Trim negative-space comments per mara's review

Don't state what a function/module doesn't do and where that happens instead — just describe what it does. Cut the "not something this function decides" / "not affected by this" / "not a placeholder for a later commit" asides from the doc comments touched in the last two commits.
Fix stale default-open doc comments per argus's review
2026-08-30 21:31:33 +02:00 · 2026-08-30 21:28:34 +02:00 · 2026-08-30 21:23:28 +02:00 · 2026-08-30 21:11:25 +02:00
15 changed files with 822 additions and 864 deletions
--- a/docs/terminal-rendering.md
+++ b/docs/terminal-rendering.md
@ -1,249 +1,51 @@
 # Per-agent terminal: row taxonomy (as built)
-Snapshot of how the per-agent web UI's live pane renders each
+The per-agent web UI's live pane renders one row per `TermMsg`
-event kind today. The per-tool icon/summary/category (and, for a
+(`hive-agent/src/term_msg.rs`): `{icon?, level: debug|info|warn|error,
-few rich tools, the expandable body) are pre-computed server-side by
+summary, body?, body_format?: markdown|diff, coalesce_key?}`. Classification
-`hive-agent/src/stream_enrich.rs::enrich` and stamped onto the
+(icon, summary text, whether a tool call gets an expandable body) happens
-stream-json value as `_icon`/`_summary`/`_category`/`_body`/
+server-side, once — `term_msg.rs` + `stream_enrich.rs` — and is served
-`_body_type` before SSE delivery, so the frontend just dispatches on
+identically by both `GET /api/events/history` and `GET /api/events/stream`
-those fields instead of re-deriving them. Frontend source of truth
+as `TermEnvelope { ts, seq?, msgs: TermMsg[] }` frames
-lives in `frontend/packages/agent/src/app.js` (`renderStream`,
+(`hive-agent/src/web_ui/stream.rs`).
 `renderRichToolUse`, `renderToolResult`, `renderTaskEvent`,
 `mdNode`, `detailsOpenMd`) +
 `frontend/packages/shared/src/terminal/terminal.css` (the shared
 `.live .<class>` styling) + the `marked` npm package (markdown).
-## Layout contract
+The frontend renders a `TermMsg` close to as-is
 (`frontend/packages/agent/src/components/Row.tsx`): `level` picks the
 CSS colour (`terminal.css`'s `.live .level-*`), an empty `summary` +
 markdown `body` renders as a flat row with just the body (assistant
 text), and any other bodied row is an expandable `<details>`, opened by
 default according to the operator's expand-tool-output preference
 (`getExpandDetailsPref()`) — uniformly, no per-tool override. There's no
 separate client-side row model or classification step.
-Every row — flat `<div class="row …">` and expandable
+## Layout
 `<details class="row …">` alike — shares one prefix column.
 The mechanism is `padding-left + negative text-indent` on
 `.live .row`: the row's first inline box gets pulled back
 into the column at ~0.5em, and wrapped continuation lines
 hang under the body, not under the glyph.
-Rows that carry an icon (the per-tool emoji, `🧠`/`💭`
+Every row shares one prefix column via `padding-left` + negative
-thinking, etc.) pass it as the `icon` argument to
+`text-indent` on `.live .row`; an icon (when set) sits in a fixed-width
-`row()` / `details()` / `detailsDiff()`, which puts it in a
+`.row-glyph` cell so icons of different rendered widths still line up.
-fixed-width `.row-glyph` cell (`display: inline-block;
+`<details>` summaries reuse the same metrics, with the disclosure caret
-width: 1.4em`) rather than as a bare first character. The
+leading the summary text rather than the icon.
 constant cell width means every icon's left edge lines up in
 the column regardless of the glyph's rendered width (emoji
 differ; some carry a variation selector) — a flat row's `🧠`
 and a `details` summary's `🖥️` align. Rows with a plain
 single-char glyph (`◆ · ! ←`) still pass it inline; it lands
 at the same ~0.5em left edge.
-`<details>` summaries inherit those metrics. The icon (when
+## Levels
 present) sits in the `.row-glyph` cell; the summary text lives
 in a `.summary-text` span and the disclosure caret (`▸` / `▾`)
 is supplied by CSS `.summary-text::before` so it **leads the
 text, not the icon** — a leading caret on the icon would push
 it out of the shared column. Icon-less summaries have no
 `.row-glyph`, so the caret falls into the prefix column like
 the old directional glyph. The summary text carries no
 `→` / `←`; the row colour (cyan = outbound, muted = inbound)
 carries the direction.
-Child blocks inside a row (the `.md` markdown wrapper, an
+| Level | Colour | Roughly |
-inner `<details>`) get `text-indent: 0` so their content
+|---|---|---|
-lays out from the body column instead of inheriting the
+| `debug` | muted | thinking, coalesced ticks, ambient harness chatter |
-parent's negative pull.
+| `info` | default fg | turn start/ok, assistant text, tool calls + results |
 | `warn` | amber, left rule | stderr, an unrecognised event shape, API retries |
 | `error` | red, left rule | turn failed, a tool result with `is_error: true` |
-## Row taxonomy
+Per-tool icon + summary formatting (what a `Read`/`Edit`/`send` call's row
-
+actually says) lives in `stream_enrich.rs`'s `fmt_tool_use()` family —
-| CSS class | Prefix glyph | Color | Triggered by | Source |
+read that when you need the specifics, this doc doesn't duplicate it.
 |---|---|---|---|---|
 | `.turn-start` | `◆ TURN ← <from>` | amber, left rule | `LiveEvent::TurnStart` | harness wake |
 | `.turn-body` | (child div under turn-start) | fg | same | the wake-prompt body |
 | `.turn-end-ok` | `✅ turn ok` | green, left rule | `LiveEvent::TurnEnd { ok: true }` | harness |
 | `.turn-end-fail` | `❌ turn fail — note` | red, left rule | `LiveEvent::TurnEnd { ok: false }` | harness |
 | `.turn-time` | `· HH:MM:SS` on turn-start; `· HH:MM:SS · <dur>` on turn-end (child span) | muted, smaller | per-event `ts` (unix seconds) on the live frame + history row | harness |
 | `.text` | (no prefix; markdown body) | fg | claude `assistant.content[].text` | stream-json |
 | `.thinking` | `💭 thinking …` | muted, italic | claude `assistant.content[].thinking` | stream-json |
 | `.tool-use` (flat) | `<icon> Name args…` | cyan | tool_use w/o rich renderer; `<icon>` from the backend's `tool_icon(name)` (`stream_enrich.rs`): 📤 send · 📥 recv · ⏰ remind · 🏷️ set_status · 🪢 loose-ends · ✂️ cancel_loose_end · ℹ️ get_agent_meta · ✅ ack_until · 📜 get_logs/get_host_journal · ↻ restart · ⏹️ kill · ▶️ start · 🔄 update · 📋 list_containers/list_rooms/list_room_members/list_invites · 📖 read_room/Read · 👁️ mark_read · 🛑 bash kill · 🖥️ bash other · 💬 matrix send/reply/dm · 📦 request_* · ⏱️ schedule · 🔧 default | stream-json |
 | `.tool-use` `<details>` | `✏️ Edit <path> · -N +N` (no `→`) | cyan, body is +/- diff | `renderRichToolUse` Edit | stream-json |
 | `.tool-use` `<details open>` | `📤 send → to · NL` | cyan, body is markdown | rich renderer for send | stream-json |
 | `.tool-result` (flat) | `←  <txt>` | muted | short `tool_result` (≤120c, non-recv) | stream-json |
 | `.tool-result-block` `<details>` | `Nl · headline` | muted, body is text | long generic `tool_result` | stream-json |
 | `.tool-result-block` `<details open>` | `recv ← <txt>` | muted, body is markdown | `tool_result` correlated to a prior `recv` tool_use via id | stream-json |
 | `.tool-result.error` (flat) | `✗  <msg>` | red | `tool_result` with `is_error: true` (≤120c); `<tool_use_error>` wrapper stripped | stream-json |
 | `.tool-result-block.error` `<details>` | `Nl · headline` | red, body is text | long error `tool_result` (`is_error: true`); wrapper stripped | stream-json |
 | `.tool-use` | `⌁  task <id> started · <desc> [type]` | cyan | claude Task-tool subagent start (dead path — `Task` omitted from agent allow-list) | `renderTaskEvent` |
 | `.turn-end-ok` / `.turn-end-fail` / `.tool-result` | `⌁  task <id> ✓/✗/◌ <status> · <desc> · → <output_file>` | green / red / muted | claude Task-tool result (dead path for agents) | `renderTaskEvent` |
 | `.note` | `· ⚙ plugin install · loading…` or `✓ done` | muted | `system/plugin_install` (`status` = started/completed) | stream-json |
 | `.note` | `· ⚙ commands changed · N available` (expandable list of `/name` entries) | muted | `system/commands_changed` (slash-command set updated, usually post-plugin_install) | stream-json |
 | `.note` | `· ⚙ compact · <trigger> · <pre>→<post> tokens · <dur>` | muted | `system/compact_boundary` (compaction completed; metadata includes pre/post token counts, duration, trigger) | stream-json |
 | `.note` | `· ⚙ <subtype>` | muted | other `system` subtypes (context_window_exceeded, etc.) | stream-json catch-all |
 | `.note` | `· <text>` | muted | harness chatter | `LiveEvent::Note` |
 | `.note.stderr` | `! stderr: <line>` | amber/orange | stderr lines off claude | `LiveEvent::Note` (`text` starts `stderr:`) |
 | `.note.op` | `· operator: <text>` | mauve italic | operator-initiated notes (/cancel, /compact, /model, new-session) | `LiveEvent::Note` (`text` starts `operator:`) |
 | `.sys` | `! {json…}` | amber/orange | catch-all for stream shapes `renderStream` didn't classify | catch-all |
 | Banner shimmer | mauve | turn in flight (ref-counted) | — | `setBannerActive` |
 The `.turn-time` span is appended to the turn-start / turn-end rows from
 the event's `ts` (unix seconds), which the backend serializes as a
 flattened sibling of `kind` on both the live SSE frame and each history
 row — so the same renderer path stamps live tail and replayed scrollback
 identically. Turn-end also shows the elapsed duration (end − start),
 paired against the most recent open turn-start. The read is guarded on a
 numeric `ts`: if a frame omits it the rows render without the time
 suffix, so the terminal degrades cleanly against older event shapes.
 ## Renderer dispatch
 `renderStream(v, api)` walks each stream-json line. Most of the
 per-event classification it used to do itself is now pre-computed
 server-side by `hive-agent/src/stream_enrich.rs::enrich` (stamped
 onto the value as `_category`/`_summary`/`_icon`/`_body`/
 `_body_type` at SSE-emit time, for both the live tail and history
 replay) — the client mostly just dispatches on those fields rather
 than re-deriving them from raw claude field names:
 1. `v._category === 'drop'` → dropped without rendering. Covers the
   top-level `result` / `rate_limit_event` types (`result` powers the
   `cost` badge elsewhere) and the `system` subtypes `init` /
   `result` / `rate_limit_event`.
 1a. `system` events with `_category === 'thinking_tok'`
   (`subtype == "thinking_tokens"`; claude streams a running
   `estimated_tokens` counter while thinking — many per turn) →
   collapses into a **single** `🧠 thinking … ~N tokens` `.note`
   row that updates in place, text taken verbatim from the
   backend-computed `_summary`. Consecutive ticks reuse the row only
   while it's still the last one rendered (`nextElementSibling ==
   null`); any other event after it makes the next tick start a
   fresh row. Avoids a note-per-tick scrollback flood.
 1b. `system/plugin_install` (matched on `subtype`, not `_category`,
   so start/complete can coalesce into one row) → muted note
   `⚙ plugin install · loading…` (on `started`) or
   `⚙ plugin install · ✓ done` (on `completed`), text from
   `_summary`. Emitted in pairs: started fires before the plugin
   loads, completed fires when it's ready.
 1c. `system/status` (matched on `subtype`) → muted note from
   `_summary`, except while the harness's local `turn_state` is
   `compacting`: the client overrides the text with an elapsed-time
   counter (`⚙ compact · <N>s…`) computed client-side from
   `stateSince`, since the backend can't know client wall-clock time
   at emit time.
