feat: complete Loop lifecycle Web, docs, and QA (stack 3/3) - #302

web/src/systems/loops/components/run-page/loop-run-waits-rail.tsx-22-42 (1)

22-42: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Fix the "Waiting on you" row: missing nodeId and possible label mismatch.

Two issues in this block:

1. "Needs attention" and "Quarantined" attach the single node's id when their count is 1 (lines 35, 40). "Waiting on you" does not, even when waiting === 1, though waiting is computed the same way. This is an inconsistency in the row-building logic.
2. waiting counts every node with any open wait, including timer waits. In loop-run-parked-panels.tsx, the timer wait copy explicitly states the wait "resumes by itself" — it does not need the user. Labeling this count "Waiting on you" implies user action is needed even for self-resuming timer waits.

As per path instructions, "never imply unsupported metrics, controls, states, or repair paths" for web/**/*.{tsx,md} files. Confirm whether the label should be relabeled (e.g., "Open waits") or the count should be scoped to decision-needed waits only.

🛠️ Proposed fix for the missing `nodeId`

   const rows: { label: string; value: number; nodeId?: string }[] = [
-    { label: "Waiting on you", value: waiting },
+    {
+      label: "Waiting on you",
+      value: waiting,
+      nodeId: waiting === 1 ? nodes.find(node => node.waits.some(isOpenWait))?.nodeId : undefined,
+    },

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@web/src/systems/loops/components/run-page/loop-run-waits-rail.tsx` around
lines 22 - 42, Update the row-building logic around isOpenWait so the “Waiting
on you” metric does not imply user action for self-resuming timer waits: either
scope the count to decision-needed waits or relabel it to “Open waits” if all
open waits remain included. Add the matching single-node nodeId lookup to that
row, consistent with the “Needs attention” and “Quarantined” rows.

Source: Path instructions

web/src/systems/loops/components/stories/loop-run-lifecycle-fixtures.ts-205-221 (1)

205-221: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Align age_seconds with created_at in the wait fixtures.

The wait() factory sets created_at: minutesAgo(18). Two overrides change age_seconds but keep that default. The timer wait declares 360 seconds (6 minutes) and the approval wait declares 22 minutes. A component that renders created_at and a component that renders age_seconds will then show different ages in the same capture. Override created_at together with age_seconds.

🐛 Proposed fix

       wait({
         node_id: "collect_fixes",
         kind: "timer",
         resume_at: minutesAgo(-24),
+        created_at: minutesAgo(6),
         age_seconds: 360,
       }),
       wait({ node_id: "fix_batch", item_index: 3, kind: "event", age_seconds: 18 * 60 }),
       wait({
         node_id: "finalize_round",
         kind: "approval_escalation",
         claim_state: "intervention_required",
         escalation_cursor: 1,
+        created_at: minutesAgo(22),
         age_seconds: 22 * 60,
         next_escalation_at: minutesAgo(-8),
       }),

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@web/src/systems/loops/components/stories/loop-run-lifecycle-fixtures.ts`
around lines 205 - 221, Update the timer wait and approval_escalation wait
entries in the waits fixture to override created_at alongside their age_seconds
values, using timestamps that represent 6 minutes and 22 minutes ago
respectively. Keep the existing wait factory defaults and all other fixture
fields unchanged.

web/src/systems/loops/lib/loop-node-now-view.ts-110-118 (1)

110-118: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Guard the formatted time before adding the at clause.

clockTime returns "" when requested_at is absent or unparsable. The template at line 114 always produces a non-empty string, so the .filter(Boolean) at line 116 cannot remove it. The provenance line then ends with the dangling text at. Build the clause only when a time is available. The same pattern applies to line 154.

🐛 Proposed fix

 function pausedLine(node: LoopNodeLifecycle): LoopNodeNowLine {
   const provenance = node.pauseProvenance;
   const actor = provenance ? actorLabel(provenance.actor_kind, provenance.actor_id) : "";
   const byRule = provenance?.rule_id ? `rule ${provenance.rule_id}` : "";
   const by = byRule || actor;
+  const requestedAt = provenance ? clockTime(provenance.requested_at) : "";
   return {
     nodeId: node.nodeId,
     state: "paused",
     headline: `${node.label} is paused`,
     body: `${by === "" ? "Parked" : `Paused by ${by}`} at a safe point — the rest of the run keeps working.`,
     chip: null,
     provenance: provenance
       ? [
           `paused by ${by || "unknown"}`,
           provenance.reason ? `reason “${provenance.reason}”` : null,
-          `at ${clockTime(provenance.requested_at)}`,
+          requestedAt === "" ? null : `at ${requestedAt}`,
         ]
           .filter(Boolean)
           .join(" · ")
       : null,

