plans: lead decision 97 and the Console reply-view brief (#1570, #1575)

Row 76 builds as briefed; its brief's 'replies show in the inbox' line was wrong. A read-only bus read of replies, shown on the Console receipt, is row 81 (#1575), held behind the cutover. Option (b), a Console message.receive, is refused because receive marks delivery. Co-Authored-By: Claude Opus 5.5 <[email protected]>
2026-10-11 18:52:25 -05:00
parent 36cd6bd12b
commit e16fb57849
2 changed files with 106 additions and 0 deletions
@@ -2101,3 +2101,28 @@ which stay with him. Each item names who decided it and what happened.
      the faster option. In exchange, the in-stack Mos never runs with
      its wider authority under Jason's user, with PID-only isolation,
      for more than the trial.
+
+97. **Row 76 builds as briefed; the reply view is its own row after the
+    cutover (2026-10-11).** Dewey found, building row 76 (#1570), that the
+    brief's "replies show in the inbox as today" is wrong. The bus inbox
+    lists only open decisions, and nothing reads `in_reply_to`. Sage wrote
+    that line; the error is Sage's.
+    - Row 76 goes ahead with Dewey's option (a). Its README says the PM's
+      reply to a web send shows on Jason's next `mosaic talk`, which
+      prints every undelivered message to the human first
+      (`packages/cli/src/cli.mjs`, `talk`).
+    - Option (b), a Console `message.receive` with no `message.read`, is
+      refused. `message.receive` writes a `delivered` row, so the Console
+      would take the reply away from `mosaic talk`. It is also a bus
+      change, which row 76's brief puts out of scope.
+    - The web send keeps the broker's default class, INFO. The brief puts
+      other classes out of scope, and the runner answers any class but
+      RESULT.
+    - Row 81 (#1575): a read-only bus read of the replies to a message,
+      shown on the Console receipt, plus the CLI talk timeout message,
+      which also names `mosaic inbox` wrongly. Owner Dewey after rows 76
+      and 59, reviewer Darkwing. Held behind the cutover with the other
+      bus rows (decision 94).
+    - The tradeoff: on cutover day Jason sends from the browser and reads
+      the answer in the CLI. In exchange no broker change lands before the
+      cutover and row 76 needs no extra review round.
@@ -0,0 +1,81 @@
+# Console: show the reply to a web send (2026-10-11)
+
+Status: written by Sage, lead. It follows `docs/plans/BRIEF-TEMPLATE.md`.
+From Dewey's question while building row 76 (#1570). Lead decision 97.
+The row waits until after the dogfood cutover, like the other bus rows
+(lead decision 94).
+
+## Show replies to a message without receiving them
+
+### Problem
+
+Row 76's brief said a web send's replies "show in the inbox as today".
+That was wrong. Dewey checked the code on 2026-10-11 and Sage confirmed
+it:
+
+- The bus `inbox` read returns only open decisions routed to the human
+  (`packages/bus/src/broker.mjs`, the `inbox` case). A message to the
+  human never appears there.
+- The runner's reply is a RESULT `message.send` to `human` with
+  `in_reply_to` set (`packages/harness/src/runner.mjs`).
+- `trail` on a message id returns that message, its events and its
+  deliveries. Nothing reads `in_reply_to`, so replies are not in it.
+- `message.receive` writes a `delivered` row for each message it returns.
+  A Console that received the human's messages would take them away from
+  `mosaic talk`, which prints only undelivered ones.
+
+So after a web send, Jason sees the receipt and the delivery state but not
+the PM's answer. The answer shows up on his next `mosaic talk`, which
+prints every undelivered message to the human before it waits. The CLI's
+timeout message (`packages/cli/src/cli.mjs`, the `talk` function) says
+"the next mosaic talk or mosaic inbox shows it", and the inbox half is
+wrong for the same reason.
+
+### Owner and reviewer
+
+Owner: Dewey, after row 76 and row 59 (both edit files this row reads or
+changes). Reviewer: Darkwing.
+
+### Files owned
+
+- `packages/bus/src/broker.mjs`, a read for replies only
+- `packages/bus/src/views.mjs`, if the read passes through it
+- `packages/bus/tests/`, `packages/bus/README.md`, the reads section only
+- `packages/webui/src/`, `packages/webui/README.md`,
+  `packages/webui/tests/`
+- `packages/cli/src/cli.mjs`, the `talk` timeout message only, and
+  `packages/cli/tests/`
+- `agents/dewey/work/queue-<row>/`
+
+### What ships
+
+- A bus read that, given a message id, returns the messages whose
+  `in_reply_to` is that id, oldest first, at most 50. It writes no
+  delivery or read row and changes nothing. Dewey picks whether it is a
+  new read verb or a field on `trail`, and says why in the packet.
+- The read is open to the same sessions as the other reads (human and
+  reader). An id that names no message in the business refuses
+  `message-not-found`, as `message.send` does for `in_reply_to`. (`trail`
+  returns an empty list instead; this read doesn't copy that.)
+- The Console receipt for a web send shows the replies that read
+  returns, refreshed with the page. A reply shown there still reaches
+  `mosaic talk` unchanged.
+- The CLI timeout message names only `mosaic talk`.
+- The webui README says where a reply shows and that the Console never
+  marks one delivered or read.
+- Tests: a reply listed and its delivery rows unchanged; a message with
+  no reply; the cap and order; an unknown id; `mosaic talk` still
+  receives a reply after the Console has shown it; the new timeout
+  message.
+
+### Out of scope
+
+Receiving or marking messages from the Console, any write verb, the
+human transport, the `deliveries` schema, decisions in the browser
+(REQ-DEC-3) and conversations with T3 seats (CHAT-07).
+
+### Gate
+
+Darkwing approves on the row's issue naming the candidate manifest. The
+bus, webui, control-board and cli node suites and every
+`scripts/test-*.sh` green on Sage's gate rerun.