 1d. `_category === 'details'` (currently just `system/commands_changed`)
   → collapsible `.note` details row: `_summary` as the header
   (`⚙ commands changed · N available`), `_body` (one `/name` per
   line) as the expandable content.
 1e. Other `system/` subtypes (e.g. `compact_boundary`, `api_retry`,
   `api_error`, or an unrecognised subtype) → `_category === 'note'`,
   rendered as a single muted note from `_summary` — computed by
   `system_fields()` in `stream_enrich.rs` (e.g. `compact_boundary`
   → `⚙ compact · <trigger> · <pre>→<post> tokens · <dur>` with each
   field guarded individually; an unrecognised subtype falls back to
   `⚙ <subtype>`).
 2. `subtype == "task_started" | "task_notification"` →
   `renderTaskEvent` (subagent activity gets the `⌁` glyph).
 3. `type == "assistant"` → walk `message.content[]`:
   - `text` → `.text` row with a markdown body via `mdNode`.
   - `thinking` → `.thinking` row.
   - `tool_use` → record `id → name` in `toolNameById`. The backend
     stamps every `tool_use` entry with `_icon` + `_summary` (via
     `fmt_tool_use()` in `stream_enrich.rs` — see [salient-arg
     formatting](#salient-arg-formatting) below) and, for a fixed set
     of tools, `_category: "rich"` + `_body`/`_body_type`. When
     `_category === "rich"`, `renderRichToolUse` dispatches on
     `_body_type` (`"diff"` → `api.detailsDiff`, `"markdown"` →
     `detailsOpenMd`, else `api.details`) to build the expandable
     row; otherwise it falls through to a flat `.tool-use` row using
     `_icon` + `_summary` as-is — no per-tool JS.
 4. `type == "user"` → walk `message.content[]` for
   `tool_result`; `renderToolResult` correlates via
   `tool_use_id → toolNameById` to default-open `recv`
   results with a markdown body, else short = flat /
   long = collapsed details.
 5. Unrecognised shape → `.sys` row (amber, `!` glyph).
 ### Salient-arg formatting
 Server-side (`fmt_tool_use()` and its per-tool-family helpers in
 `hive-agent/src/stream_enrich.rs`), computed into `_summary` and
 read verbatim by the client. The `short` name strips the
 `mcp__hyperhive__` / `mcp__bash__` / `mcp__matrix__` prefix and
 appends `*` (e.g. `recv*`, `run*`, `send_message*`). Unprefixed
 tools (Read, Write, etc.) keep their name as-is.
 | Tool | Rendered as |
 |------|-------------|
 | **Claude built-ins** | |
 | `Read` | `Read <path>` |
 | `Write` | flat, same shape as Read: `Write <path>` — no diff/count (`content` can be megabytes and is one-sided; open the file to inspect it) |
 | `Edit` | rich diff row `Edit <path> · -N +N` (just `+N` for a pure insert, i.e. empty `old_string`) |
 | `Glob` | `Glob <pattern>` |
 | `Grep` | `Grep <pattern>` |
 | `Bash` | `Bash [bg] $ <cmd>` (dead path — built-in `Bash` isn't in the agent allow-list either; shell execution goes through `mcp__bash__run` / `run*` below instead) |
 | `TodoWrite` | `TodoWrite (N items)` (dead path — `TodoWrite` isn't in the agent allow-list; its state lives in claude's in-process session and evaporates on `/compact`, so agents plan in `/state` notes instead) |
 | **Core hyperhive** | |
 | `send*` | rich renderer: `send* → to · NL` (default-open body) |
 | `recv*` | `recv*()` · `recv* wait Ns` · `recv* max N` |
 | `remind*` | `remind* +Xm "preview"` or `remind* at HH:MMZ "preview"` |
 | `set_status*` | `set_status* "text"` |
 | `get_loose_ends*` | `get_loose_ends*()` or `get_loose_ends* [agent]` |
 | `get_agent_meta*` | `get_agent_meta*()` or `get_agent_meta* name` |
 | `cancel_loose_end*` | `cancel_loose_end* kind #id` |
 | `ack_until*` | `ack_until* ≤N` |
 | `mark_todos_done*` | `mark_todos_done* [id1, id2, …]` (first 8 ids, `…` past that) |
 | **Lifecycle** | |
 | `kill*/restart*/start*/update*` | `kill* name` (etc.) |
 | `get_logs*` | `get_logs* name` or `get_logs* name NL` |
 | `get_host_journal*` | `get_host_journal*()` or with `[container] · [/grep/] · NL` |
 | **Approvals / config** | |
 | `request_init_config*` | `request_init_config* name` |
 | `request_update_meta_inputs*` | `request_update_meta_inputs* [inp1, …]` or `all` |
 | **Scheduling** | |
 | `list_schedules*` | `list_schedules*()` |
 | `cancel_schedule*` | `cancel_schedule* #id all` or `#id [t1, t2]` |
 | `fire_schedule_now*` | `fire_schedule_now* #id` |
 | `edit_schedule*` | `edit_schedule* #id · body · interval · next · +N tgt · -N tgt` (only changed fields shown) |
 | `request_schedule_prompt*` | `request_schedule_prompt* → t1, t2 at HH:MMZ` (+ `+Ns` if recurring) |
 | **Bash MCP** | |
 | `run*` | `run* [bg] $ cmd` (also rich renderer for full cmd body) |
 | `status*` (bash) | `status* id:xyz` or `status* id:xyz · wait Ns` |
 | `kill*` (bash) | `kill* id` or `kill* id [force]` |
 | **Matrix MCP** | |
 | `send_message*/send_dm*/send_reply*` | `send_message* → room: "body"` / `send_dm* → @user: "body"` |
 | `send_reaction*` | `send_reaction* room emoji` |
 | `read_room*` | `read_room* room` or `read_room* room [N]` |
 | `mark_read*` | `mark_read* room` |
 | `join_room*/open_dm*` | `join_room* room` / `open_dm* @user` |
 | `invite_user*` | `invite_user* @user → room` |
 | `download_file*` | `download_file* room` |
 | **Everything else** | `fmt_args_generic` — see [Extra-MCP tools](#extra-mcp-tools) |
 ## Markdown
-`mdNode(text)` wraps `marked.parse(text)` (the `marked` npm dep,
+`frontend/packages/agent/src/lib/markdown.ts`'s `renderMarkdown()` runs
-bundled by esbuild into the page's `app.js`) in a `<div
+`marked.parse()` through DOMPurify into a `<div class="md">`. Applied to
-class="md">`. CSS in `terminal.css` scopes paragraph / code /
+any `TermMsg` with `body_format: "markdown"`.
 list / blockquote / link styling under `.live .row .md` so
 the markdown body doesn't bleed into the row's own
 text-indent. Falls back to plain text if `marked` didn't
 load. Applied to `text` rows and to send / recv message bodies.
 ## Extra-MCP tools
 `fmt_args_generic(name, input)` (`hive-agent/src/stream_enrich.rs`)
 is the fallback when a tool isn't in the built-in `fmt_tool_use`
 switch, computed into `_summary` server-side:
 - single string field → `name k: "v"`
 - single number/bool field → `name k: v`
 - multi-field → first 4 pairs trimmed to `k: "v"` /
  `k: [N]` / `k: {…}` with a `…+N` overflow
 This keeps less-frequent tools that don't have a specific
 `fmt_tool_use` case from dumping raw JSON. Common matrix and
 hyperhive tools have their own cases and skip this path.
 ## Dashboard side (not covered here)
-The main dashboard's message-flow pane is a different
+The main dashboard's message-flow pane is a different shape: broker
-shape: broker messages render as `.msgrow` grid lines (ts /
+messages render as `.msgrow` grid lines, not agent-terminal rows.
 arrow / from / → / to / body) with their own styling.
 `.live .msgrow` explicitly resets `text-indent: 0` so the
 per-agent terminal's hanging-indent metrics don't leak into
 the flex-grid broker rows.
--- a/frontend/packages/agent/src/Root.tsx
+++ b/frontend/packages/agent/src/Root.tsx
@ -240,7 +240,7 @@ export function Root() {
        {state.status === 'needs_login_idle' || state.status === 'needs_login_in_progress' ? (
          <LoginFlow status={state.status} session={state.session} onRefresh={refresh} />
        ) : null}
-        <LiveStream ref={liveStreamRef} onLiveTurnBoundary={refresh} />
+        <LiveStream ref={liveStreamRef} />
      </main>
      {termInput}
      {panel}
--- a/frontend/packages/agent/src/components/LiveStream.tsx
+++ b/frontend/packages/agent/src/components/LiveStream.tsx
@ -28,12 +28,6 @@ import { Row } from './Row.js';
 const NEAR_BOTTOM_PX = 48;
 const LOAD_MORE_SCROLL_PX = 80;
 export interface LiveStreamProps {
  /** Forwarded to useLiveStream — fires on live turn_start/turn_end so
   * the caller can refresh `/api/state` sooner than its poll interval. */
  onLiveTurnBoundary?: () => void;
 }
 /** Imperative escape hatch for TermInput's local-only slash commands
 * (`/help`, `/clear`) — same shape as app.js's old `termAPI` object,
 * kept as a ref handle rather than lifting the whole row array up to
@ -43,11 +37,8 @@ export interface LiveStreamHandle {
  clear: () => void;
 }
-export const LiveStream = forwardRef<LiveStreamHandle, LiveStreamProps>(function LiveStream(
+export const LiveStream = forwardRef<LiveStreamHandle, object>(function LiveStream(_props, ref) {
-  { onLiveTurnBoundary },
+  const { rows, hasMore, loadingMore, loadMore, pushLocalNote, clearLocal } = useLiveStream();
  ref,
 ) {
  const { rows, hasMore, loadingMore, loadMore, pushLocalNote, clearLocal } = useLiveStream({ onLiveTurnBoundary });
  useImperativeHandle(ref, () => ({ pushNote: pushLocalNote, clear: clearLocal }), [pushLocalNote, clearLocal]);
  const logRef = useRef<HTMLDivElement>(null);
  const [stickToBottom, setStickToBottom] = useState(true);
--- a/frontend/packages/agent/src/components/Row.tsx
+++ b/frontend/packages/agent/src/components/Row.tsx
@ -1,13 +1,16 @@
-// Renders one `StreamRow` — flat `<div class="row …">` or expandable
+// Renders one `TermRow` — flat `<div class="row …">` or expandable
-// `<details class="row …">`, matching @hive/shared/terminal.css's
+// `<details class="row …">`, driven straight off the wire shape
-// existing row-kind classes exactly (see docs/terminal-rendering.md).
+// (`lib/termMsg.ts`'s `TermMsg`, mirroring hive-agent's `term_msg.rs`):
-// Reuses that stylesheet as-is (imported once by LiveStream.tsx) — the
+// `level` picks the colour class, an empty `summary` + markdown `body`
-// taxonomy's visual language isn't what mara asked to change, the
+// is a flat row with just the body (assistant text), anything else with
-// component *model* underneath it is.
