@agent-native/core-corpus 0.221.0-same-host-mcp-20261027-20261027081850 → 0.221.0

package/corpus/README.md

@@ -32,4 +32,4 @@ rg -n "defineAction|useActionQuery" node_modules/@agent-native/core-corpus/corpu
 
 ## Generated Counts
 
-- template files: 14842
+- template files: 14961

package/corpus/templates/analytics/.agents/skills/ab-readout/SKILL.md

@@ -0,0 +1,19 @@
+---
+name: ab-readout
+description: >-
+  Read the first-run Builder credits experiment output. Use for first-run
+  onboarding experiment analysis or the weekly radar report.
+scope: dev
+---
+
+# A/B readout
+
+- Run `pnpm --filter analytics radar:ab --experiment first_run_builder_credits`.
+- Verify the sample-ratio result and each arm's denominator before comparing
+  rates. Under the configured minimum arm size, report not enough data.
+- Read the 95% confidence interval and two-proportion p-value beside the raw
+  counts. D1 and D7 only use mature cohorts.
+- Keep per-app entry denominators separate from experiment-wide exposure counts.
+  State the observation window and note the overlapping Clips recording
+  experiment when interpreting Clips results.
+- Report association only. Do not claim the credits variant caused a change.

package/corpus/templates/analytics/.agents/skills/feedback-ingest/SKILL.md

@@ -0,0 +1,22 @@
+---
+name: feedback-ingest
+description: >-
+  Collect product feedback themes for the first-touch radar. Use during a weekly
+  report or after a fired first-session check.
+scope: dev
+---
+
+# Feedback ingest
+
+- Read `.tmp/first-touch-radar/slack-watermark.json`; if absent, use the prior
+  24 hours for the first scan.
+- Search `#product-agent-native-feedback`, `#dev-agent-native`,
+  `#dev-agent-native-feedback`, `#design-an-feedback`, `#qa-agent-native`, and
+  `#plg-agent-native` from the watermark through now using Slack MCP.
+- Keep the search read-only: do not post, react, mark messages read, or open the
+  Slack desktop app.
+- Group feedback into themes with counts, app, dates, and permalinks. Join a
+  theme to a fired radar check only when the app and date overlap; otherwise mark
+  it unlinked.
+- Store only the new watermark and aggregate theme notes in the private radar
+  run directory. Do not copy full messages, emails, or prompt text.

package/corpus/templates/analytics/.agents/skills/first-touch-radar/SKILL.md

@@ -0,0 +1,51 @@
+---
+name: first-touch-radar
+description: >-
+  Run the first-touch product radar for Design, Slides, and Clips. Use for daily
+  watch, replay triage, weekly growth reports, or first-session incidents.
+scope: dev
+---
+
+# First-touch radar
+
+1. Run `pnpm --filter analytics radar:watch --window daily` and read both the
+   JSON and Markdown under `.tmp/first-touch-radar/runs/`.
+2. Treat exit 0 as clear, exit 1 as one or more fired checks, and exit 2 as
+   incomplete evidence. Never turn a missing source or failed query into zero.
+3. On a fire, run `pnpm --filter analytics radar:triage` and inspect the
+   outcome-ranked sessions. Use `metric-watch` and `replay-triage` for details.
+4. For a weekly review, run `pnpm --filter analytics radar:report --week
+   <Monday>` and read the coverage flags, denominators, campaign annotations,
+   previous-issue recovery, and A/B maturity before drawing conclusions.
+   The `campaign-followup` check runs in daily watch and weekly report; compare
+   identified recipients' post-send activity with the seven days before each
+   send, split by recipient app and activity app. Weekly reports cap cumulative
+   BigQuery dry-run estimates at 100 GB; use `--max-total-gb` to set a different
+   report-wide cap, and treat a reached cap as incomplete evidence.
+5. Run `feedback-ingest` for Slack themes. Searches are read-only and begin at
+   the saved watermark.
+6. Send one Telegram to Steve only when a check fired or the weekly report is
+   ready. Include counts and the private run-file path. Send nothing on a quiet
+   day. Never send Slack or email.
+7. Keep recipient emails and prompt text out of logs, reports, Telegram, and
+   commits. The campaign check reads its mode-0600 recipient file in memory and
+   emits aggregate counts only. With a PostgreSQL sink, recipient identifiers
+   are parameters to the read-only query and caching is disabled. A BigQuery
+   sink cannot run campaign follow-up until a keyed recipient matching
+   function is configured; report it as incomplete rather than treating it as
+   zero. Unknown signup campaign labels are grouped as `other`. Anonymous
+   landings cannot be assigned to a recipient; do not interpret identified
+   counts as a complete click total.
+8. This radar changes scripts and skills only. Do not make user-visible UI
+   changes. If UI work becomes necessary, stop that scope and request Steve's
+   screenshot approval before implementing it.
+9. A fired metric may create a local brief through `fix-brief`; no fix starts
+   until Steve approves it.
+10. Interpret the first-touch journeys separately: Design web excludes
+    `/mcp/oauth%` signup paths; Design MCP is signup → `agent_connected` → first
+    `$mcp_tool_call` → `design_created`. Clips activation is signup →
+    `recording_ready` within 24 hours. Treat 2026-10-21 as a definition break:
+    `recording_started` before then includes local-only recordings.
+11. Read the weekly paid-acquisition notes alongside signup counts: Slides ads
+    stopped after 2026-10-16, Bing CPC ended around 2026-10-20 21:00 UTC for
+    Clips/Design, and Google paid spend was about 45% lower from 2026-10-21.

