fix: exclude failed WIP from live capacity holders

Failed parks left in the WIP column still counted as running agents and
file-scope lease holders, so they consumed maxWorktrees/maxConcurrent and
could serialize unrelated todos. Match review-lane semantics: status
failed is never a live top-level holder.

packages/core/src/__tests__/live-agent-count.test.ts

@@ -30,6 +30,21 @@ describe("live agent count predicates", () => {
     expect(isRunningAgentTask(task({ column: "in-progress", columnCountsTowardWip: true, userPaused: true }))).toBe(false);
   });
 
+  it("does not count failed WIP (or any failed row) as a live capacity holder", () => {
+    // Failed parks remain in WIP until rebound/operator action but must free maxWorktrees/maxConcurrent.
+    expect(isRunningAgentTask(task({ column: "in-progress", columnCountsTowardWip: true, status: "failed" }))).toBe(false);
+    expect(isRunningAgentTask(task({ column: "working", columnCountsTowardWip: true, status: "failed" }))).toBe(false);
+    expect(isRunningAgentTask(task({
+      column: "in-progress",
+      columnCountsTowardWip: true,
+      status: "failed",
+      workflowStepResults: [{ workflowStepId: "code-review", workflowStepName: "Code Review", status: "pending" as const, startedAt: "2026-08-01T00:00:00.000Z" }],
+    }))).toBe(false);
+    expect(isRunningAgentTask(task({ column: "in-review", columnIsReviewOrMerge: true, status: "failed" }))).toBe(false);
+    // Still Waiting only for intake/hold — failed WIP is neither running nor waiting.
+    expect(isWaitingAgentTask(task({ column: "in-progress", columnCountsTowardWip: true, status: "failed" }))).toBe(false);
+  });
+
   it("counts only active review/merge statuses and excludes terminal columns", () => {
     for (const status of ["merging", "merging-pr", "merging-fix", "reviewing", "landing", "fixing"]) {
       expect(isRunningAgentTask(task({ column: "review", status, columnIsReviewOrMerge: true }))).toBe(true);

packages/core/src/live-agent-count.ts

@@ -171,6 +171,8 @@ function terminalKind(task: RunningAgentTaskShape): ColumnTerminalKind {
  * Active review/merge statuses count only in review/merge columns.
  * A live `pending` workflow-step lease (e.g. an in-flight Code Review gate)
  * counts in any non-terminal column, since gate sessions run with null status.
+ * A parked `failed` row never counts, even in WIP — failed WIP is operator-
+ * visible debris, not a live agent (must not consume maxWorktrees/maxConcurrent).
  */
 /*
 FNXC:ConcurrencyIndicators 2026-07-30-03:40 DELIBERATE-LITERAL: the no-enrichment fallback only.
@@ -180,6 +182,15 @@ guess, and a wrong guess under-reports the queued total. Fix at the CALLER by pa
 */
 export function isRunningAgentTask(task: RunningAgentTaskShape): boolean {
   if (task.paused || task.userPaused || terminalKind(task) !== "none") return false;
+  /*
+  FNXC:ConcurrencyIndicators 2026-08-01-19:22:
+  Failed WIP parks (honest-blocked, graph parse failure, exhausted recovery) remain in the
+  WIP column until an operator or rebound moves them, but they are not live agents. Counting
+  them filled maxWorktrees/maxConcurrent and blocked ready work (e.g. FN-8704 sitting
+  in-progress/failed while todos waited on capacity). Match review-lane semantics, which
+  already exclude status:"failed" from active merge holders.
+  */
+  if (task.status === "failed") return false;
   if (task.status === "planning") return true;
   // Review statuses are not globally live: a stale status in intake/WIP must not consume capacity.
   if (ACTIVE_IN_REVIEW_AGENT_STATUSES.has(String(task.status ?? ""))) {

packages/engine/src/scheduler.ts

@@ -563,6 +563,13 @@ export function shouldHoldActiveFileScopeLease(
   */
   if (task.paused || task.userPaused) return false;
   /*
+  FNXC:OverlapScheduling 2026-08-01-19:22:
+  A failed park is not live work. Review already dropped status:"failed"; WIP must too, or a
+  stranded failed in-progress card keeps its file-scope lease and serializes unrelated todos
+  while consuming no real agent (same class as the capacity-holder fix in isRunningAgentTask).
+  */
+  if (task.status === "failed") return false;
+  /*
   DELIBERATE-LITERAL — the documented default for an unconverted caller, reviewed 2026-07-30-20:40.
   Both scheduler call sites now pass the resolved answer, so these defaults are dead on the
   scheduler's own path; they exist for the self-healing / repair callers this predicate is shared
@@ -2168,7 +2175,15 @@ export class Scheduler {
       */
       const isReviewColumnTask = (task: Task): boolean =>
         isReviewColumnRole(columnFlagsForTask(task), task.column);
-      const wipTaskIds = tasks.filter(isWipColumnTask).map((task) => task.id);
+      /*
+      FNXC:ConcurrencyIndicators 2026-08-01-19:22:
+      Failed WIP is not a live holder (isRunningAgentTask). Keep the WIP id list aligned so
+      diagnostic maxConcurrent holders and any WIP-only arithmetic do not re-count stranded failed
+      parks that the worktree ledger already excludes.
+      */
+      const wipTaskIds = tasks
+        .filter((task) => isWipColumnTask(task) && task.status !== "failed")
+        .map((task) => task.id);
       /*
       FNXC:WorktreeCapacity 2026-08-01-04:38 (inactive retained-worktree capacity inversion):
       Worktree capacity is a LIVE-TASK budget, not a count of directories retained on disk. The