+// a body is an expandable details row gated by the operator's
 // expand-tool-output preference. No separate classification step —
 // mara: "StreamRow should now match what the server sends in TermMsg."
 import { useEffect, useRef } from 'preact/hooks';
-import type { StreamRow } from '../lib/streamRow.js';
+import type { TermRow } from '../lib/termMsg.js';
 import { linkifyToNodes } from '../lib/linkify.js';
 import { renderMarkdown } from '../lib/markdown.js';
 import { getExpandDetailsPref } from '@hive/shared/prefs.js';
 function MarkdownBody({ text }: { text: string }) {
  const ref = useRef<HTMLDivElement>(null);
@ -41,31 +44,39 @@ function DiffBody({ text }: { text: string }) {
  );
 }
-export function Row({ row }: { row: StreamRow }) {
+export function Row({ row }: { row: TermRow }) {
-  if (row.details) {
+  const cssClass = 'level-' + row.level;
  const icon = row.icon != null && row.icon !== '' && <span className="row-glyph">{row.icon}</span>;
  if (row.body == null) {
    return (
-      <details className={`row ${row.cssClass}`} open={row.defaultOpen || undefined}>
+      <div className={`row ${cssClass}`}>
-        <summary>
+        {icon}
-          {row.icon != null && row.icon !== '' && <span className="row-glyph">{row.icon}</span>}
+        {linkifyToNodes(row.summary)}
          <span className="summary-text">{row.text}</span>
        </summary>
        {row.diffBody != null && <DiffBody text={row.diffBody} />}
        {row.markdownBody != null && <MarkdownBody text={row.markdownBody} />}
        {row.plainBody != null && <pre className="tool-body">{linkifyToNodes(row.plainBody)}</pre>}
      </details>
    );
  }
  return (
    <div className={`row ${row.cssClass}`}>
      {row.icon != null && row.icon !== '' && <span className="row-glyph">{row.icon}</span>}
      {row.text != null && linkifyToNodes(row.text)}
      {row.meta?.map((m) => (
        <span key={m.cls} className={m.cls}>
          {m.text}
        </span>
      ))}
      {row.childText != null && <div className={row.childText.cls}>{row.childText.text}</div>}
      {row.markdownBody != null && <MarkdownBody text={row.markdownBody} />}
      </div>
    );
  }
  // Empty summary + markdown body → the body itself is the whole row
  // (assistant text), no summary prefix line, never collapsible.
  if (row.body_format === 'markdown' && row.summary === '') {
    return (
      <div className={`row ${cssClass}`}>
        {icon}
        <MarkdownBody text={row.body} />
      </div>
    );
  }
  return (
    <details className={`row ${cssClass}`} open={getExpandDetailsPref() || undefined}>
      <summary>
        {icon}
        <span className="summary-text">{row.summary}</span>
      </summary>
      {row.body_format === 'diff' && <DiffBody text={row.body} />}
      {row.body_format === 'markdown' && <MarkdownBody text={row.body} />}
      {row.body_format == null && <pre className="tool-body">{linkifyToNodes(row.body)}</pre>}
    </details>
  );
 }
--- a/frontend/packages/agent/src/hooks/useAgentState.ts
+++ b/frontend/packages/agent/src/hooks/useAgentState.ts
@ -1,16 +1,5 @@
-// useAgentState — polls `GET /api/state` and exposes the latest
+// useAgentState — polls `GET /api/state` every 4s and exposes the latest
-// snapshot + loading/error status. Mirrors app.js's old `refreshState`
+// snapshot + loading/error status.
 // data fetch (not yet its exact re-poll cadence — see the interval
 // comment below).
 //
 // Cadence note: the old page only re-polls on a timer while a login is
 // in flight, and otherwise waits for an SSE `turn_end` event to trigger
 // one-shot refreshes — this avoids clobbering the operator's half-typed
 // message in the term-input field. That field doesn't exist in this
 // rewrite yet (a later commit on this same PR), so there's nothing to
 // clobber yet; this hook uses a flat interval for now and switches to
 // the SSE-triggered model in the commit that adds TermInput + the live
 // stream, matching the original behavior once it's actually needed.
 import { useEffect, useRef, useState } from 'preact/hooks';
 import type { AgentState } from '../types.js';
--- a/frontend/packages/agent/src/hooks/useLiveStream.ts
+++ b/frontend/packages/agent/src/hooks/useLiveStream.ts
@ -1,37 +1,30 @@
 // Backfill + live SSE for the agent's event stream, reduced to a plain
-// `StreamRow[]` — the Preact-data sibling of
+// `TermRow[]` — the Preact-data sibling of @hive/shared/terminal.js's
-// @hive/shared/terminal.js's `create()`. Scroll behaviour is
+// `create()`. Scroll behaviour is deliberately NOT this hook's job (see
-// deliberately NOT this hook's job (see components/LiveStream.tsx):
+// components/LiveStream.tsx): mara flagged the old page's scroll-while-
-// mara flagged the old page's scroll-while-streaming bug explicitly as
+// streaming bug explicitly as something not to copy 1:1, and keeping
-// something not to copy 1:1, and keeping "what rows exist" separate
+// "what rows exist" separate from "where the viewport is" is what makes
-// from "where the viewport is" is what makes that fixable — this hook
+// that fixable — this hook only ever appends/prepends to `rows`, the
-// only ever appends/prepends to `rows`, the DOM-owning component
+// DOM-owning component decides whether that should move the scroll
-// decides whether that should move the scroll position.
+// position.
 //
 // Same subscribe → buffer → fetch-history → seq-dedupe → flush dance as
-// terminal.js's `start()`: a live event landing between EventSource-open
+// terminal.js's `start()`: a live envelope landing between EventSource-
-// and the history response resolving is buffered, not dropped or
+// open and the history response resolving is buffered, not dropped or
-// double-counted (seq <= history.seq AND the event's kind appeared in
+// double-counted (`envelope.seq <= history.seq` → already covered by the
-// the history replay → already covered, drop it from the buffer).
+// initial history page, drop it from the buffer). `seq` alone is the
 // whole dedup signal, see `TermEnvelope`'s doc in hive-agent's
 // `web_ui/stream.rs`.
 import { useEffect, useRef, useState } from 'preact/hooks';
-import { classifyEvent, createClassifyCtx, type ClassifyCtx } from '../lib/classifyEvent.js';
+import type { TermEnvelope, TermRow } from '../lib/termMsg.js';
 import type { StreamRow } from '../lib/streamRow.js';
 // eslint-disable-next-line @typescript-eslint/no-explicit-any -- raw SSE payload, dynamically shaped
 type AnyEvent = any;
 export interface UseLiveStreamOptions {
  historyUrl?: string;
  streamUrl?: string;
  /** Fired on every LIVE (not history-replay) turn_start/turn_end, so the
   * caller can trigger a snappier `/api/state` refresh than the plain
   * poll interval — mirrors app.js's turn_end → refreshState()/
   * refreshTodos(). */
  onLiveTurnBoundary?: () => void;
 }
 export interface UseLiveStreamResult {
-  rows: StreamRow[];
+  rows: TermRow[];
  hasMore: boolean;
  loadingMore: boolean;
  loadMore: () => void;
@ -44,12 +37,12 @@ export interface UseLiveStreamResult {
  clearLocal: () => void;
 }
-function appendRow(rows: StreamRow[], row: StreamRow): StreamRow[] {
+function appendRow(rows: TermRow[], row: TermRow): TermRow[] {
  const last = rows[rows.length - 1];
  // Coalesce in place only while the coalescible row is still the last
  // one in the list — any other row landing in between starts a fresh
  // one, same rule as terminal.js's makeCoalescer.
-  if (row.coalesceKey && last && last.coalesceKey === row.coalesceKey) {
+  if (row.coalesce_key && last && last.coalesce_key === row.coalesce_key) {
    const next = rows.slice(0, -1);
    next.push({ ...row, key: last.key });
    return next;
@ -57,59 +50,60 @@ function appendRow(rows: StreamRow[], row: StreamRow): StreamRow[] {
  return [...rows, row];
 }
 function appendMany(rows: StreamRow[], newRows: StreamRow[]): StreamRow[] {
  let next = rows;
  for (const r of newRows) next = appendRow(next, r);
  return next;
 }
 export function useLiveStream(opts: UseLiveStreamOptions = {}): UseLiveStreamResult {
  const historyUrl = opts.historyUrl ?? 'events/history';
  const streamUrl = opts.streamUrl ?? 'events/stream';
-  const [rows, setRows] = useState<StreamRow[]>([]);
+  const [rows, setRows] = useState<TermRow[]>([]);
  const [hasMore, setHasMore] = useState(false);
  const [loadingMore, setLoadingMore] = useState(false);
-  const ctxRef = useRef<ClassifyCtx | null>(null);
+  const keySeqRef = useRef(0);
-  if (!ctxRef.current) ctxRef.current = createClassifyCtx();
+  function nextKey(): string {
    keySeqRef.current += 1;
    return 'r' + keySeqRef.current;
  }
  function toRows(env: TermEnvelope, fromHistory: boolean): TermRow[] {
    return env.msgs.map((m) => ({ ...m, key: nextKey(), fromHistory }));
  }
  function appendEnvelope(rows: TermRow[], env: TermEnvelope, fromHistory: boolean): TermRow[] {
    let next = rows;
    for (const row of toRows(env, fromHistory)) next = appendRow(next, row);
    return next;
  }
  const minIdRef = useRef<number | null>(null);
  const liveRef = useRef(false);
-  const bufferedRef = useRef<AnyEvent[]>([]);
+  const bufferedRef = useRef<TermEnvelope[]>([]);
  const onBoundaryRef = useRef(opts.onLiveTurnBoundary);
  onBoundaryRef.current = opts.onLiveTurnBoundary;
  useEffect(() => {
    let cancelled = false;
-    function pushLive(ev: AnyEvent) {
+    function pushLive(env: TermEnvelope) {
-      const newRows = classifyEvent(ev, false, ctxRef.current!);
+      if (env.msgs.length) setRows((prev) => appendEnvelope(prev, env, false));
      if (newRows.length) setRows((prev) => appendMany(prev, newRows));
      if (ev.kind === 'turn_start' || ev.kind === 'turn_end') onBoundaryRef.current?.();
    }
    const es = new EventSource(streamUrl);
    es.onmessage = (e) => {
-      let ev: AnyEvent;
+      let env: TermEnvelope;
      try {
-        ev = JSON.parse(e.data);
+        env = JSON.parse(e.data);
      } catch {
        setRows((prev) => appendRow(prev, {
-          key: 'parse-err-' + Date.now(), cssClass: 'note', fromHistory: false,
+          key: 'parse-err-' + Date.now(), level: 'warn', summary: '[parse err] ' + e.data, fromHistory: false,
          text: '[parse err] ' + e.data,
        }));
        return;
      }
      if (!liveRef.current) {
-        bufferedRef.current.push(ev);
+        bufferedRef.current.push(env);
        return;
      }
-      pushLive(ev);
+      pushLive(env);
    };
    es.onerror = () => {
      setRows((prev) => appendRow(prev, {
-        key: 'conn-note', cssClass: 'note', fromHistory: false, coalesceKey: 'conn-status',
+        key: 'conn-note', level: 'warn', fromHistory: false, coalesce_key: 'conn-status',
-        text: es.readyState === 0 /* CONNECTING */ ? '[reconnecting…]' : '[disconnected]',
+        summary: es.readyState === 0 /* CONNECTING */ ? '[reconnecting…]' : '[disconnected]',
      }));
    };
@ -118,30 +112,29 @@ export function useLiveStream(opts: UseLiveStreamOptions = {}): UseLiveStreamRes
        const resp = await fetch(historyUrl);
        if (!resp.ok) throw new Error('http ' + resp.status);
        const body = await resp.json();
-        const events: AnyEvent[] = Array.isArray(body) ? body : body.events || [];
+        const events: TermEnvelope[] = Array.isArray(body) ? body : body.events || [];
        const boundarySeq: number | null = Array.isArray(body) ? null : (body.seq ?? null);
        if (!Array.isArray(body)) {
          setHasMore(!!body.has_more);
          if (typeof body.min_id === 'number') minIdRef.current = body.min_id;
        }
        const historyKinds = new Set(events.map((e) => e.kind));
        if (cancelled) return;
-        let initial: StreamRow[] = [];
+        let initial: TermRow[] = [];
-        for (const ev of events) initial = appendMany(initial, classifyEvent(ev, true, ctxRef.current!));
+        for (const env of events) initial = appendEnvelope(initial, env, true);
        initial = events.length
-          ? appendRow(initial, { key: 'live-sep', cssClass: 'note', fromHistory: true, text: '─── live (older above) ───' })
+          ? appendRow(initial, { key: 'live-sep', level: 'debug', fromHistory: true, summary: '─── live (older above) ───' })
-          : [{ key: 'placeholder', cssClass: 'note', fromHistory: true, text: '(connected — waiting for events)' }];
+          : [{ key: 'placeholder', level: 'debug', fromHistory: true, summary: '(connected — waiting for events)' }];
        setRows(initial);
        const drained = bufferedRef.current;
        bufferedRef.current = [];
        liveRef.current = true;
-        for (const ev of drained) {
+        for (const env of drained) {
-          if (boundarySeq != null && typeof ev.seq === 'number' && ev.seq <= boundarySeq && historyKinds.has(ev.kind)) {
+          // Already covered by the initial history page — drop it from
-            continue;
+          // the buffer rather than rendering it twice.
