline-block; min-width:14px; padding:0 3px; border-radius:4px; background:var(--bs-primary-bg-subtle); color:var(--bs-primary-text-emphasis); text-align:center; font-weight:600; }
+.chips{ display:flex; flex-wrap:wrap; gap:6px; margin:6px 0; }
+.rchip{ display:inline-flex; align-items:center; gap:6px; max-width:100%; font-size:12.5px; font-weight:500; padding:5px 10px; border-radius:8px;
+  border:1px solid var(--bs-primary-border-subtle); background:var(--bs-primary-bg-subtle); color:var(--bs-primary-text-emphasis); text-decoration:none; overflow-wrap:anywhere; }
+.rchip:hover{ filter:brightness(.97); }
+.rchip.fact{ border-color:var(--bs-border-color); background:var(--bs-tertiary-bg); color:var(--bs-body-color); }
+.ask-quote{ margin:8px 0; padding:8px 10px; border-left:3px solid var(--bs-info-border-subtle); background:var(--bs-tertiary-bg); border-radius:0 8px 8px 0; }
+.ask-quote figcaption{ color:var(--bs-secondary-color); margin-bottom:4px; }
+.ask-quote-text{ margin:0; overflow-wrap:anywhere; }
+.ask-note{ color:var(--bs-secondary-color); margin:4px 0; }
+.how{ margin-top:8px; font-size:13px; }
+.how summary{ cursor:pointer; color:var(--bs-primary-text-emphasis); }
+.how-list{ padding-left:18px; margin:8px 0; }
+.how-list li + li{ margin-top:6px; }
+.how-facts{ display:grid; grid-template-columns:auto 1fr; gap:2px 10px; margin:0; }
+.how-facts dt{ font-weight:600; color:var(--bs-secondary-color); }
+.how-facts dd{ margin:0; overflow-wrap:anywhere; }
+.ask-thumbs{ display:flex; flex-wrap:wrap; align-items:center; gap:6px; margin-top:10px; }
+.thumb{ display:inline-grid; place-items:center; width:30px; height:30px; border-radius:7px; border:1px solid var(--bs-border-color); background:var(--bs-body-bg); color:var(--bs-secondary-color); padding:0; }
+.thumb[aria-pressed="true"]{ background:var(--bs-emphasis-color); border-color:var(--bs-emphasis-color); color:var(--bs-body-bg); }
+.ask-fb{ flex-basis:100%; }
+.ask-fb fieldset{ display:flex; flex-direction:column; gap:4px; margin-top:6px; }
+.ask-form{ border-top:1px solid var(--bs-border-color); padding:10px 16px 12px; display:flex; flex-direction:column; gap:6px; background:var(--bs-body-bg); }
+.ask-row{ display:flex; gap:8px; align-items:flex-end; }
+.ask-row textarea{ flex:1; min-width:0; font-size:14px; line-height:1.4; border:1px solid var(--bs-border-color); border-radius:10px; padding:9px 11px; resize:none;
+  background:var(--bs-body-bg); color:var(--bs-body-color); }
+.ask-ctx{ display:inline-flex; align-items:center; gap:4px; align-self:flex-start; max-width:100%; font-size:12.5px; padding:3px 4px 3px 10px;
+  border-radius:8px; background:var(--bs-primary-bg-subtle); color:var(--bs-primary-text-emphasis); font-weight:500; }
+.ask-ctx .cx{ border:0; background:transparent; color:inherit; font-size:15px; line-height:1; padding:2px 6px; border-radius:5px; }
+.hist li{ padding:8px 0; border-bottom:1px solid var(--bs-border-color-translucent); }
+.hist-q{ border:0; background:none; padding:0; color:var(--bs-primary-text-emphasis); text-align:left; }
+@media (max-width: 576px){
+  .askbox{ width:38px; padding:0; justify-content:center; background:var(--bs-body-bg); }
+  .askbox .ph, .askbox kbd{ display:none; }
+  .ask-drawer{ width:100%; border-left:0; box-shadow:none; }
+}