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@web/src/systems/loops/lib/loop-node-now-view.ts` around lines 110 - 118,
Update the provenance formatting in the visible block and the corresponding
block around the second occurrence to call clockTime(provenance.requested_at)
once, then add the “at” clause only when that formatted value is non-empty;
preserve the existing filtering and joining behavior so no dangling “at” remains
for missing or unparsable timestamps.

web/src/systems/loops/components/run-page/loop-quarantine-sheet.tsx-222-233 (1)

222-233: 📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win

Add unit/e2e coverage for the requeue gating rule.

The LoopQuarantineSheet only uses node.quarantined for the loop-quarantine-requeue button, and existing Stories/rg output do not show a canonical assertion for the non-quarantined branch. A test should cover both quarantined and already-out-of-quarantine states so refactors do not expose a rejected requeue action.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@web/src/systems/loops/components/run-page/loop-quarantine-sheet.tsx` around
lines 222 - 233, Add unit or end-to-end coverage for the requeue control in
LoopQuarantineSheet, asserting that loop-quarantine-requeue is available for
quarantined nodes and absent or unavailable for nodes where node.quarantined is
false; preserve the existing isRequeuePending disabled behavior.

web/src/systems/loops/lib/loop-reliability.ts-60-73 (1)

60-73: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Do not present the first node limits as one loop-wide policy.

firstDefined can select timeout from one node and deadline from another node. The resulting text can describe a time budget that no node has.

Collect limits by node. If values differ, identify the affected nodes or state that only some steps use each limit.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@web/src/systems/loops/lib/loop-reliability.ts` around lines 60 - 73, Update
deadlineTile to aggregate timeout and deadline values per node instead of
selecting independent firstDefined values across graph.nodes. When limits differ
or apply only to some steps, describe the affected nodes or explicitly indicate
partial coverage, ensuring the generated clauses and codes never imply a
loop-wide policy that no node has.

web/src/systems/loops/lib/loop-graph.ts-101-104 (1)

101-104: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Read wait expiry from params.expires.

Wait controls store their configuration in params. The projection reads root-level expires, so an authored expiry never reaches waitTile. The test fixture repeats this invalid shape and hides the defect.

• web/src/systems/loops/lib/loop-graph.ts#L101-L104: read after from record.params.expires.
• web/src/systems/loops/lib/__tests__/loop-reliability.test.ts#L50-L59: create the wait expiry under params and retain the expiry assertion.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@web/src/systems/loops/lib/loop-graph.ts` around lines 101 - 104, Read wait
expiry from record.params.expires in the loop-graph projection so authored
values reach waitTile. Update web/src/systems/loops/lib/loop-graph.ts lines
101-104 accordingly; in
web/src/systems/loops/lib/__tests__/loop-reliability.test.ts lines 50-59, move
the fixture expiry under params and retain the existing expiry assertion.

web/src/systems/loops/components/editor/loop-editor-publish-rejected-strip.tsx-34-39 (1)

34-39: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Show the version-kept status independently of the issue list, and do not hardcode 422.

The badge renders only when issues.length > 0. A rejection that carries a message but no issue list then hides the statement that the compare-and-swap version was kept, which is the main point stated in the component docstring. The literal 422 also assumes the status code; a rejection returned with another status displays a wrong code. Pass the status as a prop, or drop the code from the text.

🩹 Proposed change

       <p className="flex items-center gap-2 text-form-label text-danger">
         <AlertCircle aria-hidden="true" className="size-3.5 shrink-0" />
         {message}
-        {issues.length > 0 ? (
-          <span className="ml-auto font-mono text-mono-id text-subtle">
-            422 · expected_version v{version ?? "?"} kept
-          </span>
-        ) : null}
+        <span className="ml-auto font-mono text-mono-id text-subtle">
+          expected_version v{version ?? "?"} kept
+        </span>
       </p>

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In
`@web/src/systems/loops/components/editor/loop-editor-publish-rejected-strip.tsx`
around lines 34 - 39, Update the version-kept badge in the component rendering
around `issues` so it renders independently of `issues.length`, including when a
rejection has a message but no issues. Remove the hardcoded `422` by either
accepting and rendering the actual rejection status through the component props
or omitting the status code from the badge text.

web/src/systems/loops/lib/loop-runs-view.ts-174-177 (1)

174-177: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Align the fallback order for kind and reference.

kind prefers started_origin_kind, while ref prefers started_by_ref. When a run carries both pairs and they differ, the line pairs the origin kind with the actor reference, for example schedule · pedro. That reads as provenance the daemon never recorded. Pick one pair first and fall back to the other pair as a unit.