-          }
+          if (boundarySeq != null && typeof env.seq === 'number' && env.seq <= boundarySeq) continue;
-          pushLive(ev);
+          pushLive(env);
        }
      } catch (err) {
        console.warn('history backfill failed', err);
@ -149,7 +142,7 @@ export function useLiveStream(opts: UseLiveStreamOptions = {}): UseLiveStreamRes
        const drained = bufferedRef.current;
        bufferedRef.current = [];
        liveRef.current = true;
-        for (const ev of drained) pushLive(ev);
+        for (const env of drained) pushLive(env);
      }
    }
    backfill();
@ -168,14 +161,14 @@ export function useLiveStream(opts: UseLiveStreamOptions = {}): UseLiveStreamRes
      const resp = await fetch(historyUrl + sep + 'before=' + minIdRef.current);
      if (!resp.ok) return;
      const body = await resp.json();
-      const events: AnyEvent[] = Array.isArray(body) ? body : body.events || [];
+      const events: TermEnvelope[] = Array.isArray(body) ? body : body.events || [];
      setHasMore(!!body.has_more);
      if (typeof body.min_id === 'number') minIdRef.current = body.min_id;
      if (events.length) {
-        let older: StreamRow[] = [];
+        let older: TermRow[] = [];
-        for (const ev of events) older = appendMany(older, classifyEvent(ev, true, ctxRef.current!));
+        for (const env of events) older = appendEnvelope(older, env, true);
        older = appendRow(older, {
-          key: 'older-sep-' + minIdRef.current, cssClass: 'note', fromHistory: true, text: '─── older above ───',
+          key: 'older-sep-' + minIdRef.current, level: 'debug', fromHistory: true, summary: '─── older above ───',
        });
        setRows((prev) => [...older, ...prev]);
      }
@ -186,10 +179,8 @@ export function useLiveStream(opts: UseLiveStreamOptions = {}): UseLiveStreamRes
    }
  }
  const localKeyRef = useRef(0);
  function pushLocalNote(text: string) {
-    localKeyRef.current += 1;
+    setRows((prev) => appendRow(prev, { key: 'local-' + nextKey(), level: 'info', fromHistory: false, summary: text }));
    setRows((prev) => appendRow(prev, { key: `local-${localKeyRef.current}`, cssClass: 'note', fromHistory: false, text }));
  }
  function clearLocal() {
    setRows([]);
--- a/frontend/packages/agent/src/lib/classifyEvent.ts
+++ b/frontend/packages/agent/src/lib/classifyEvent.ts
@ -1,268 +0,0 @@
 // Turns one raw harness/stream-json event into zero or more `StreamRow`s.
 // Preact-data port of app.js's `renderStream` / `renderRichToolUse` /
 // `renderToolResult` / `renderTaskEvent` — see docs/terminal-rendering.md
 // for the row taxonomy this mirrors. Almost all per-tool classification
 // is already done server-side (`hive-agent/src/stream_enrich.rs::enrich`
 // stamps `_icon`/`_summary`/`_category`/`_body`/`_body_type`), so this
 // mostly just dispatches on those fields rather than re-deriving them.
 //
 // One deliberate simplification vs. app.js: the client-side "compacting…
 // <N>s" live override on `system/status` ticks (computed from the
 // harness's local `stateSince`) isn't ported — this just shows the
 // backend's `_summary` as-is. `turn_state`/`state_since` (and therefore
 // the "compacting" badge itself) still update correctly via
 // useAgentState's poll; only that one status row's live elapsed-seconds
 // text loses its client-side tick. Revisit if that's missed in practice.
 import type { StreamRow, StreamRowMeta } from './streamRow.js';
 import { fmtAge, fmtClock } from './format.js';
 import { getExpandDetailsPref } from '@hive/shared/prefs.js';
 // eslint-disable-next-line @typescript-eslint/no-explicit-any -- stream-json
 // content is dynamically-shaped JSON, same as app.js's untyped handling.
 type AnyEvent = any;
 export interface ClassifyCtx {
  toolNameById: Map<string, string>;
  pendingTurnStartTs: { current: number | null };
  keySeq: { current: number };
 }
 export function createClassifyCtx(): ClassifyCtx {
  return { toolNameById: new Map(), pendingTurnStartTs: { current: null }, keySeq: { current: 0 } };
 }
 function nextKey(ctx: ClassifyCtx): string {
  ctx.keySeq.current += 1;
  return 'r' + ctx.keySeq.current;
 }
 function trim(s: string, n: number): string {
  return s.length > n ? s.slice(0, n) + '…' : s;
 }
 export function classifyEvent(ev: AnyEvent, fromHistory: boolean, ctx: ClassifyCtx): StreamRow[] {
  switch (ev.kind) {
    case 'turn_start':
      return [classifyTurnStart(ev, fromHistory, ctx)];
    case 'turn_end':
      return [classifyTurnEnd(ev, fromHistory, ctx)];
    case 'note':
      return [classifyNote(ev, fromHistory, ctx)];
    case 'stream': {
      const v = { ...ev };
      delete v.kind;
      return classifyStream(v, fromHistory, ctx);
    }
    default:
      // status_changed / model_changed / effort_changed /
      // token_usage_changed / turn_state_changed drive badges elsewhere
      // (useAgentState's poll — see Root.tsx) rather than rows here.
      return [];
  }
 }
 function classifyTurnStart(ev: AnyEvent, fromHistory: boolean, ctx: ClassifyCtx): StreamRow {
  const meta: StreamRowMeta[] = [];
  if (typeof ev.ts === 'number') {
    ctx.pendingTurnStartTs.current = ev.ts;
    meta.push({ cls: 'turn-time', text: '· ' + fmtClock(ev.ts) });
  }
  if (ev.unread > 0) {
    meta.push({ cls: 'unread-badge', text: '· ' + ev.unread + ' unread' });
  }
  return {
    key: nextKey(ctx),
    cssClass: 'turn-start',
    fromHistory,
    text: '◆ TURN ← ' + ev.from,
    meta,
    childText: { cls: 'turn-body', text: String(ev.body ?? '') },
  };
 }
 function classifyTurnEnd(ev: AnyEvent, fromHistory: boolean, ctx: ClassifyCtx): StreamRow {
  const meta: StreamRowMeta[] = [];
  if (typeof ev.ts === 'number') {
    let label = '· ' + fmtClock(ev.ts);
    if (ctx.pendingTurnStartTs.current != null && ev.ts >= ctx.pendingTurnStartTs.current) {
      label += ' · ' + fmtAge((ev.ts - ctx.pendingTurnStartTs.current) * 1000);
    }
    meta.push({ cls: 'turn-time', text: label });
  }
  ctx.pendingTurnStartTs.current = null;
  return {
    key: nextKey(ctx),
    cssClass: ev.ok ? 'turn-end-ok' : 'turn-end-fail',
    fromHistory,
    text: (ev.ok ? '✅' : '❌') + ' turn ' + (ev.ok ? 'ok' : 'fail') + (ev.note ? ' — ' + ev.note : ''),
    meta,
  };
 }
 function classifyNote(ev: AnyEvent, fromHistory: boolean, ctx: ClassifyCtx): StreamRow {
  const t = String(ev.text ?? '');
  if (t.startsWith('stderr:')) return { key: nextKey(ctx), cssClass: 'note stderr', fromHistory, text: '! ' + t };
  if (t.startsWith('operator:')) return { key: nextKey(ctx), cssClass: 'note op', fromHistory, text: '· ' + t };
  return { key: nextKey(ctx), cssClass: 'note', fromHistory, text: '· ' + t };
 }
 function classifyStream(v: AnyEvent, fromHistory: boolean, ctx: ClassifyCtx): StreamRow[] {
  if (v._category === 'drop') return [];
  if (v.type === 'system') {
    const cat = v._category;
    const summary = v._summary;
    if (cat === 'thinking_tok') {
      return [{
        key: nextKey(ctx), cssClass: 'note', icon: '🧠', fromHistory,
        text: summary || 'thinking…', coalesceKey: 'thinking-tok',
      }];
    }
    if (v.subtype === 'plugin_install') {
      return [{
        key: nextKey(ctx), cssClass: 'note', fromHistory,
        text: summary || '⚙ plugin install', coalesceKey: 'plugin-install',
      }];
    }
    if (v.subtype === 'status') {
      return [{
        key: nextKey(ctx), cssClass: 'note', fromHistory,
        text: summary || '⚙ status', coalesceKey: 'status-tick',
      }];
    }
    if (cat === 'details') {
      return [{
        key: nextKey(ctx), cssClass: 'note', fromHistory, details: true,
        text: summary || '⚙ ' + (v.subtype || ''), plainBody: v._body || '',
      }];
    }
    return [{ key: nextKey(ctx), cssClass: 'note', fromHistory, text: summary || '⚙ ' + (v.subtype || 'system') }];
  }
  if (v.subtype === 'task_started' || v.subtype === 'task_notification') {
    const row = classifyTaskEvent(v, fromHistory, ctx);
    if (row) return [row];
  }
  if (v.type === 'assistant' && v.message && v.message.content) {
    const rows: StreamRow[] = [];
    for (const c of v.message.content) {
      if (c.type === 'text' && c.text && String(c.text).trim()) {
        rows.push({ key: nextKey(ctx), cssClass: 'text', fromHistory, markdownBody: c.text });
      } else if (c.type === 'thinking') {
        const txt = String(c.thinking || c.text || '').trim();
        rows.push({ key: nextKey(ctx), cssClass: 'thinking', icon: '💭', fromHistory, text: txt || 'thinking …' });
      } else if (c.type === 'tool_use') {
        if (c.id && c.name) ctx.toolNameById.set(c.id, c.name);
        rows.push(classifyToolUse(c, fromHistory, ctx));
      }
    }
    return rows;
  }
  if (v.type === 'user' && v.message && v.message.content) {
    const rows: StreamRow[] = [];
    for (const c of v.message.content) {
      if (c.type === 'tool_result') rows.push(classifyToolResult(c, fromHistory, ctx));
    }
    return rows;
  }
  return [{ key: nextKey(ctx), cssClass: 'sys', fromHistory, text: '! ' + trim(JSON.stringify(v), 200) }];
 }
 // `_category === 'rich'` tools get an expandable row: diff body (Edit),
 // default-open markdown body (send/recv-shaped, always open regardless
 // of the preference below — matches app.js), or a plain
 // body (collapsed unless the operator's "expand tool output" preference
 // says otherwise — @hive/shared/prefs.js's getExpandDetailsPref(), read
 // fresh per row so a mid-session preference change applies going
 // forward without a reload) — all pre-computed server-side, no per-tool
 // JS needed.
 function classifyToolUse(c: AnyEvent, fromHistory: boolean, ctx: ClassifyCtx): StreamRow {
  const icon = c._icon || '🔧';
  const name = c.name || '';
  if (c._category === 'rich' && c._body != null) {
    const summary = c._summary || name || '?';
    if (c._body_type === 'diff') {
      return {
        key: nextKey(ctx), cssClass: 'tool-use', fromHistory, details: true, defaultOpen: getExpandDetailsPref(),
        icon, text: summary, diffBody: c._body,
      };
    }
    if (c._body_type === 'markdown') {
      return {
        key: nextKey(ctx), cssClass: 'tool-use', fromHistory, details: true, defaultOpen: true,
        icon, text: summary, markdownBody: c._body,
      };
    }
    return {
      key: nextKey(ctx), cssClass: 'tool-use', fromHistory, details: true, defaultOpen: getExpandDetailsPref(),
      icon, text: summary, plainBody: c._body,
    };
  }
  return { key: nextKey(ctx), cssClass: 'tool-use', fromHistory, icon, text: c._summary || name || '?' };
 }
 function classifyToolResult(c: AnyEvent, fromHistory: boolean, ctx: ClassifyCtx): StreamRow {
  const rawTxt = Array.isArray(c.content) ? c.content.map((p: AnyEvent) => p.text || '').join('') : c.content || '';
  const isError = !!c.is_error;
  // Strip the <tool_use_error>…</tool_use_error> wrapper — implementation
  // detail claude emits on failed tool calls, adds nothing for the operator.