package/corpus/templates/analytics/.agents/skills/fix-brief/SKILL.md

@@ -0,0 +1,19 @@
+---
+name: fix-brief
+description: >-
+  Turn a fired radar check into a local, evidence-backed repair brief. Use when
+  Steve asks for a fix brief or approves one.
+scope: dev
+---
+
+# Fix brief
+
+- Choose one fired check and write `.tmp/first-touch-radar/briefs/<id>.md`.
+- Include its query window, numerator and denominator, comparison baseline,
+  affected app, relevant redacted replay link, and likely source files from
+  `git grep origin/main`.
+- Make acceptance measurable: the detecting query returns under threshold with
+  sufficient coverage on a fresh run.
+- Do not change application code until Steve explicitly approves the fix task.
+  Keep UI changes out; any proposed UI repair needs Steve's screenshot approval.
+- Do not include recipient emails, prompt text, or copied user content.

package/corpus/templates/analytics/.agents/skills/metric-watch/SKILL.md

@@ -0,0 +1,29 @@
+---
+name: metric-watch
+description: >-
+  Inspect first-touch radar checks, their query coverage, thresholds, and
+  backtests. Use when a watch fires or a metric needs calibration.
+scope: dev
+---
+
+# Metric watch
+
+- Run `pnpm --filter analytics radar:watch --window daily`; use `--window
+  hourly` for provisioning and environment-key checks.
+- Read the check status, numerator, denominator, seven-day comparison, and
+  coverage. `insufficient` is not a pass, and `unavailable` means the run exits
+  2.
+- Compare daily thresholds with the incident window and an ordinary week before
+  changing them. Record both windows and the observed counts in the PR or report.
+- Action success uses a weighted `action.response` rate. Keep its app-specific
+  threshold below the quiet-week distribution; do not copy the 95% starting
+  value without a backtest.
+- Read signup campaign tags beside signup drops. Paid search changes are context,
+  not proof that product behavior changed.
+- Event first-seen checks require 10 events that day; went-to-zero checks require
+  at least 50 events across the preceding seven days.
+- Read `campaign-followup` as identified recipient counts after send versus the
+  same recipients' prior seven days, split by recipient and activity app.
+  Anonymous landings cannot be attributed to recipients.
+- The hourly Analytics alert rules cannot express ratios or trailing baselines;
+  do not replace these checks with simple count rules.