🩹 Proposed fix

-  const kind = run.started_origin_kind || run.started_by_kind || "";
-  const ref = run.started_by_ref || run.started_origin_ref || "";
+  const originKind = run.started_origin_kind || "";
+  const originRef = run.started_origin_ref || "";
+  const kind = originKind || run.started_by_kind || "";
+  const ref = originKind ? originRef : run.started_by_ref || "";
   if (kind && ref) return `${kind} · ${ref}`;
   return kind || ref || "—";

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@web/src/systems/loops/lib/loop-runs-view.ts` around lines 174 - 177, Update
the run label construction so the kind and reference values are selected from
the same provenance pair: prefer started_origin_kind with started_origin_ref,
then fall back together to started_by_kind with started_by_ref. Preserve the
existing formatting and em-dash fallback when no complete pair is available.

web/src/systems/loops/lib/loop-start-kinds.ts-13-17 (1)

13-17: 📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win

Expose thisFormDeclared in the run-form UI or remove it.

LoopRunWaysToStart only displays thisForm and alsoDeclared, while every run post is always http. Add kind/http to this form when thisFormDeclared is false, or remove the unused field.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@web/src/systems/loops/lib/loop-start-kinds.ts` around lines 13 - 17, Update
LoopRunWaysToStart to use thisFormDeclared: when the current run form is not
declared, include the http kind in the displayed run-start options; otherwise
preserve the existing thisForm and alsoDeclared behavior. Alternatively, remove
thisFormDeclared from the relevant type and producer if it is not needed.

web/src/systems/loops/hooks/use-loop-node-inventory.ts-43-43 (1)

43-43: 🩺 Stability & Availability | 🟡 Minor | ⚡ Quick win

Use query.isLoading for the inventory skeleton state.

When no inventory state is selected, the query is disabled and v5 keeps status: "pending" with fetchStatus: "idle", so query.isPending is true while no request runs. Consumers that render a loading view from isLoading can show it permanently in that state; use query.isLoading, which is isPending && isFetching.

🐛 Proposed fix

-    isLoading: query.isPending,
+    isLoading: query.isLoading,

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@web/src/systems/loops/hooks/use-loop-node-inventory.ts` at line 43, Update
the inventory query state mapping in useLoopNodeInventory to expose
query.isLoading instead of query.isPending for isLoading, preserving the
intended skeleton behavior when the disabled query is pending without fetching.

web/src/systems/loops/components/run-page/loop-run-control-dialog.tsx-17-18 (1)

17-18: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

The error prop is declared but the only consumer does not pass it.

The doc for error states that the daemon's rejection is shown inline. In web/src/systems/os/apps/loops/loop-run-detail-location.tsx, LoopRunControlDialog receives generation, isPending, onConfirm, onOpenChange, runId, status, and verb, but no error. LoopNodeControlDialog in the same file does receive error={nodeControls.answer}. A rejected run cancel or kill therefore shows no inline answer, unlike the node path.

Pass the run action's rejection text into error at the consumer, or remove the prop if the run path reports failures elsewhere.

Also applies to: 50-50

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@web/src/systems/loops/components/run-page/loop-run-control-dialog.tsx` around
lines 17 - 18, Update the LoopRunControlDialog invocation in
loop-run-detail-location.tsx to pass the run action’s rejection text through its
existing error prop, matching LoopNodeControlDialog’s use of
nodeControls.answer. Preserve the current dialog behavior and use the run
control state’s corresponding answer/error symbol.

web/src/systems/loops/lib/loop-node-verb-copy.ts-124-134 (1)

124-134: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Do not state an episode number when the payload carries no quarantine entry.

(node.quarantineEntry?.episodes.length ?? 1) + 1 renders "episode 2" when quarantineEntry is absent. The payload never reported an episode in that case. loopNodeStateStrip uses ?? 0 for the same field and omits the episode when it is zero, so the two helpers disagree. The file header states that a missing detail must shorten the sentence rather than be invented.

Drop the episode sentence when the entry is absent.