  const txt = isError
    ? String(rawTxt).replace(/^<tool_use_error>([\s\S]*)<\/tool_use_error>$/, '$1').trim()
    : String(rawTxt);
  const sourceName = c.tool_use_id ? ctx.toolNameById.get(c.tool_use_id) : null;
  const isMessageBearing = sourceName === 'mcp__hyperhive__recv';
  const trimmed = txt.replace(/\s+/g, ' ').trim();
  const summaryBody = (() => {
    if (!trimmed) return '(empty)';
    if (trimmed.length <= 120) return trimmed;
    const lines = txt.split('\n').filter((l: string) => l.length).length;
    const headline = trimmed.slice(0, 90) + '…';
    return `${lines}L · ${headline}`;
  })();
  if (isError) {
    if (!txt.trim() || txt.length <= 120) {
      return { key: nextKey(ctx), cssClass: 'tool-result error', fromHistory, text: '✗  ' + summaryBody };
    }
    return {
      key: nextKey(ctx), cssClass: 'tool-result-block error', fromHistory, details: true,
      defaultOpen: getExpandDetailsPref(), text: summaryBody, plainBody: txt,
    };
  }
  if (isMessageBearing && txt.trim()) {
    return {
      key: nextKey(ctx), cssClass: 'tool-result-block', fromHistory, details: true, defaultOpen: true,
      text: 'recv ← ' + summaryBody, markdownBody: txt,
    };
  }
  if (!txt.trim() || txt.length <= 120) {
    return { key: nextKey(ctx), cssClass: 'tool-result', fromHistory, text: '←  ' + summaryBody };
  }
  return {
    key: nextKey(ctx), cssClass: 'tool-result-block', fromHistory, details: true,
    defaultOpen: getExpandDetailsPref(), text: summaryBody, plainBody: txt,
  };
 }
 // Subagent (claude `Task`-tool) activity — dead path for agents today
 // (`Task` is omitted from the allow-list) but kept for parity, same as
 // app.js's `renderTaskEvent`. Glyph stays embedded in the row text
 // (not routed through the icon column) — matches the original exactly.
 function classifyTaskEvent(v: AnyEvent, fromHistory: boolean, ctx: ClassifyCtx): StreamRow | null {
  const id = String(v.task_id || '').slice(0, 8);
  const kind = v.task_type ? ` [${v.task_type}]` : '';
  const desc = v.description || v.summary || '(no description)';
  if (v.subtype === 'task_started') {
    return { key: nextKey(ctx), cssClass: 'tool-use', fromHistory, text: `⌁  task ${id} started · ${desc}${kind}` };
  }
  if (v.subtype === 'task_notification') {
    const status = v.status || 'unknown';
    const glyph = status === 'completed' ? '✓' : status === 'failed' ? '✗' : '◌';
    const cssClass = status === 'completed' ? 'turn-end-ok' : status === 'failed' ? 'turn-end-fail' : 'tool-result';
    const out = v.output_file ? ` · → ${v.output_file}` : '';
    return { key: nextKey(ctx), cssClass, fromHistory, text: `⌁  task ${id} ${glyph} ${status} · ${desc}${out}` };
  }
  return null;
 }
--- a/frontend/packages/agent/src/lib/format.ts
+++ b/frontend/packages/agent/src/lib/format.ts
@ -15,10 +15,3 @@ export function fmtAge(ms: number): string {
  const h = Math.floor(m / 60);
  return h + 'h ' + (m % 60) + 'm';
 }
 /** Wall-clock HH:MM:SS (UTC) from a unix-seconds value — labels
 * turn-start / turn-end rows in the live stream (ported from app.js's
 * `fmtClock`). */
 export function fmtClock(sec: number): string {
  return new Date(sec * 1000).toISOString().slice(11, 19);
 }
--- a/frontend/packages/agent/src/lib/streamRow.ts
+++ b/frontend/packages/agent/src/lib/streamRow.ts
@ -1,49 +0,0 @@
 // Row model for the live event stream. One `StreamRow` = one rendered
 // line/panel in the terminal pane; `classifyEvent` (classifyEvent.ts)
 // turns a raw harness/stream-json event into zero or more of these, and
 // `<Row>` (components/Row.tsx) renders one. Kept as plain data (not JSX)
 // so the append/coalesce bookkeeping in useLiveStream.ts stays pure —
 // see docs/terminal-rendering.md for the exact taxonomy this mirrors.
 export interface StreamRowMeta {
  cls: string;
  text: string;
 }
 export interface StreamRow {
  /** Stable across re-renders; reused in place when a row is coalesced
   * (e.g. the thinking-token counter) so Preact updates rather than
   * remounts it. */
  key: string;
  /** Space-joined class names appended after "row" / "row details", e.g.
   * "turn-start", "tool-result error" — taken verbatim from
   * @hive/shared/terminal.css's row-kind taxonomy. */
  cssClass: string;
  icon?: string;
  fromHistory: boolean;
  /** When set, an event that maps to the same coalesceKey and lands
   * while this row is still the last one in the list replaces it in
   * place instead of appending a new row (thinking-token ticks, status
   * ticks, plugin_install started→completed). */
  coalesceKey?: string;
  /** `false`/absent → flat `<div class="row">`; `true` → `<details
   * class="row">` with `.summary-text` + optional body. */
  details?: boolean;
  defaultOpen?: boolean;
  /** Flat-row prefix text, or the details `<summary>` text. Linkified,
   * never markdown. */
  text?: string;
  /** Small trailing spans after the prefix text (turn-time, unread
   * count) — flat rows only. */
  meta?: StreamRowMeta[];
  /** Plain-text child block under a flat row (just the turn-start wake
   * body today) — its own class, no markdown parsing. */
  childText?: StreamRowMeta;
  /** Sanitized-markdown body: appended under a flat row (assistant
   * text) or inside an open details row (send/recv bodies). */
  markdownBody?: string;
  /** Plain `<pre>` body inside a details row (generic long tool output). */
  plainBody?: string;
  /** `+`/`-`/context diff body inside a details row (Edit tool). */
  diffBody?: string;
 }
--- a/frontend/packages/agent/src/lib/termMsg.ts
+++ b/frontend/packages/agent/src/lib/termMsg.ts
@ -0,0 +1,33 @@
 // Wire types for the agent's terminal stream — mirrors hive-agent's
 // `term_msg.rs`/`web_ui/stream.rs` field-for-field. The frontend renders
 // a `TermMsg` close to as-is (see components/Row.tsx); there's no
 // separate client-side row model or classification step any more —
 // mara: "StreamRow should now match what the server sends in TermMsg."
 export type Level = 'debug' | 'info' | 'warn' | 'error';
 export interface TermMsg {
  icon?: string;
  level: Level;
  summary: string;
  body?: string;
  body_format?: 'markdown' | 'diff';
  coalesce_key?: string;
 }
 /** One SSE frame / history array entry — `hive-agent`'s `TermEnvelope`.
 * `seq` is the live per-event dedup counter (`BusEvent::seq`); absent on
 * history-replayed envelopes, see useLiveStream.ts's backfill dance. */
 export interface TermEnvelope {
  ts: number;
  seq?: number;
  msgs: TermMsg[];
 }
 /** A `TermMsg` plus the bookkeeping Preact needs to render a list —
 * stable identity for coalescing/keys, and whether it came from history
 * replay vs. the live tail. Not a separate model: everything content-wise
 * is still exactly the wire shape. */
 export interface TermRow extends TermMsg {
  key: string;
  fromHistory: boolean;
 }
--- a/frontend/packages/shared/src/terminal/terminal.css
+++ b/frontend/packages/shared/src/terminal/terminal.css
@ -97,55 +97,34 @@
  display: inline-block;
  width: 1.4em;
 }
-/* Row-kind colours. Pages register renderers that emit these classes;
+/* Row colours, keyed by severity `level` (hive-agent's `term_msg.rs`),
-   any class no page emits is just dead CSS, which is fine. Turn-framing
+   not by row kind any more — mara's terminal-message redesign dropped
-   classes carry their signal entirely on the coloured border-left rule —
+   the server-side `kind` tag (turn-start/tool-use/tool-result/etc) in
-   no bold, no top/bottom margins, no background tint. The chrome was
+   favour of one uniform shape, `{icon, level, summary, body,
-   overweight for what's just a "this is a boundary" marker. */
+   body_format, coalesce_key}`. What used to be a dozen-odd per-kind
-.live .turn-start    { color: var(--amber); border-left-color: var(--amber); }
+   classes (`.turn-start`, `.tool-use`, `.tool-result.error`, `.sys`, …)
-/* turn-body is a child block under turn-start carrying the wake-prompt
+   is now four: the four severities every row already carries. Structural
-   body; reset text-indent so wrapped content stays under its own column
+   identity (this is a turn boundary, this is a tool call) is carried by
-   instead of pulling back into the parent's prefix. */
+   the row's icon (`◆`, `🔧`, `✅`, …) and summary text instead of colour —
-.live .turn-body     { color: var(--fg); text-indent: 0; margin-top: 0.15em; }
+   see docs/terminal-rendering.md. */
-/* Any child block (markdown body, nested details) resets the parent
+.live .level-debug { color: var(--muted); }
-   row's hanging indent so the content lays out from column 0 of the
+.live .level-info  { color: var(--fg); }
-   body area. */
+.live .level-warn  { color: var(--amber); border-left-color: var(--amber); }
-.live .row .md, .live .row > details { text-indent: 0; }
+.live .level-error { color: var(--red);   border-left-color: var(--red);  }
-.live .turn-end-ok   { color: var(--green); border-left-color: var(--green); }
+/* `badge-pulse` itself is no longer used by any terminal row (the
-.live .turn-end-fail { color: var(--red);   border-left-color: var(--red); }
+   turn-start `unread` count it animated is gone — see term_msg.rs's
-/* Wall-clock time (+ duration on turn-end) appended to the turn-start /
+   module doc), but agent.css's `.state-badge.state-thinking`/
-   turn-end rows. Dim + smaller so the boundary glyph stays the focus and
+   `.state-compacting` badges still reuse this keyframe via
-   the timestamp reads as metadata. */
+   `@import "@hive/shared/terminal.css"` — keep the definition here, drop
-.live .turn-time   { color: var(--muted); font-size: 0.85em; margin-left: 0.5em; }
+   only the terminal-specific `.unread-badge` selector that used it. */
 .live .text        { color: var(--fg);    }
 .live .thinking    { color: var(--muted); font-style: italic; }
 .live .tool-use    { color: var(--cyan);  }
 .live .tool-result              { color: var(--muted); }
 .live .tool-result.error        { color: var(--red);   }
 .live .tool-result-block.error  { color: var(--red);   }
 .live .result      { color: var(--green); }
 .live .note        { color: var(--muted); }
 /* Distinguish stderr lines (orange) and operator-initiated notes
   (mauve, lightly emphasised) from ambient harness chatter so the
   eye picks out anomalies + operator actions in the scrollback. */
 .live .note.stderr { color: var(--amber); }
 .live .note.op     { color: var(--purple); font-style: italic; }
 /* The .sys catch-all fires when renderStream landed an event shape it
   couldn't classify. Make it visually loud so silently-dropped event
   types surface for follow-up. */
 .live .sys         { color: var(--amber); }
 .live .unread-badge {
  color: var(--amber);
  font-weight: normal;
  margin-left: 0.6em;
  font-size: 0.85em;
  text-shadow: 0 0 6px color-mix(in srgb, var(--amber) 55%, transparent);
  animation: badge-pulse 1.4s ease-in-out infinite;
 }
@keyframes badge-pulse {
  0%, 100% { opacity: 1;   text-shadow: 0 0 6px color-mix(in srgb, var(--amber) 55%, transparent); }
  50%      { opacity: 0.7; text-shadow: 0 0 14px color-mix(in srgb, var(--amber) 95%, transparent); }
 }
 /* Any child block (markdown body, nested details) resets the parent
   row's hanging indent so the content lays out from column 0 of the
   body area. */
 .live .row .md, .live .row > details { text-indent: 0; }
 /* "↓ N new" pill: shown when new rows arrive while the operator is
   scrolled up; click to jump to bottom. Positioned by the wrapper's
   `position: relative` (terminal-wrap supplies it; pages that skip the
--- a/hive-agent/src/main.rs
+++ b/hive-agent/src/main.rs
@ -28,6 +28,7 @@ mod serve_common;
 mod state_entry_watch;
 mod stats;
 mod stream_enrich;
 mod term_msg;
 mod todo_server;
 mod todos;
 mod turn;
--- a/hive-agent/src/stream_enrich.rs
+++ b/hive-agent/src/stream_enrich.rs
@ -1,75 +1,97 @@
-//! Enrich raw claude stream-json values before SSE delivery.