package/corpus/templates/analytics/.agents/skills/replay-triage/SKILL.md

@@ -0,0 +1,19 @@
+---
+name: replay-triage
+description: >-
+  Review outcome-ranked first-touch sessions and their replay evidence. Use when
+  radar checks or the weekly report identify a first-session failure.
+scope: dev
+---
+
+# Replay triage
+
+- Run `pnpm --filter analytics radar:triage` or pass `--date YYYY-MM-DD` for a
+  UTC day.
+- Start with `prompted_no_output`, `recording_started_no_ready_clip`, and
+  `onboarding_abandoned`. Friction and performance fields explain outcomes; do
+  not rank sessions by friction score.
+- Review at most the top 15–30 sessions. When a bearer is available, fetch replay
+  timelines only for those recording ids through Analytics' deployed action.
+- Use event names, offsets, counts, trouble labels, and performance summaries.
+  Never include emails, prompt text, URLs, or other user content in a report.

package/corpus/templates/analytics/.agents/skills/storing-data/SKILL.md

@@ -29,6 +29,24 @@ Agent-Native apps use Drizzle ORM over PostgreSQL. Local development uses PGlite
 
 For app code, use Drizzle's schema/query DSL by default. Raw SQL is an escape hatch for additive migrations, health checks, or one-off maintenance, not the normal way to build features.
 
+### Typed raw SQL fragments
+
+When Drizzle cannot express a PostgreSQL query, import `sql` from
+`@agent-native/core/db` and pass its fragment directly to
+`getDbExec().execute()`. Use `sql.value(value)` for bound values,
+`sql.ident(name)` for validated lowercase identifiers, `sql.in(values)` for a
+nonempty list (it includes parentheses), and `sql.join(fragments, separator)`
+or nested fragments for composition. The tag rejects plain string interpolation
+at compile time; do not build `{ sql: `...${value}...` }` or pass interpolated
+templates directly to `execute()`. The diff-scoped guard checks new
+interpolated SQL reaching `getDbExec().execute()` through the helper and wrapper
+paths it analyzes. A reviewed exception needs `// raw-sql-ok: <reason>` as
+a standalone comment immediately before the sink statement or after it on the
+same line. Markers inside SQL text, callbacks, and object properties do not opt
+out. The guard does not audit Drizzle `sql.raw()` or raw adapter `query()` calls,
+so use bound Drizzle fragments or the typed kernel instead of interpolating
+dynamic values through those APIs.
+
 ### Schema and migration ownership
 
 Define PostgreSQL tables with Drizzle's `drizzle-orm/pg-core` exports and use

package/corpus/templates/analytics/CHANGELOG.md

@@ -5,12 +5,23 @@ time from the command menu (Cmd+K → "What's new") or from Settings.
 
 ## 2026-10-24
 
+### Added
+
+- Pick dashboard panels for the agent with the Select button in the chat composer.
+
 ### Improved
 
+- An empty Ask page shows the large composer, and the agent panel opens on the app's default model
+- The Content dashboard ranks articles by signups, adds Authors, New and rising, and Sources tabs, and names missing data sources in one notice.
 - Organization roster searches now find row-level dbt membership data more reliably.
 
 ### Fixed
 
+- Analytics now clearly indicates incomplete Gong user lists and HubSpot deal pages.
+- Date columns in dashboard tables no longer show the previous day in time zones west of UTC.
+- Gong transcript reads preserve the transcript alias for existing callers.
+- Phones can open the agent from the top bar on every page outside Ask
+- HubSpot and Gong searches keep their filters and page totals visible.
 - Analytics avoids showing partial warehouse results when a metric or plan request is ambiguous.
 - Analytics preserves verified results from successful time ranges when another requested range is unavailable.
 - Analytics uses a warehouse fallback for a failed metric only when requested and supported by a complete metric definition and result.
@@ -18,8 +29,16 @@ time from the command menu (Cmd+K → "What's new") or from Settings.
 
 ## 2026-10-23
 
+### Improved
+
+- Metric tiles can show their change from the period before, tables give the spare width to their first text column and right-align numbers, and callouts read as a list with a severity icon per row, in dashboards and emailed reports.
+
 ### Fixed
 
+- Content pageviews use their own GA4 property without changing other dashboards.
+- Content signup collection keeps actual query usage and the configured cap, including when collection fails
+- External agents need approval before changing recurring index builds.
+- Query cost tracking distinguishes unknown BigQuery usage from zero-byte results.
 - Analytics figures require complete, scope-matched warehouse evidence
 - Analytics keeps malformed metric values out of verified summaries
 - Analytics keeps verified metrics available when another requested dbt metric is unavailable
@@ -29,8 +48,23 @@ time from the command menu (Cmd+K → "What's new") or from Settings.
 - Analytics withholds warehouse figures when an owner-defined dbt metric query fails.
 - Warehouse figures stay hidden when a failed dbt query cannot verify the calculation.
 