🐛 Proposed fix

-  requeue: node => ({
-    eyebrow: "Requeue node",
-    title: `Send ${node.nodeId} back to the queue?`,
-    body: `The lane leaves quarantine and runs again through the normal path, with all limits applying. If it fails again, it returns here as episode ${
-      (node.quarantineEntry?.episodes.length ?? 1) + 1
-    }.`,
+  requeue: node => {
+    const episodes = node.quarantineEntry?.episodes.length;
+    const nextEpisode =
+      episodes === undefined ? "" : ` If it fails again, it returns here as episode ${episodes + 1}.`;
+    return {
+      eyebrow: "Requeue node",
+      title: `Send ${node.nodeId} back to the queue?`,
+      body: `The lane leaves quarantine and runs again through the normal path, with all limits applying.${nextEpisode}`,
       confirmLabel: "Requeue lane",
       cancelLabel: "Keep quarantined",
       tone: "accent",
       micro: "node_requeued · origin requeue · with provenance",
-  }),
+    };
+  },

As per coding guidelines: "never imply unsupported metrics, controls, states, or repair paths".

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@web/src/systems/loops/lib/loop-node-verb-copy.ts` around lines 124 - 134,
Update the requeue confirmation builder to omit the episode sentence when
node.quarantineEntry is absent, rather than defaulting its count to produce an
invented episode number. Preserve the existing episode calculation and wording
when quarantineEntry is present, and follow the shortening behavior used by
loopNodeStateStrip.

Source: Coding guidelines

web/src/systems/loops/lib/loop-run-story-lifecycle-rows.ts-176-186 (1)

176-186: 📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win

Separate the raw cause from the following sentence.

cause comes straight from the payload's failure.cause and is usually a machine token, for example connection_refused. The template joins it with a single space before a capitalized sentence, so the row reads "connection_refused Its clock is suspended…". Add an explicit separator so the two parts do not read as one sentence.

🐛 Proposed fix

     sub:
-      `${cause === "" ? "" : `${cause} `}` +
+      `${cause === "" ? "" : `${cause} — `}` +
       "Its clock is suspended and no attempts are being spent; the rest of the run keeps going.",

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@web/src/systems/loops/lib/loop-run-story-lifecycle-rows.ts` around lines 176
- 186, Update the node_quarantined row construction to explicitly separate the
raw cause from the following sentence in the sub text. Adjust the cause prefix
logic near the node_quarantined return so non-empty causes receive appropriate
punctuation or delimiter, while preserving the no-cause output and existing
sentence text.

web/src/systems/loops/components/run-page/loop-node-row-actions.tsx-39-52 (1)

39-52: 📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win

Use the requeue icon when the promoted verb is requeue.

The promoted button always renders Play. LoopNodeControlMenu renders RotateCcw for the same requeue verb. The same action therefore shows two different icons in the same row.

🐛 Proposed fix to match the menu icon

-import { Play } from "lucide-react";
+import { Play, RotateCcw } from "lucide-react";

       {primary ? (
         <Button
           className="min-h-6"
           data-testid={`loop-node-primary-${primary}-${node.nodeId}`}
           disabled={isPending}
           onClick={() => onVerb(primary, node)}
           size="sm"
           type="button"
           variant="outline"
         >
-          <Play aria-hidden="true" className="size-3" />
-          {primary === "resume" ? "Resume" : "Requeue…"}
+          {primary === "resume" ? (
+            <Play aria-hidden="true" className="size-3" />
+          ) : (
+            <RotateCcw aria-hidden="true" className="size-3" />
+          )}
+          {primary === "resume" ? "Resume" : "Requeue…"}
         </Button>
       ) : null}

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@web/src/systems/loops/components/run-page/loop-node-row-actions.tsx` around
lines 39 - 52, Update the primary action button in the loop-node row actions to
render the same icon as LoopNodeControlMenu for each promoted verb: keep Play
for resume and use RotateCcw for requeue. Preserve the existing button behavior,
labels, and accessibility attributes.

web/src/systems/loops/lib/loop-run-progress.ts-140-153 (1)

140-153: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Derive the parked noun from the fan-out node, not from every parked node.

parkedNoun scans all lifecycle nodes. The parked segments belong to one node only, the widest fan-out node. If that node is paused while an unrelated node is quarantined, states.size === 2 and the meta line says "parked" instead of "paused". latestFanOut already resolves widestId, so return it and read that node's state.