+//! Classify raw claude stream-json values into [`crate::term_msg::TermMsg`]
 //! rows before SSE delivery.
 //!
-//! A single [`enrich`] function stamps `_icon`, `_summary`, `_category`,
+//! [`classify_stream_value`] is the entry point — it walks one raw claude
-//! and optionally `_body` onto [`crate::events::LiveEvent::Stream`] payloads
+//! `stream-json` line (the payload of a [`crate::events::LiveEvent::Stream`])
-//! so the frontend can read pre-computed fields instead of duplicating the
+//! and returns zero or more terminal rows. Applied at SSE-emit time in
-//! dispatch logic in JavaScript.
+//! `crate::web_ui::stream` so both the live tail (`events/stream`) and the
 //! history replay (`events/history`) endpoints deliver the same classified
 //! shape — the sqlite event log stores the raw, unclassified event, so the
 //! DB never needs migration when classification logic changes.
 //!
-//! The sqlite event log stores raw (un-enriched) events — the DB never needs
+//! The per-tool icon/summary formatting below (`tool_icon`, `fmt_tool_use`
-//! migration when the enrichment logic changes. Enrichment is applied at
+//! and its per-family helpers, `rich_tool_body`) is reused as-is from
-//! SSE-emit time in `crate::web_ui::stream` so both the live tail
+//! before mara's terminal-message redesign — that logic (what does
-//! (`events/stream`) and the history replay (`events/history`) endpoints
+//! `mcp__hyperhive__send`'s row say, which tools get an expandable body)
-//! deliver the same enriched shape.
+//! didn't change; only the shape it gets packed into did.
 //!
 //! # Migration (two-phase)
 //!
 //! **Phase 1** (this change): backend stamps `_icon`/`_summary`/`_category`
 //! fields; the client reads them when present and falls back to its own JS
 //! tables when absent. Zero user-visible change — a no-op for clients that
 //! haven't yet been updated.
 //!
 //! **Phase 2** (follow-up): the client drops the JS tables once phase 1 is
 //! deployed everywhere.
 use crate::term_msg::{BodyFormat, ClassifyCtx, Level, TermMsg};
 use serde_json::{Value, json};
-/// Stamp enrichment fields onto a raw claude stream-json [`Value`].
+/// Classify one raw claude stream-json line into zero or more terminal rows.
 ///
-/// - `type="system"` events get `_category` + `_summary` (and `_body` for
+/// Dispatch order mirrors the old client-side `classifyEvent.ts` (before
-///   expandable detail, e.g. `commands_changed`).
+/// this classification moved server-side):
-/// - `type="assistant"` events get `_icon`, `_summary`, and optionally
+/// top-level drop-noise types first, then `type="system"`, then task events
-///   `_category: "rich"` stamped onto each `message.content[]` entry that
+/// (matched on `subtype` regardless of `type`), then `assistant`/`user`
-///   has `type="tool_use"`.
+/// content, with an unrecognised shape falling through to a loud
-///
+/// warn-level catch-all so a silently-dropped event type stays visible.
-/// No-ops for unknown/unhandled top-level types. Existing `_`-prefixed fields
+pub fn classify_stream_value(v: &Value, ctx: &mut ClassifyCtx) -> Vec<TermMsg> {
-/// are left unchanged so the call is idempotent (history replay may hit
+    let vtype = v.get("type").and_then(Value::as_str).unwrap_or("");
-/// already-enriched values if the DB is ever pre-populated by a future phase).
+    if matches!(vtype, "result" | "rate_limit_event") {
-pub fn enrich(v: &mut Value) {
+        return vec![];
    match v.get("type").and_then(Value::as_str).unwrap_or("") {
        // Top-level result/rate_limit_event are drop-category noise — stamp
        // the same category the frontend uses to silently discard them.
        "result" | "rate_limit_event" => {
            if let Some(obj) = v.as_object_mut() {
                obj.entry("_category").or_insert_with(|| json!("drop"));
    }
    if vtype == "system" {
        return classify_system(v);
    }
-        "system" => enrich_system(v),
+    let subtype = v.get("subtype").and_then(Value::as_str);
-        "assistant" => enrich_assistant(v),
+    if matches!(subtype, Some("task_started" | "task_notification")) {
-        _ => {}
+        return classify_task_event(v).into_iter().collect();
    }
    if vtype == "assistant" {
        return v
            .get("message")
            .and_then(|m| m.get("content"))
            .and_then(Value::as_array)
            .map_or_else(Vec::new, |content| classify_assistant_content(content, ctx));
    }
    if vtype == "user" {
        return v
            .get("message")
            .and_then(|m| m.get("content"))
            .and_then(Value::as_array)
            .map_or_else(Vec::new, |content| classify_user_content(content, ctx));
    }
    vec![TermMsg::new(Level::Warn, trim_str(&v.to_string(), 200)).icon("!")]
 }
 // ---------------------------------------------------------------------------
 // system events
 // ---------------------------------------------------------------------------
-fn enrich_system(v: &mut Value) {
+fn classify_system(v: &Value) -> Vec<TermMsg> {
-    if v.get("_category").is_some() {
+    let subtype = v.get("subtype").and_then(Value::as_str).unwrap_or("");
-        return; // idempotent
+    let (category, summary, body) = system_fields(v, subtype);
-    }
+    match category {
-    let subtype = v
+        "drop" => vec![],
-        .get("subtype")
+        "thinking_tok" => vec![
-        .and_then(Value::as_str)
+            TermMsg::new(
-        .unwrap_or("")
+                Level::Debug,
-        .to_owned();
+                summary.unwrap_or_else(|| "thinking…".to_owned()),
-    let (category, summary, body) = system_fields(v, &subtype);
+            )
-    let Some(obj) = v.as_object_mut() else { return };
+            .icon("🧠")
-    obj.insert("_category".to_owned(), json!(category));
+            .coalesce("thinking-tok"),
-    if let Some(s) = summary {
+        ],
-        obj.insert("_summary".to_owned(), json!(s));
+        "details" => {
-    }
+            let mut m = TermMsg::new(Level::Debug, summary.unwrap_or_default());
            if let Some(b) = body {
-        obj.insert("_body".to_owned(), json!(b));
+                m = m.body(b, None);
            }
            vec![m]
        }
        // "note" category — ambient harness/system chatter. Level + coalesce
        // key vary by subtype; everything else defaults to a plain debug note.
        _ => {
            let s = summary.unwrap_or_default();
            let m = match subtype {
                "api_error" => TermMsg::new(Level::Error, s),
                "api_retry" => TermMsg::new(Level::Warn, s),
                "plugin_install" => TermMsg::new(Level::Debug, s).coalesce("plugin-install"),
                "status" => TermMsg::new(Level::Debug, s).coalesce("status-tick"),
                _ => TermMsg::new(Level::Debug, s),
            };
            vec![m]
        }
    }
 }
@ -197,51 +219,197 @@ fn system_fields(v: &Value, subtype: &str) -> (&'static str, Option<String>, Opt
 }
 // ---------------------------------------------------------------------------
-// assistant events
+// assistant events (claude's own output: text / thinking / tool calls)
 // ---------------------------------------------------------------------------
-fn enrich_assistant(v: &mut Value) {
+fn classify_assistant_content(content: &[Value], ctx: &mut ClassifyCtx) -> Vec<TermMsg> {
-    // Navigate message.content[] — absent on text-only turns.
+    let mut rows = Vec::new();
-    let Some(content) = v
+    for c in content {
-        .get_mut("message")
+        match c.get("type").and_then(Value::as_str) {
-        .and_then(|m| m.get_mut("content"))
+            Some("text") => {
-        .and_then(Value::as_array_mut)
+                let text = c.get("text").and_then(Value::as_str).unwrap_or("");
-    else {
+                if !text.trim().is_empty() {
-        return;
+                    // No separate summary line — the markdown body is the
                    // whole row, matching the old `.text` row shape.
                    rows.push(
                        TermMsg::new(Level::Info, String::new())
                            .body(text.to_owned(), Some(BodyFormat::Markdown)),
                    );
                }
            }
            Some("thinking") => {
                let txt = c
                    .get("thinking")
                    .or_else(|| c.get("text"))
                    .and_then(Value::as_str)
                    .unwrap_or("")
                    .trim()
                    .to_owned();
                let summary = if txt.is_empty() {
                    "thinking …".to_owned()
                } else {
                    txt
                };
-    for entry in content.iter_mut() {
+                rows.push(TermMsg::new(Level::Debug, summary).icon("💭"));
-        enrich_tool_use_entry(entry);
+            }
            Some("tool_use") => {
                if let (Some(id), Some(name)) = (
                    c.get("id").and_then(Value::as_str),
                    c.get("name").and_then(Value::as_str),
                ) {
                    ctx.record_tool_use(id, name);
                }
                rows.push(classify_tool_use(c));
            }
            _ => {}
        }
    }
    rows
 }
 /// `_category === 'rich'` tools used to get an expandable row (diff body
 /// for Edit, markdown body for send, plain for everything else with a
 /// body). That's now just "does this row have a body" — `body.is_some()`
 /// on the returned [`TermMsg`] *is* the expandable signal, no separate
 /// flag.
 fn classify_tool_use(c: &Value) -> TermMsg {
    let name = c.get("name").and_then(Value::as_str).unwrap_or("");
    let input = c.get("input").cloned().unwrap_or_else(|| json!({}));
    let mut m = TermMsg::new(Level::Info, fmt_tool_use(name, &input)).icon(tool_icon(name));
    if let Some((body, body_type)) = rich_tool_body(name, &input) {
        let format = match body_type {
            "diff" => Some(BodyFormat::Diff),
            "markdown" => Some(BodyFormat::Markdown),
            _ => None, // "plain"
        };
        m = m.body(body, format);
    }
    m
 }
 // ---------------------------------------------------------------------------
 // user events (tool_result — claude's own tool calls answered)
 // ---------------------------------------------------------------------------
 fn classify_user_content(content: &[Value], ctx: &ClassifyCtx) -> Vec<TermMsg> {
    content
        .iter()
        .filter(|c| c.get("type").and_then(Value::as_str) == Some("tool_result"))
        .map(|c| classify_tool_result(c, ctx))
        .collect()
 }
 /// `<tool_use_error>…</tool_use_error>` is claude's own wrapper on failed
 /// tool calls — implementation detail, adds nothing for the operator.
 fn strip_tool_use_error_wrapper(s: &str) -> String {
    let trimmed = s.trim();
    trimmed
        .strip_prefix("<tool_use_error>")
        .and_then(|rest| rest.strip_suffix("</tool_use_error>"))
        .map_or_else(|| trimmed.to_owned(), |inner| inner.trim().to_owned())
 }
 fn classify_tool_result(c: &Value, ctx: &ClassifyCtx) -> TermMsg {
    let raw_txt = match c.get("content") {
        Some(Value::Array(parts)) => parts
            .iter()
            .filter_map(|p| p.get("text").and_then(Value::as_str))
            .collect::<String>(),
        Some(Value::String(s)) => s.clone(),
        _ => String::new(),
    };
    let is_error = c.get("is_error").and_then(Value::as_bool).unwrap_or(false);
    let txt = if is_error {
        strip_tool_use_error_wrapper(&raw_txt)
    } else {
        raw_txt
    };
    let tool_use_id = c.get("tool_use_id").and_then(Value::as_str);
    let source_name = tool_use_id.and_then(|id| ctx.tool_name(id));
    let is_message_bearing = source_name == Some("mcp__hyperhive__recv");
    let trimmed: String = txt.split_whitespace().collect::<Vec<_>>().join(" ");
    let summary = summarize_tool_result(&txt, &trimmed);
    let short = txt.trim().is_empty() || txt.chars().count() <= 120;
    if is_error {
        let m = TermMsg::new(Level::Error, summary);
        return if short {
            m.icon("✗")
        } else {
            m.body(txt, None)
        };
    }
    if is_message_bearing && !txt.trim().is_empty() {
        return TermMsg::new(Level::Info, format!("recv ← {summary}"))
            .body(txt, Some(BodyFormat::Markdown));
    }
    let m = TermMsg::new(Level::Info, summary);
    if short {
        m.icon("←")
    } else {
        m.body(txt, None)
    }
 }
-fn enrich_tool_use_entry(entry: &mut Value) {
+/// `(empty)` / a short trimmed line / `"NL · headline…"` for a long one —
-    if entry.get("type").and_then(Value::as_str) != Some("tool_use") {
+/// matches the old client-side `summaryBody` computation exactly.