+## 2026-10-22
+
+### Added
+
+- Tables, metric tiles and charts can show a number of seconds as a duration such as 16 min or 7.4 hours, and duration columns sort by time.
+
+### Improved
+
+- Approval links from a connected agent, such as Claude Code, now open the review inside the app, in its own theme.
+- Extension panels on dashboards are as tall as their content, so a one-line extension no longer leaves an empty band below it
+
 ## 2026-10-21
 
+### Added
+
+- You can pin a shared chat just for yourself, or hide it from your list and undo that from the toast.
+
 ### Improved
 
 - Analytics recognizes questions about connected data, product analytics, sessions, dashboards, and charts.
@@ -43,6 +77,10 @@ time from the command menu (Cmd+K → "What's new") or from Settings.
 
 ## 2026-10-20
 
+### Added
+
+- Custom SQL charts can open configured record details on click
+
 ### Fixed
 
 - Blog signup attribution excludes signups before the first visit and URLs on unrelated hosts

package/corpus/templates/analytics/actions/set-index-schedule.ts

@@ -5,6 +5,7 @@ import {
 } from "@agent-native/core/server";
 import {
   defineAutomation,
+  isOutsideAutomationOrigin,
   listAutomationDefinitions,
   updateAutomation,
 } from "@agent-native/core/triggers";
@@ -26,6 +27,7 @@ const INDEX_RUN_BODY = [
 ].join("\n");
 
 export default defineAction({
+  needsApproval: (_input, ctx) => isOutsideAutomationOrigin(ctx),
   mcpAnnotations: {
     readOnlyHint: false,
     destructiveHint: false,
@@ -65,7 +67,14 @@ export default defineAction({
     const current = await readIndexSchedule(admin.orgId);
     const name =
       current.automationName ?? DEFAULT_INDEX_SCHEDULE_AUTOMATION_NAME;
-    const actor = { userEmail: admin.userEmail, orgId: admin.orgId, appId };
+    const actor = {
+      userEmail: admin.userEmail,
+      orgId: admin.orgId,
+      appId,
+      caller: ctx.caller,
+      networkProtocol: ctx.networkProtocol,
+      automation: ctx.automation,
+    };
     const definitions = await listAutomationDefinitions(actor, "organization");
     const exists = definitions.some((definition) => definition.name === name);
 

package/corpus/templates/analytics/app/components/dashboard/SqlChart.tsx

@@ -37,7 +37,7 @@ import {
   type ChartPointLimit,
   type HeatmapKeys,
 } from "@shared/panel-render-contract";
-import { numberTableColumns } from "@shared/table-columns";
+import { numberTableColumns, tableQuantity } from "@shared/table-columns";
 import {
   FUNNEL_OUTCOME_FILLS,
   FUNNEL_OUTCOME_SEGMENTS,
@@ -2480,6 +2480,8 @@ export function sortTableRows(
   rows: Record<string, unknown>[],
   sorts: TableSort[],
   columns: TableColumnConfig[] = [],
+  // The columns that line up as numbers, when the caller already has them.
+  numberColumns?: ReadonlySet<string>,
 ): Record<string, unknown>[] {
   if (sorts.length === 0) return rows;
 
@@ -2489,8 +2491,30 @@ export function sortTableRows(
   const isEmpty = (key: string, value: unknown) =>
     formats.get(key) === "duration" ? isEmptyDuration(value) : value == null;
 