♻️ Proposed fix

 interface FanOutBranches {
   /** Branch statuses ordered by ascending `item_index`. */
   statuses: string[];
   /** True when the node owning these branches is parked by the daemon. */
   parked: boolean;
+  /** The node that owns these branches. */
+  nodeId: string;
 }

-  return { statuses, parked: widestId !== null && parked.has(widestId) };
+  return { statuses, parked: widestId !== null && parked.has(widestId), nodeId: widestId ?? "" };

-function parkedNoun(nodes: readonly LoopNodeLifecycle[]): string {
-  const states = new Set<LoopNodeLifecycle["state"]>();
-  for (const node of nodes) {
-    if (node.parked) states.add(node.state);
-  }
-  if (states.size === 1) {
-    const [only] = [...states];
-    if (only === "paused") return "paused";
-    if (only === "quarantined") return "quarantined";
-    if (only === "waiting") return "waiting";
-  }
-  return "parked";
-}
+function parkedNoun(nodes: readonly LoopNodeLifecycle[], nodeId: string): string {
+  const state = nodes.find(node => node.parked && node.nodeId === nodeId)?.state;
+  if (state === "paused" || state === "quarantined" || state === "waiting") return state;
+  return "parked";
+}

-  const noun = parkedNoun(nodes);
+  const noun = parkedNoun(nodes, fanOut.nodeId);

Also applies to: 192-200

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@web/src/systems/loops/lib/loop-run-progress.ts` around lines 140 - 153,
Update parkedNoun to derive the noun from the node identified by latestFanOut’s
resolved widestId, rather than aggregating states across all parked lifecycle
nodes. Read that fan-out node’s state and return paused, quarantined, or waiting
when applicable, otherwise return parked; preserve the existing fallback
behavior when the node or state is unavailable.

web/src/systems/loops/lib/loop-node-inventory.ts-73-108 (1)

73-108: 📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win

Format the timestamps before showing them to operators.

waitReason prints wait.resume_at and retryReason prints output.next_attempt_at directly. These fields are raw API timestamps, so the row reads like timer — resumes 2026-08-03T14:00:00Z. Reuse the existing time formatting helper used elsewhere in the loops system, or express the value as a relative duration.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@web/src/systems/loops/lib/loop-node-inventory.ts` around lines 73 - 108,
Update waitReason and retryReason to format wait.resume_at and
output.next_attempt_at through the existing loops-system time formatting helper
before displaying them. Preserve the current message structure and fallback
behavior, but ensure operators see a human-readable or relative timestamp
instead of raw API ISO values.

web/src/systems/loops/components/run-page/loop-run-outcome-card.tsx-172-178 (1)

172-178: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Treat an empty cause as absent.

cause ?? failure?.code keeps an empty string. If the terminal frame carries cause: "", causeLabel becomes "", the segment is dropped, and the available failure.code is never shown. Use a truthiness fallback.

🐛 Proposed fix

-  const causeLabel = cause ?? failure?.code;
+  const causeLabel = cause || failure?.code;

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@web/src/systems/loops/components/run-page/loop-run-outcome-card.tsx` around
lines 172 - 178, Update the causeLabel assignment in the loop-run outcome card
to use a truthiness fallback, so an empty cause is treated as absent and
failure?.code is used instead. Preserve the existing precedence for non-empty
causes and the subsequent micro segment behavior.

web/src/systems/os/apps/loops/loop-runs-location.tsx-128-147 (1)

128-147: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Source loopOptions from the loop catalog before building the inventory view.

LoopNodeInventoryViewProps.loopOptions says the filter comes from the workspace catalog, but this passes only loop names from loaded runs. If runsQuery loads first but the inventory query is still pending, the node inventory renders with no loop options. Use the catalog query for this option list, or rename the prop contract to “loop names seen in the loaded runs.”

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@web/src/systems/os/apps/loops/loop-runs-location.tsx` around lines 128 - 147,
Update the loopOptions value passed to LoopNodeInventoryView to source loop
names from the workspace loop catalog query rather than runs, ensuring options
remain available while inventory data is pending. Preserve the existing sorting
and deduplication behavior, and use the catalog query’s established data symbol.

web/src/systems/loops/components/run-page/loop-run-story-timeline.tsx (1)

65-76: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Use distinct icons for "retry" and "requeued".

retry and requeued both map to RotateCcw. These are different lifecycle concepts: a scheduled automatic retry and an actor-initiated requeue. The companion type file documents a "one fixed lucide icon per concept" rule. Reusing the same icon for two different concepts makes them indistinguishable in the timeline.