-        return;
+fn summarize_tool_result(txt: &str, trimmed: &str) -> String {
    if trimmed.is_empty() {
        return "(empty)".to_owned();
    }
-    if entry.get("_icon").is_some() {
+    if trimmed.chars().count() <= 120 {
-        return; // idempotent
+        return trimmed.to_owned();
    }
-    let name = entry
+    let lines = txt.lines().filter(|l| !l.is_empty()).count();
-        .get("name")
+    let headline: String = trimmed.chars().take(90).collect();
    format!("{lines}L · {headline}…")
 }
 // ---------------------------------------------------------------------------
 // subagent (claude Task-tool) activity — dead path for agents today (`Task`
 // is omitted from the allow-list) but kept for parity with the old
 // client-side `classifyTaskEvent`.
 // ---------------------------------------------------------------------------
 fn classify_task_event(v: &Value) -> Option<TermMsg> {
    let id: String = v
        .get("task_id")
        .and_then(Value::as_str)
        .unwrap_or("")
-        .to_owned();
+        .chars()
-    let input = entry.get("input").cloned().unwrap_or_else(|| json!({}));
+        .take(8)
-    let icon = tool_icon(&name);
+        .collect();
-    let summary = fmt_tool_use(&name, &input);
+    let kind = v
-    let rich = is_rich_tool(&name);
+        .get("task_type")
-    let body = rich_tool_body(&name, &input);
+        .and_then(Value::as_str)
-    let Some(obj) = entry.as_object_mut() else {
+        .map(|t| format!(" [{t}]"))
-        return;
+        .unwrap_or_default();
-    };
+    let desc = v
-    obj.insert("_icon".to_owned(), json!(icon));
+        .get("description")
-    obj.insert("_summary".to_owned(), json!(summary));
+        .or_else(|| v.get("summary"))
-    if rich {
+        .and_then(Value::as_str)
-        obj.insert("_category".to_owned(), json!("rich"));
+        .unwrap_or("(no description)");
    match v.get("subtype").and_then(Value::as_str) {
        Some("task_started") => {
            Some(TermMsg::new(Level::Info, format!("task {id} started · {desc}{kind}")).icon("⌁"))
        }
-    if let Some((b, bt)) = body {
+        Some("task_notification") => {
-        obj.insert("_body".to_owned(), json!(b));
+            let status = v.get("status").and_then(Value::as_str).unwrap_or("unknown");
-        obj.insert("_body_type".to_owned(), json!(bt));
+            let (glyph, level) = match status {
                "completed" => ("✓", Level::Info),
                "failed" => ("✗", Level::Error),
                _ => ("◌", Level::Info),
            };
            let out = v
                .get("output_file")
                .and_then(Value::as_str)
                .map(|f| format!(" · → {f}"))
                .unwrap_or_default();
            Some(TermMsg::new(level, format!("task {id} {glyph} {status} · {desc}{out}")).icon("⌁"))
        }
        _ => None,
    }
 }
@ -249,22 +417,14 @@ fn enrich_tool_use_entry(entry: &mut Value) {
 // tool helpers
 // ---------------------------------------------------------------------------
 /// Whether this tool is rendered by the frontend's *rich* renderer (diff
 /// view, body expansion) rather than the flat `_summary` row. Flagged with
 /// `_category: "rich"` so the client can distinguish without re-implementing
 /// the tool name list.
 fn is_rich_tool(name: &str) -> bool {
    matches!(name, "Edit" | "mcp__bash__run" | "mcp__hyperhive__send")
 }
 /// Pre-compute the expandable body for rich tool entries.
 ///
-/// Returns `Some((body, body_type))` where `body_type` tells the frontend
+/// Returns `Some((body, body_type))` where `body_type` becomes the
-/// which renderer to use:
+/// `TermMsg`'s `body_format` and tells the frontend which renderer to use:
-/// - `"diff"` → `api.detailsDiff` (colour-coded `+`/`-` lines)
+/// - `"diff"` → colour-coded `+`/`-` lines
-/// - `"plain"` → `api.details` (plain `<pre>` block)
+/// - `"plain"` (`None` on the wire) → a plain `<pre>` block
-/// - `"markdown"` → `api.detailsOpenMd` (markdown rendered via marked + `DOMPurify`,
+/// - `"markdown"` → rendered via `marked` + `DOMPurify`; used for
-///   default-open; used for message-bearing tools: `send`)
+///   message-bearing tools: `send`
 ///
 /// Returns `None` for tools that have no body at all.
 ///
--- a/hive-agent/src/term_msg.rs
+++ b/hive-agent/src/term_msg.rs
@ -0,0 +1,283 @@
 //! Terminal-message wire shape: what the per-agent web UI's live/history
 //! endpoints actually serve for the "terminal" event stream, as opposed to
 //! agent-state changes (`StatusChanged`/`ModelChanged`/`EffortChanged`/
 //! `TokenUsageChanged`/`TurnStateChanged`), which have never rendered as
 //! terminal rows (the header/badges poll `/api/state`, not this stream) and
 //! produce zero [`TermMsg`]s here.
 //!
 //! One [`crate::events::LiveEvent`] maps to zero or more `TermMsg`s — most
 //! map to exactly one, but `LiveEvent::Stream` (one raw claude
 //! `stream-json` line) can expand to several: an `assistant` message with
 //! both a text block and a `tool_use` block produces two rows.
 //!
 //! Design history: mara's terminal-message redesign (six rounds of
 //! negotiation on the forge) collapsed what was an 11-field frontend-side
 //! row shape (`StreamRow`, `frontend/packages/agent/src/lib/streamRow.ts`)
 //! plus raw claude-JSON passthrough into this 6-field shape, with
 //! classification moved server-side so the client-side `classifyEvent.ts` —
 //! a large per-tool dispatch table — mostly goes away. `kind`, `unread`,
 //! `from`, and `expanded_default` were all considered and dropped along the
 //! way; `level` replaces free-text CSS-class styling.
 use serde::Serialize;
 use std::collections::HashMap;
 use crate::events::LiveEvent;
 #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize)]
 #[serde(rename_all = "snake_case")]
 pub enum Level {
    /// Low-signal / ambient chatter — thinking, progress ticks, harness
    /// housekeeping. The client's default rendering can dim/de-emphasize
    /// these without hiding them outright.
    Debug,
    /// Routine substantive content — turn boundaries, assistant text, tool
    /// calls/results, message bodies.
    Info,
    /// Heads-up, not necessarily broken — stderr lines, an unclassified
    /// event shape landing (the old `.sys` catch-all), API retries.
    Warn,
    /// Something actually failed — a turn ending non-ok, a tool result with
    /// `is_error: true`.
    Error,
 }
 #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize)]
 #[serde(rename_all = "snake_case")]
 pub enum BodyFormat {
    Markdown,
    Diff,
 }
 /// One terminal row. `body_format: None` with `body: Some(_)` means plain
 /// text (the common case — no explicit tag on the wire for it, same logic
 /// as `body` itself being absent meaning "nothing to expand").
 #[derive(Debug, Clone, Serialize)]
 pub struct TermMsg {
    #[serde(skip_serializing_if = "Option::is_none")]
    pub icon: Option<String>,
    pub level: Level,
    pub summary: String,
    #[serde(skip_serializing_if = "Option::is_none")]
    pub body: Option<String>,
    #[serde(skip_serializing_if = "Option::is_none")]
    pub body_format: Option<BodyFormat>,
    #[serde(skip_serializing_if = "Option::is_none")]
    pub coalesce_key: Option<String>,
 }
 impl TermMsg {
    pub fn new(level: Level, summary: impl Into<String>) -> Self {
        Self {
            icon: None,
            level,
            summary: summary.into(),
            body: None,
            body_format: None,
            coalesce_key: None,
        }
    }
    #[must_use]
    pub fn icon(mut self, icon: impl Into<String>) -> Self {
        self.icon = Some(icon.into());
        self
    }
    #[must_use]
    pub fn body(mut self, body: impl Into<String>, format: Option<BodyFormat>) -> Self {
        self.body = Some(body.into());
        self.body_format = format;
        self
    }
    #[must_use]
    pub fn coalesce(mut self, key: impl Into<String>) -> Self {
        self.coalesce_key = Some(key.into());
        self
    }
 }
 /// Per-connection/per-request classification state. A live SSE stream keeps
 /// one of these alive for the connection's lifetime — `tool_use` id → name
 /// correlation, so a `tool_result` can tell it's answering a `recv` call and
 /// render its body as markdown instead of plain text. The history endpoint
 /// uses a fresh one per page: correlation only works within the page
 /// actually returned, not across the live/history boundary. Accepted
 /// degradation — the only user-visible effect is a `recv` result whose
 /// `tool_use` fell on the other side of a page/reconnect boundary rendering
 /// its body as plain text instead of markdown.
 #[derive(Default)]
 pub struct ClassifyCtx {
    tool_name_by_id: HashMap<String, String>,
 }
 impl ClassifyCtx {
    pub fn record_tool_use(&mut self, id: &str, name: &str) {
        self.tool_name_by_id.insert(id.to_owned(), name.to_owned());
    }
    #[must_use]
    pub fn tool_name(&self, id: &str) -> Option<&str> {
        self.tool_name_by_id.get(id).map(String::as_str)
    }
 }
 /// Classify one [`LiveEvent`] into zero or more terminal rows.
 pub fn classify(ev: &LiveEvent, ctx: &mut ClassifyCtx) -> Vec<TermMsg> {
    match ev {
        LiveEvent::TurnStart { from, body, .. } => {
            // `unread` (the third field) was dropped — mara: "i can see
            // the todo/inbox count in toolbar and dont need the N unread".
            let mut m = TermMsg::new(Level::Info, format!("TURN ← {from}")).icon("◆");
            if !body.trim().is_empty() {
                m = m.body(body.clone(), None);
            }
            vec![m]
        }
        LiveEvent::TurnEnd { ok, note } => {
            let msg = if *ok {
                TermMsg::new(Level::Info, "turn ok").icon("✅")
            } else {
                let summary = note
                    .as_deref()
                    .filter(|n| !n.is_empty())
                    .map_or_else(|| "turn fail".to_owned(), |n| format!("turn fail — {n}"));
                TermMsg::new(Level::Error, summary).icon("❌")
            };
            vec![msg]
        }
        LiveEvent::Note { text } => vec![classify_note(text)],
        LiveEvent::Stream(v) => crate::stream_enrich::classify_stream_value(v, ctx),
        // Agent-state transitions never render as terminal rows — the
        // header/badges read `/api/state`, not this stream (see module doc).
        LiveEvent::StatusChanged { .. }
        | LiveEvent::ModelChanged { .. }
        | LiveEvent::EffortChanged { .. }
        | LiveEvent::TokenUsageChanged { .. }
        | LiveEvent::TurnStateChanged { .. } => vec![],
    }
 }
 fn classify_note(text: &str) -> TermMsg {
    if let Some(rest) = text.strip_prefix("stderr:") {
        TermMsg::new(Level::Warn, format!("stderr:{rest}"))
    } else if let Some(rest) = text.strip_prefix("operator:") {
        TermMsg::new(Level::Info, format!("operator:{rest}"))
    } else {
        // Ambient harness chatter (session archived, plugin loaded, etc.) —
        // routine, not worth the same visual weight as a tool call or
        // assistant text.
