docs(agents): trim more excessive compaction comments

Tighten the over-long comments in the compaction indicator helper + component
and the reconstruction watermark block, dropping a redundant inline restatement.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: kevin-dp

packages/agents-runtime/src/timeline-context.ts

@@ -497,10 +497,8 @@ export function timelineMessages(
     if (item.at < since) {
       continue
     }
-    // The watermark is the order of the last summarized item (the snapshot
-    // head), so everything at or below it is folded into the summary.
     if (item.at <= compactionWatermark) {
-      continue
+      continue // folded into the summary
     }
     // Place the summary at the watermark boundary, before the first kept item.
     emitSummary()

packages/agents-server-ui/src/components/CompactionIndicator.tsx

@@ -8,19 +8,11 @@ import {
 import styles from './CompactionIndicator.module.css'
 
 /**
- * Live indicator shown while a context compaction is in flight.
- *
- * The runtime writes a compaction checkpoint row with `attrs.status: "running"`
- * before summarizing and supersedes it with a `complete` (or `failed`) row when
- * done. We read the latest such row reactively; while it's `running` we show
- * "Compacting context…" so the user understands the pause. It clears the moment
- * compaction completes.
- *
- * A summarize is bounded by a hard timeout (~120s) after which a terminal row is
- * always written, so a `running` row that lingers well past that is orphaned —
- * its process crashed before writing the terminal row. We stop showing the
- * spinner for such a row (and self-clear via a timer so it disappears even with
- * no further events). See `lib/compactionIndicator` for the staleness rule.
+ * Spinner shown while a context compaction is in flight. The runtime writes a
+ * `running` compaction checkpoint before summarizing and supersedes it with
+ * `complete`/`failed` when done; we show the spinner while the latest is
+ * `running`. Orphaned (crashed) running rows are cleared after a timeout — see
+ * `lib/compactionIndicator`.
  */
 
 interface CheckpointRow {

packages/agents-server-ui/src/lib/compactionIndicator.ts

@@ -1,23 +1,14 @@
-/**
- * Staleness logic for the "Compacting…" indicator, extracted so it can be unit
- * tested without a DOM/render harness (this package has no React-render test
- * setup).
- */
+// Staleness logic for the "Compacting…" indicator, split out so it's unit
+// testable without a render harness.
 
-/**
- * Beyond this age a still-`running` compaction checkpoint is treated as orphaned
- * (its process crashed before writing a terminal `complete`/`failed` row). A
- * summarize is bounded by a ~120s hard timeout after which a terminal row is
- * always written, so 150s comfortably clears only genuinely-crashed runs.
- */
+/** A `running` checkpoint older than this is orphaned (its process crashed
+ * before a terminal row); comfortably above the summarize timeout. */
 export const STALE_RUNNING_MS = 150_000
 
 /**
- * Whether a `running` checkpoint with the given `timestamp` (ISO string) should
- * be treated as orphaned at `now` (ms). A missing or unparseable timestamp is
- * treated as NOT orphaned — we can't prove staleness, and hiding a genuinely
- * in-flight compaction is worse than briefly over-showing one. (insertContext
- * always stamps a timestamp, so this only guards against a schema regression.)
+ * Whether a `running` checkpoint with `timestamp` (ISO) is orphaned at `now`
+ * (ms). A missing/unparseable timestamp counts as live — we can't prove
+ * staleness, and hiding an in-flight compaction is worse than over-showing one.
  */
 export function isRunningCheckpointOrphaned(
   timestamp: string | undefined,