Assign a separate icon to one of the two states.

As per coding guidelines, packages/ui/**/*-adjacent web conventions call for distinguishable, consistent iconography per concept; the in-file comment states: "One fixed lucide icon per concept across the domain, per DESIGN-LESSONS L7."

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@web/src/systems/loops/components/run-page/loop-run-story-timeline.tsx` around
lines 65 - 76, Update the icon mapping so the retry and requeued states use
distinct Lucide icons, preserving the existing icon for one state and assigning
an appropriate unused icon to the other. Keep the one-fixed-icon-per-concept
convention in the status mapping.

internal/loop/coordinator_wait.go (1)

226-227: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Consider reusing generationOutputRefForPayload here.

The code sets the reference, the runtime payload, and later appends a GenerationOutputBlob (Line 245) by hand. generationOutputRefForPayload(payload, outputBlobs, at) already performs the reference and blob steps in one place, as used in internal/loop/coordinator_watch_events.go Line 281. Reusing it keeps one owner for externalized payload persistence and prevents future drift, for example a missed blob append.

The current order is correct, because setGenerationOutputRef clears runtimePayload.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@internal/loop/coordinator_wait.go` around lines 226 - 227, Replace the manual
payload reference and blob handling in the coordinator wait flow with
generationOutputRefForPayload, passing payload, outputBlobs, and the existing at
value. Remove the redundant setGenerationOutputRef, runtimePayload assignment,
and later GenerationOutputBlob append while preserving the current ordering and
resulting output.

internal/loop/generation_output_runtime.go (1)

34-57: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Accept a typed reader instead of any.

generationOutputRuntimeView takes reader any and asserts the interface at runtime. A typed parameter moves the contract to compile time and removes the assertion branch. Callers that hold a store value can pass it directly; the function can still reject a nil reader when a content-addressed reference exists.

Also consider renaming generationOutputRuntimeRef: it returns a payload document when runtimePayload is set, not a reference.

As per coding guidelines: "Define interfaces where they are consumed, use direct interface calls".

♻️ Proposed typing change

 func generationOutputRuntimeView(
 	ctx context.Context,
-	reader any,
+	reader GenerationOutputPayloadReader,
 	scope generationOutputRuntimeScope,
 	outputs []GenerationOutput,
 ) ([]GenerationOutput, error) {
 	view := cloneGenerationOutputs(outputs)
 	if len(view) == 0 {
 		return view, nil
 	}
-	var payloadReader GenerationOutputPayloadReader
 	resolved := make(map[GenerationOutputPayloadKey]json.RawMessage)
 	for index := range view {
 		ref := strings.TrimSpace(view[index].OutputRef)
 		if !OutputRefLooksContentAddressed(ref) {
 			continue
 		}
-		if payloadReader == nil {
-			var ok bool
-			payloadReader, ok = reader.(GenerationOutputPayloadReader)
-			if !ok || payloadReader == nil {
-				return nil, fmt.Errorf("%w: generation output payload reader is required", ErrValidation)
-			}
-		}
+		if reader == nil {
+			return nil, fmt.Errorf("%w: generation output payload reader is required", ErrValidation)
+		}

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@internal/loop/generation_output_runtime.go` around lines 34 - 57, Change
generationOutputRuntimeView to accept GenerationOutputPayloadReader directly
instead of reader any, remove the runtime type assertion and related
payloadReader initialization, and call the typed reader interface directly when
resolving content-addressed outputs. Preserve validation by rejecting a nil
reader only when such a reference requires payload access. Rename
generationOutputRuntimeRef to reflect that it returns a payload document when
runtimePayload is set, and update all call sites.

Source: Coding guidelines

web/src/systems/loops/lib/loop-node-schema-types.ts (1)

97-98: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Export one shared WAIT_MODES list next to LoopWaitMode.

The same vocabulary is declared twice as a value: web/src/systems/loops/lib/loop-node-wait-fields.ts Line 14 and web/src/systems/loops/components/editor/loop-editor-wait-mode.tsx Line 15. If the DSL adds a fourth wait mode, one list can be updated and the other missed. Declare the list once and derive the type from it, then import the list in both consumers.