+  // Text that reads as a number sorts by it, and a duration such as
+  // "4.8 days" by how long it is, in a column that lines up as numbers.
+  const numeric =
+    numberColumns ??
+    numberTableColumns(
+      sorts.map(
+        (sort) =>
+          columns.find((col) => col.key === sort.key) ?? { key: sort.key },
+      ),
+      rows,
+    );
+  const quantity = (key: string, value: unknown) =>
+    numericTableValue(value) ??
+    (numeric.has(key) ? tableQuantity(value) : null);
+
   return rows
-    .map((row, index) => ({ row, index }))
+    .map((row, index) => ({
+      row,
+      index,
+      // Read each sorted cell once, not on every comparison.
+      quantities: new Map<string, number | null>(
+        sorts.map((sort) => [sort.key, quantity(sort.key, row[sort.key])]),
+      ),
+    }))
     .sort((a, b) => {
       for (const sort of sorts) {
         const av = a.row[sort.key];
@@ -2501,8 +2525,8 @@ export function sortTableRows(
         if (aEmpty) return 1;
         if (bEmpty) return -1;
 
-        const an = numericTableValue(av);
-        const bn = numericTableValue(bv);
+        const an = a.quantities.get(sort.key) ?? null;
+        const bn = b.quantities.get(sort.key) ?? null;
 
         if (an !== null && bn === null) return -1;
         if (an === null && bn !== null) return 1;
@@ -3070,11 +3094,17 @@ function TableRenderer({
   const [requestedPage, setPage] = useState(0);
   const [pageSize, setPageSize] = useState(DEFAULT_TABLE_PAGE_SIZE);
 
+  // Formats, not widths, decide these, so resizing a column doesn't redo it.
+  const numberColumns = useMemo(
+    () => numberTableColumns(savedColumns, limitedRows),
+    [savedColumns, limitedRows],
+  );
+
   const sortedRows = useMemo(() => {
     return sortable
-      ? sortTableRows(limitedRows, sortColumns, savedColumns)
+      ? sortTableRows(limitedRows, sortColumns, savedColumns, numberColumns)
       : limitedRows;
-  }, [limitedRows, sortColumns, sortable, savedColumns]);
+  }, [limitedRows, sortColumns, sortable, savedColumns, numberColumns]);
 
   const pageCount = Math.ceil(sortedRows.length / pageSize);
   // A refresh can return fewer rows than the page a viewer was on; show the
@@ -3082,10 +3112,6 @@ function TableRenderer({
   const page = Math.min(requestedPage, Math.max(pageCount - 1, 0));
   const displayRows = sortedRows.slice(page * pageSize, (page + 1) * pageSize);
 
-  const numberColumns = useMemo(
-    () => numberTableColumns(columns, limitedRows),
-    [columns, limitedRows],
-  );
   const fillKey = tableFillColumn(columns, numberColumns);
   // The fill column's max-width of 0 drops its content from the table's
   // minimum width, so it takes only the spare width; a width set on it (saved

package/corpus/templates/analytics/changelog/2026-10-22-approval-links-from-a-connected-agent-such-as-claude-code-no.md

@@ -0,0 +1,6 @@
+---
+type: improved
+date: 2026-10-22
+---
+
+Approval links from a connected agent, such as Claude Code, now open the review inside the app, in its own theme.

package/corpus/templates/analytics/changelog/2026-10-23-external-agents-need-approval-before-changing-recurring-inde.md

@@ -0,0 +1,6 @@
+---
+type: fixed
+date: 2026-10-23
+---
+
+External agents need approval before changing recurring index builds.

package/corpus/templates/analytics/package.json

@@ -14,6 +14,11 @@
     "typecheck": "agent-native typecheck",
     "migrate:production": "tsx scripts/migrate-production.ts",
     "journey:capture": "tsx scripts/journey-capture.ts",
+    "radar:query": "tsx scripts/radar/query.ts",
+    "radar:watch": "tsx scripts/radar/watch.ts",
+    "radar:triage": "tsx scripts/radar/triage.ts",
+    "radar:ab": "tsx scripts/radar/ab.ts",
+    "radar:report": "tsx scripts/radar/report.ts",
     "journey:snapshot": "tsx scripts/journey-snapshot.ts",
     "source-index": "tsx scripts/build-source-index.ts",
     "backfill:first-party-bigquery": "tsx scripts/first-party-analytics-bigquery-backfill.ts",