        TermMsg::new(Level::Debug, text.to_owned())
    }
 }
 #[cfg(test)]
 mod tests {
    use super::{ClassifyCtx, Level, classify};
    use crate::events::LiveEvent;
    #[test]
    fn turn_start_carries_from_and_body_no_unread() {
        let ev = LiveEvent::TurnStart {
            from: "operator".into(),
            body: "sweep the backlog".into(),
            unread: 3,
        };
        let mut ctx = ClassifyCtx::default();
        let msgs = classify(&ev, &mut ctx);
        assert_eq!(msgs.len(), 1);
        assert_eq!(msgs[0].summary, "TURN ← operator");
        assert_eq!(msgs[0].body.as_deref(), Some("sweep the backlog"));
        assert_eq!(msgs[0].level, Level::Info);
    }
    #[test]
    fn turn_start_empty_body_has_no_body() {
        let ev = LiveEvent::TurnStart {
            from: "reminder".into(),
            body: String::new(),
            unread: 0,
        };
        let mut ctx = ClassifyCtx::default();
        let msgs = classify(&ev, &mut ctx);
        assert!(msgs[0].body.is_none());
    }
    #[test]
    fn turn_end_ok_is_info() {
        let ev = LiveEvent::TurnEnd {
            ok: true,
            note: None,
        };
        let mut ctx = ClassifyCtx::default();
        let msgs = classify(&ev, &mut ctx);
        assert_eq!(msgs[0].level, Level::Info);
        assert_eq!(msgs[0].summary, "turn ok");
    }
    #[test]
    fn turn_end_fail_is_error_with_note() {
        let ev = LiveEvent::TurnEnd {
            ok: false,
            note: Some("rate limited".into()),
        };
        let mut ctx = ClassifyCtx::default();
        let msgs = classify(&ev, &mut ctx);
        assert_eq!(msgs[0].level, Level::Error);
        assert_eq!(msgs[0].summary, "turn fail — rate limited");
    }
    #[test]
    fn note_stderr_is_warn() {
        let ev = LiveEvent::Note {
            text: "stderr: warning: deprecated flag".into(),
        };
        let mut ctx = ClassifyCtx::default();
        let msgs = classify(&ev, &mut ctx);
        assert_eq!(msgs[0].level, Level::Warn);
    }
    #[test]
    fn note_operator_is_info() {
        let ev = LiveEvent::Note {
            text: "operator: /compact requested".into(),
        };
        let mut ctx = ClassifyCtx::default();
        let msgs = classify(&ev, &mut ctx);
        assert_eq!(msgs[0].level, Level::Info);
    }
    #[test]
    fn note_plain_is_debug() {
        let ev = LiveEvent::Note {
            text: "created fresh session".into(),
        };
        let mut ctx = ClassifyCtx::default();
        let msgs = classify(&ev, &mut ctx);
        assert_eq!(msgs[0].level, Level::Debug);
    }
    #[test]
    fn agent_state_events_produce_no_rows() {
        let mut ctx = ClassifyCtx::default();
        assert!(
            classify(
                &LiveEvent::StatusChanged {
                    status: "online".into()
                },
                &mut ctx
            )
            .is_empty()
        );
        assert!(
            classify(
                &LiveEvent::ModelChanged {
                    model: "opus".into()
                },
                &mut ctx
            )
            .is_empty()
        );
    }
 }
--- a/hive-agent/src/web_ui/stream.rs
+++ b/hive-agent/src/web_ui/stream.rs
@ -9,13 +9,36 @@ use serde::{Deserialize, Serialize};
 use tokio_stream::{Stream, StreamExt, wrappers::BroadcastStream};
 use super::AppState;
 use crate::term_msg::{ClassifyCtx, Level, TermMsg, classify};
 /// One classified envelope on the wire: transport-level metadata (a sibling
 /// of the terminal-row payload, not part of it) plus the zero-or-more rows
 /// the raw event classified into. An event that classifies to zero rows (an
 /// agent-state change — `StatusChanged`/`ModelChanged`/etc. — or
 /// drop-category noise) never reaches the wire at all; see
 /// `crate::term_msg` for why.
 ///
 /// `seq` is the live per-event dedup counter (`BusEvent::seq`) — `Some` on
 /// the SSE path, `None` on history replay (a stored row has no live seq).
 /// Same category of plumbing as `ts`: the client already used it to drop
 /// buffered live traffic it's about to see again in the initial history
 /// page, and that need didn't go away just because rows lost their `kind`
 /// tag — dropping it here would silently reintroduce duplicate rows across
 /// the live/history boundary.
 #[derive(Serialize)]
 pub(super) struct TermEnvelope {
    ts: i64,
    #[serde(skip_serializing_if = "Option::is_none")]
    seq: Option<u64>,
    msgs: Vec<TermMsg>,
 }
 /// Response body for `GET /api/events/history`. `seq` is omitted from the
 /// wire entirely on a paginated (non-initial) load — matches the old
 /// `json!` shape, which only ever set the `"seq"` key when `Some`.
 #[derive(Serialize)]
 pub(super) struct EventsHistoryBody {
-    events: Vec<crate::events::StoredEvent>,
+    events: Vec<TermEnvelope>,
    min_id: Option<i64>,
    has_more: bool,
    #[serde(skip_serializing_if = "Option::is_none")]
@ -52,15 +75,28 @@ pub(super) async fn events_history(
    };
    let (events, min_id, has_more) = state.bus.history_page(before, limit);
-    // Apply the same enrichment as the live SSE path so history replay
+    // Classify with the same function the live SSE path uses so history
-    // and live tail deliver identical shapes. The DB stores raw events.
+    // replay and live tail deliver identical shapes. The DB stores raw
-    let events: Vec<_> = events
+    // events; classification is applied at read time here (see
    // `crate::term_msg`). One `ClassifyCtx` for the whole page — tool_use→
    // name correlation (for markdown-vs-plain `recv` result bodies) only
    // works within a single page/connection, not across the live/history
    // boundary; see that module's doc for why that's an accepted
    // degradation.
    let mut ctx = ClassifyCtx::default();
    let events: Vec<TermEnvelope> = events
        .into_iter()
-        .map(|mut se| {
+        .filter_map(|se| {
-            if let crate::events::LiveEvent::Stream(ref mut v) = se.event {
+            let msgs = classify(&se.event, &mut ctx);
-                crate::stream_enrich::enrich(v);
+            if msgs.is_empty() {
                None
            } else {
                Some(TermEnvelope {
                    ts: se.ts,
                    seq: None,
                    msgs,
                })
            }
            se
        })
        .collect();
    Json(EventsHistoryBody {
@ -81,23 +117,29 @@ pub(super) async fn events_stream(
    // stream rather than emitted to the bus — a bus emit would spam every
    // already-connected client with a spurious note each time anyone opens
    // the stream.
-    let hello = Event::default().data(
+    let hello_envelope = TermEnvelope {
-        serde_json::to_string(&crate::events::LiveEvent::Note {
+        ts: chrono::Utc::now().timestamp(),
-            text: "live stream attached".into(),
+        seq: None,
-        })
+        msgs: vec![TermMsg::new(Level::Debug, "live stream attached")],
-        .unwrap_or_default(),
+    };
-    );
+    let hello = Event::default().data(serde_json::to_string(&hello_envelope).unwrap_or_default());
-    let live = BroadcastStream::new(rx).filter_map(|res| {
+    // One `ClassifyCtx` per connection, moved into the closure — tool_use→
-        let mut ev = res.ok()?;
+    // name correlation persists for the connection's lifetime (see
-        // Enrich stream-json values with pre-computed display fields
+    // `crate::term_msg::ClassifyCtx`'s doc for the history-page boundary
-        // (`_icon`, `_summary`, `_category`) so the frontend doesn't need to
+    // this doesn't cross).
-        // duplicate the dispatch logic. The DB stores raw events; enrichment
+    let mut ctx = ClassifyCtx::default();
-        // is applied here so both the live tail and the history endpoint
+    let live = BroadcastStream::new(rx).filter_map(move |res| {
-        // deliver the same shape (see `events_history` above).
+        let ev = res.ok()?;
-        if let crate::events::LiveEvent::Stream(ref mut v) = ev.event {
+        let msgs = classify(&ev.event, &mut ctx);
-            crate::stream_enrich::enrich(v);
+        if msgs.is_empty() {
            return None;
        }
-        let json = serde_json::to_string(&ev).ok()?;
+        let envelope = TermEnvelope {
            ts: ev.ts,
            seq: Some(ev.seq),
            msgs,
        };
        let json = serde_json::to_string(&envelope).ok()?;
        Some(Ok(Event::default().data(json)))
    });
    let stream = tokio_stream::once(Ok(hello)).chain(live);
Author	SHA1	Message	Date
iris	5930efc29b	Trim negative-space comments per mara's review Don't state what a function/module doesn't do and where that happens instead — just describe what it does. Cut the "not something this function decides" / "not affected by this" / "not a placeholder for a later commit" asides from the doc comments touched in the last two commits.	2026-08-30 21:31:33 +02:00
iris	907567ef76	Fix stale default-open doc comments per argus's review ClassifyCtx's tool_use-id correlation gates markdown-vs-plain body format for a recv result, not open/collapsed state — that's always the operator's uniform preference now. 5 backend comments still described it as controlling "default-open" rendering, contradicting the actual render path and this PR's own rewritten docs. Also fixed useAgentState.ts's stale comment promising an SSE-triggered refresh model "in a later commit" — that's permanently off the table now that the terminal stream's kind tag is gone by design (argus flagged this as a drive-by, not blocking, but it's a one-line cause-and-effect of this same PR so fixing it here).	2026-08-30 21:28:34 +02:00
iris	cebf3c6ced	Drop classifyEvent.ts, render TermMsg directly Per review: StreamRow was meant to match what the server sends in TermMsg, not be a separate model needing a translation step. - classifyEvent.ts and streamRow.ts deleted; termMsg.ts holds the wire types (TermMsg/TermEnvelope) plus TermRow, a TermMsg with just the key/fromHistory bookkeeping Preact needs for list rendering. - Row.tsx renders a TermRow directly: level -> CSS class, empty summary + markdown body -> flat row, everything else with a body -> expandable details gated by the operator's preference. No separate classification step. - useLiveStream.ts drops ClassifyCtx (a single incrementing key counter didn't need a whole context object) and maps envelopes to rows inline. - docs/terminal-rendering.md trimmed substantially — was documenting more implementation detail than useful; points at stream_enrich.rs for the per-tool specifics instead of duplicating them in prose.	2026-08-30 21:23:28 +02:00
iris	5eefaa951d	Simplify terminal message shape to a uniform TermMsg Move terminal-row classification server-side into a new hive-agent/src/term_msg.rs, replacing the old JSON-mutation enrich()/stamped-field approach in stream_enrich.rs with one uniform wire shape: {icon?, level: debug\|info\|warn\|error, summary, body?, body_format?: markdown\|diff, coalesce_key?}. No more per-row `kind` tag or raw claude-JSON passthrough — every row is the same shape, with structural identity carried by icon + summary text instead of a CSS class per row kind. hive-agent/src/web_ui/stream.rs's history + SSE endpoints now both call term_msg::classify() and serve TermEnvelope{ts, seq?, msgs} frames; events that classify to zero rows (agent-state changes, drop-noise) never reach the wire. Frontend: classifyEvent.ts collapses from a large per-tool dispatch tree to a thin TermMsg -> StreamRow adapter. streamRow.ts/Row.tsx drop the now-dead meta/childText fields. terminal.css switches from a dozen-odd per-row-kind classes to four level-based color rules. Expand/collapse of a bodied row is now a uniform client-side decision (the operator's preference), no server-side per-tool override. docs/terminal-rendering.md rewritten to match.	2026-08-30 21:11:25 +02:00