♻️ Proposed single source of truth

-/** The closed `wait` discriminator vocabulary (`dsl.WaitParams`). */
-export type LoopWaitMode = "for" | "until" | "event";
+/** The closed `wait` discriminator vocabulary (`dsl.WaitParams`). */
+export const WAIT_MODES = ["for", "until", "event"] as const;
+export type LoopWaitMode = (typeof WAIT_MODES)[number];

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@web/src/systems/loops/lib/loop-node-schema-types.ts` around lines 97 - 98,
Define and export a shared WAIT_MODES constant beside LoopWaitMode, deriving
LoopWaitMode from its literal values instead of maintaining a separate union.
Update loop-node-wait-fields.ts and loop-editor-wait-mode.tsx to import and
reuse WAIT_MODES, removing their duplicated wait-mode value lists.

internal/loop/generation_snapshot.go (1)

44-47: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Document that runtimePayload is not serialized.

ExpectedEpoch carries a comment that explains its non-durable role. runtimePayload has no comment. encoding/json ignores unexported fields, so this payload is dropped on every snapshot marshal or unmarshal. Add a short comment that states the field is runtime-only and is hydrated in-package, so future readers do not expect it to survive a store round-trip.

📝 Proposed comment

 	ExpectedEpoch  *int64 `json:"expected_epoch,omitempty"`
+	// runtimePayload holds the hydrated output payload for in-process reads only.
+	// It is intentionally unexported and never serialized into the snapshot.
 	runtimePayload json.RawMessage

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@internal/loop/generation_snapshot.go` around lines 44 - 47, Add a concise
comment above runtimePayload stating that it is runtime-only, is not serialized
by encoding/json, and is hydrated in-package rather than preserved across
snapshot store round-trips.

web/src/systems/loops/components/stories/loop-run-scenario-props.ts (1)

39-51: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Remove the duplicate nodeLifecycles assignment.

Line 40 spreads view, which already contains nodeLifecycles. Line 49 sets the same key again from the same source. Keeping both creates two places to change if the source ever differs.

♻️ Proposed change

     workspaceLabel: "Home",
     versionLabel: `v${run.definition_version} · pinned`,
-    nodeLifecycles: view.nodeLifecycles,
     nodeNowLines: buildNodeNowLines(view.nodeLifecycles, view.graph, live.retrySchedules),

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@web/src/systems/loops/components/stories/loop-run-scenario-props.ts` around
lines 39 - 51, Remove the explicit nodeLifecycles property from the returned
object in the scenario props builder, leaving the value supplied by the ...view
spread while preserving the derived nodesById and nodeNowLines fields.

internal/loop/goal/types_binding.go (1)

139-154: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

State the mutual exclusion on CellFence in the contract.

validateBindingActivationOwner rejects a request that sets CellFence together with CheckpointKey, ExpectedControlEpoch, or GrantID. That rule is only visible in the store package. Add a doc comment on CellFence that names the exclusive alternative, so callers learn the constraint from the contract type.

📝 Proposed comment

 type ActivateBindingRequest struct {
 	Key                  BindingKey
 	CheckpointKey        *TurnKey
+	// CellFence activates from an ordinary loop cell. When it is set, the Goal
+	// owner fields (CheckpointKey, ExpectedControlEpoch, GrantID) must be empty.
 	CellFence            *BindingCellFence

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@internal/loop/goal/types_binding.go` around lines 139 - 154, Add a doc
comment to the CellFence field in ActivateBindingRequest stating that it is
mutually exclusive with CheckpointKey, ExpectedControlEpoch, and GrantID,
matching the validation in validateBindingActivationOwner.

web/src/systems/loops/components/run-page/loop-node-control-menu.tsx (1)

52-52: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Consider one source of truth for "destructive".

Two definitions now describe destructive verbs. DESTRUCTIVE_VERBS here controls placement below the separator and includes cancel and kill. LOOP_NODE_VERB_PRESENTATION controls the danger styling and marks only kill as destructive: true. The result is intentional today, but a future edit to one list will not update the other.

Derive the placement set from the presentation map, or rename the local set to state that it controls order only.

♻️ Suggested rename to separate placement from styling

-/** Destructive verbs sit below a separator so they are never the reflex click. */
-const DESTRUCTIVE_VERBS = new Set<LoopNodeVerb>(["cancel", "kill"]);
+/**
+ * Verbs that end work. They sit below a separator so they are never the reflex
+ * click. Danger styling is separate and comes from
+ * `LOOP_NODE_VERB_PRESENTATION[verb].destructive`.
+ */
+const STOPPING_VERBS = new Set<LoopNodeVerb>(["cancel", "kill"]);

Also applies to: 112-112

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@web/src/systems/loops/components/run-page/loop-node-control-menu.tsx` at line
52, Clarify the purpose of DESTRUCTIVE_VERBS in the loop-node control menu by
renaming it to indicate that it controls menu ordering or separator placement,
not danger styling. Update its references while preserving the existing
cancel/kill placement behavior and the separate LOOP_NODE_VERB_PRESENTATION
styling definitions.

web/src/systems/loops/components/__tests__/loop-run-page.test.tsx (1)

546-571: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Duplicated LoopNodeLifecycle fixtures across the loops test suites. Both files hand-build the full lifecycle literal because no shared fixture factory exists, so every new lifecycle field needs an edit in each copy.

• web/src/systems/loops/components/__tests__/loop-run-page.test.tsx#L546-L571: replace this factory and the two further literals in the file (Lines 646-670 and 738-762) with one shared factory that accepts overrides.
• web/src/systems/loops/lib/__tests__/loop-run-progress.test.ts#L82-L108: import the same shared factory and pass state, paused, and quarantined as overrides.

As per path instructions: "Check for shared test utilities usage to avoid duplication".

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@web/src/systems/loops/components/__tests__/loop-run-page.test.tsx` around
lines 546 - 571, Create one shared LoopNodeLifecycle fixture factory, move the
duplicated defaults from loop-run-page.test.tsx into it, and replace all three
lifecycle literals in that file with factory calls using overrides. In
web/src/systems/loops/lib/__tests__/loop-run-progress.test.ts lines 82-108,
import the shared factory and pass state, paused, and quarantined through
overrides.

Source: Path instructions

packages/ui/src/components/__tests__/confirm-dialog.test.tsx (1)

188-192: 🎯 Functional Correctness | 🔵 Trivial | ⚡ Quick win

Guard the order assertions against missing elements.

indexOf returns -1 for an absent test id. If the note stops rendering, -1 < order.indexOf("confirm-body") still passes, so the test reports success for a broken dialog. Assert presence first.

♻️ Proposed assertion hardening

+    expect(order).toContain("confirm-note");
+    expect(order).toContain("confirm-body");
+    expect(order).toContain("cancel-action");
     expect(order.indexOf("confirm-note")).toBeLessThan(order.indexOf("confirm-body"));
     expect(order.indexOf("confirm-body")).toBeLessThan(order.indexOf("cancel-action"));

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@packages/ui/src/components/__tests__/confirm-dialog.test.tsx` around lines
188 - 192, Harden the order assertions in the confirm-dialog test by first
asserting that confirm-note, confirm-body, and cancel-action are present in the
queried test IDs. Then retain the existing relative-order checks so missing
elements cannot produce a false pass through indexOf returning -1.

web/src/systems/loops/components/editor/loop-editor-json-field.tsx (1)

24-37: 🎯 Functional Correctness | 🔵 Trivial | ⚡ Quick win

Note: this component does not resync text when value changes without a remount.

text is derived from value only at mount, through the lazy useState initializer. If a consumer changes value for the same component instance (no key change), the textarea keeps showing the old text. This is the mechanism behind the stale-JSON issue flagged in loop-editor-effects.tsx for the shape-switch flow.

The fix for that call site is a key={shape}. If a future call site reuses this component where the same instance can receive a different underlying value (for example, a JSON field bound to a node field that changes without a selectionKey remount), the same staleness will recur. Consider that when adding new call sites.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@web/src/systems/loops/components/editor/loop-editor-json-field.tsx` around
lines 24 - 37, The current LoopEditorJsonField state initializes text only once,
so later value changes leave stale JSON displayed. At the
loop-editor-effects.tsx call site, force remounting by supplying key={shape};
when adding other call sites, also provide a key that changes whenever the
underlying value identity changes.

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

web/src/systems/loops/lib/loop-reliability.ts (1)

1-2: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Use the required @/* import alias.

Replace the relative imports with @/systems/loops/... imports.

Proposed change

-import type { LoopGraph } from "./loop-graph";
-import type { LoopDetail } from "../types";
+import type { LoopGraph } from "`@/systems/loops/lib/loop-graph`";
+import type { LoopDetail } from "`@/systems/loops/types`";

As per coding guidelines, use the @/* path alias for ./src/*.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@web/src/systems/loops/lib/loop-reliability.ts` around lines 1 - 2, Update the
type imports for LoopGraph and LoopDetail in loop-reliability.ts to use the
configured `@/systems/loops/`... path alias instead of relative paths, preserving
the existing imported symbols and type-only behavior.

Source: Coding guidelines

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes