diff --git a/src/services/agent-action-executor.ts b/src/services/agent-action-executor.ts
index abacfc62d2..78169a7744 100644
--- a/src/services/agent-action-executor.ts
+++ b/src/services/agent-action-executor.ts
@@ -45,6 +45,16 @@ const AGENT_ACTOR = "gittensory";
 // already used for mergeBlockedReason (db/repositories.ts) and the merge_blocked audit metadata below.
 const AUDIT_REASON_MAX_LENGTH = 280;
 
+function boundAuditReason(detail: string): string {
+  return detail.length > AUDIT_REASON_MAX_LENGTH ? `${detail.slice(0, AUDIT_REASON_MAX_LENGTH)}…` : detail;
+}
+
+function closeReasonsForAudit(action: PlannedAgentAction): string[] | undefined {
+  if (action.actionClass !== "close") return undefined;
+  const rawReasons = action.closeReasons?.length ? action.closeReasons : [action.reason];
+  return rawReasons.map((reason) => boundAuditReason(reason));
+}
+
 // The PR-visible action classes that require an elevated GitHub App write permission. Most use
 // `pull_requests: write`; merge uses `contents: write`; `label` mutates through the Issues API, so it is exempt
 // from this readiness gate.
@@ -203,11 +213,11 @@ export async function executeAgentMaintenanceActions(env: Env, ctx: AgentActionE
       // merge_blocked path below, db/repositories.ts's mergeBlockedReason) -- a heuristic close's reason is
       // built by joining every blocker title, so a PR with many blockers could otherwise write an arbitrarily
       // large, un-truncated string into audit_events.detail (#terminal-outcome-audit).
-      const boundedDetail = detail.length > AUDIT_REASON_MAX_LENGTH ? `${detail.slice(0, AUDIT_REASON_MAX_LENGTH)}…` : detail;
+      const boundedDetail = boundAuditReason(detail);
       outcomes.push({ actionClass: action.actionClass, outcome, detail: boundedDetail });
       return recordAuditEvent(
         env,
-        buildAgentActionAudit({ actionClass: action.actionClass, autonomyLevel, mode, outcome: auditOutcome, repoFullName: ctx.repoFullName, targetKey, actor: AGENT_ACTOR, reason: boundedDetail }),
+        buildAgentActionAudit({ actionClass: action.actionClass, autonomyLevel, mode, outcome: auditOutcome, repoFullName: ctx.repoFullName, targetKey, actor: AGENT_ACTOR, reason: boundedDetail, closeReasons: closeReasonsForAudit(action) }),
       );
     };
 
@@ -548,11 +558,11 @@ export async function executeIssueMaintenanceActions(env: Env, ctx: IssueActionE
       // merge_blocked path below, db/repositories.ts's mergeBlockedReason) -- a heuristic close's reason is
       // built by joining every blocker title, so a PR with many blockers could otherwise write an arbitrarily
       // large, un-truncated string into audit_events.detail (#terminal-outcome-audit).
-      const boundedDetail = detail.length > AUDIT_REASON_MAX_LENGTH ? `${detail.slice(0, AUDIT_REASON_MAX_LENGTH)}…` : detail;
+      const boundedDetail = boundAuditReason(detail);
       outcomes.push({ actionClass: action.actionClass, outcome, detail: boundedDetail });
       return recordAuditEvent(
         env,
-        buildAgentActionAudit({ actionClass: action.actionClass, autonomyLevel, mode, outcome: auditOutcome, repoFullName: ctx.repoFullName, targetKey, actor: AGENT_ACTOR, reason: boundedDetail }),
+        buildAgentActionAudit({ actionClass: action.actionClass, autonomyLevel, mode, outcome: auditOutcome, repoFullName: ctx.repoFullName, targetKey, actor: AGENT_ACTOR, reason: boundedDetail, closeReasons: closeReasonsForAudit(action) }),
       );
     };
 
@@ -701,6 +711,7 @@ export function actionParams(action: PlannedAgentAction): AgentPendingActionPara
     ...(action.reviewBody !== undefined ? { reviewBody: action.reviewBody } : {}),
     ...(action.mergeMethod !== undefined ? { mergeMethod: action.mergeMethod } : {}),
     ...(action.closeComment !== undefined ? { closeComment: action.closeComment } : {}),
+    ...(action.closeReasons !== undefined ? { closeReasons: action.closeReasons } : {}),
     ...(action.expectedHeadSha !== undefined ? { expectedHeadSha: action.expectedHeadSha } : {}),
     ...(action.dismissStaleApproval !== undefined ? { dismissStaleApproval: action.dismissStaleApproval } : {}),
     // Round-trip closeKind so a staged close's kind survives to accept-time — without it, the close-precision
diff --git a/src/settings/agent-actions.ts b/src/settings/agent-actions.ts
index 6d7d780be0..faeec09a16 100644
--- a/src/settings/agent-actions.ts
+++ b/src/settings/agent-actions.ts
@@ -78,6 +78,10 @@ export type PlannedAgentAction = {
   reviewBody?: string;
   mergeMethod?: AutoMergeMethod;
   closeComment?: string;
+  // For a `close` action: the individual reasons that justified closure. `reason` stays as the flat,
+  // human-readable summary for legacy callers/public notification text; this structured list is persisted into
+  // audit metadata so a close caused by multiple independent signals remains historically reconstructable.
+  closeReasons?: string[];
   // For a `close` action: WHICH kind of close this is, so the close-precision circuit-breaker can scope itself.
   // "linked-issue-hard-rule" = the DETERMINISTIC flag-then-close state machine (zero hallucination risk — and on
   // the verify path it posts a comment PROMISING closure); "heuristic" = a verdict-driven close (gate-verdict /
@@ -493,6 +497,7 @@ export function planAgentMaintenanceActions(input: AgentActionPlanInput): Planne
         actionClass: "close",
         requiresApproval: approval("close"),
         reason: "blacklisted contributor",
+        closeReasons: ["blacklisted contributor"],
         closeComment: sanitizePublicComment(blacklistCloseMessage()),
         closeKind: "blacklist",
         // Pin like merge/approve (#2452): for an auto_with_approval stage this travels into the pending row so
@@ -519,6 +524,7 @@ export function planAgentMaintenanceActions(input: AgentActionPlanInput): Planne
         actionClass: "close",
         requiresApproval: approval("close"),
         reason: "over the per-contributor open-item cap",
+        closeReasons: ["over the per-contributor open-item cap"],
         closeComment: sanitizePublicComment(contributorCapCloseMessage(authorLogin, openCount, cap, itemKind, scope)),
         closeKind: "contributor_cap",
       });
@@ -543,6 +549,7 @@ export function planAgentMaintenanceActions(input: AgentActionPlanInput): Planne
         actionClass: "close",
         requiresApproval: approval("close"),
         reason: "review-nag cooldown",
+        closeReasons: ["review-nag cooldown"],
         closeComment: sanitizePublicComment(reviewNagCloseMessage(authorLogin, pingCount, maxPings)),
         closeKind: "review_nag",
         ...(input.pr.headSha ? { expectedHeadSha: input.pr.headSha } : {}),
@@ -826,6 +833,7 @@ export function planAgentMaintenanceActions(input: AgentActionPlanInput): Planne
       actionClass: "close",
       requiresApproval: approval("close"),
       reason,
+      closeReasons: [reason],
       closeComment: closeMessage([reason]),
       closeKind: "linked-issue-hard-rule",
       // Pin like merge/approve (#2452): lets the accept-time supersede check detect a force-push after staging.
@@ -860,6 +868,7 @@ export function planAgentMaintenanceActions(input: AgentActionPlanInput): Planne
       actionClass: "close",
       requiresApproval: approval("close"),
       reason: closeReasons.join("; "),
+      closeReasons,
       closeComment: closeMessage(closeReasons),
       closeKind: "heuristic",
       closeConcreteEvidence: hasConcreteCloseEvidence(input, ciFailed, isConflict),
diff --git a/src/settings/agent-execution.ts b/src/settings/agent-execution.ts
index 8eb2875b2c..ded6485f1a 100644
--- a/src/settings/agent-execution.ts
+++ b/src/settings/agent-execution.ts
@@ -55,7 +55,9 @@ export function buildAgentActionAudit(input: {
   targetKey?: string | null | undefined;
   actor?: string | null | undefined;
   reason?: string | null | undefined;
+  closeReasons?: readonly string[] | null | undefined;
 }): AuditEventRecord {
+  const closeReasons = input.actionClass === "close" && input.closeReasons?.length ? [...input.closeReasons] : null;
   return {
     eventType: `agent.action.${input.actionClass}`,
     actor: input.actor ?? null,
@@ -67,6 +69,7 @@ export function buildAgentActionAudit(input: {
       actionClass: input.actionClass,
       autonomyLevel: input.autonomyLevel,
       mode: input.mode,
+      ...(closeReasons ? { closeReasons, closeReasonCount: closeReasons.length } : {}),
     },
   };
 }
diff --git a/src/types.ts b/src/types.ts
index 71ae850d08..cb6ca4dbe4 100644
--- a/src/types.ts
+++ b/src/types.ts
@@ -997,6 +997,9 @@ export type AgentPendingActionParams = {
   reviewBody?: string;
   mergeMethod?: AutoMergeMethod;
   closeComment?: string;
+  // Individual close reasons, persisted for approval-queue replay so the eventual audit row keeps the structured
+  // reason list rather than only the flattened `reason` field.
+  closeReasons?: string[];
   // Which kind of close this is (see PlannedAgentAction.closeKind), persisted so it round-trips through staging:
   // the close-precision circuit-breaker still scopes itself correctly when a staged close is later accepted
   // (#2127), and the actuation-time live-CI re-check (#2364) — which only applies to a heuristic close — still
diff --git a/test/unit/agent-action-executor.test.ts b/test/unit/agent-action-executor.test.ts
index 1045a5b28e..1ad862dabd 100644
--- a/test/unit/agent-action-executor.test.ts
+++ b/test/unit/agent-action-executor.test.ts
@@ -115,6 +115,22 @@ describe("executeAgentMaintenanceActions (#778 gate stack)", () => {
     expect(replayed).toMatchObject({ actionClass: "label", autonomyClass: "close", requiresApproval: false, reason: "blacklisted contributor", label: "slop", labelOp: "add" });
   });
 
+  it("actionParams round-trips structured closeReasons so approval replay preserves every close cause", () => {
+    const closeWithReasons: PlannedAgentAction = {
+      actionClass: "close",
+      requiresApproval: true,
+      reason: "CI failed; blocker",
+      closeComment: "closing",
+      closeReasons: ["CI failed", "blocker"],
+    };
+
+    const persisted = actionParams(closeWithReasons);
+    const replayed = pendingActionToPlanned({ actionClass: "close", params: persisted, reason: closeWithReasons.reason });
+
+    expect(persisted).toEqual({ closeComment: "closing", closeReasons: ["CI failed", "blocker"] });
+    expect(replayed).toMatchObject({ actionClass: "close", requiresApproval: false, reason: "CI failed; blocker", closeComment: "closing", closeReasons: ["CI failed", "blocker"] });
+  });
+
   it("LIVE: executes each action class via its GitHub primitive and audits completed", async () => {
     const env = createTestEnv({});
     const outcomes = await executeAgentMaintenanceActions(env, ctx(), [label, requestChanges, approve, merge, close, updateBranch]);
@@ -145,9 +161,12 @@ describe("executeAgentMaintenanceActions (#778 gate stack)", () => {
     expect(outcomes[0]?.outcome).toBe("completed");
     expect(outcomes[0]?.detail.length).toBe(281); // 280 chars + the "…" truncation marker
     expect(outcomes[0]?.detail.endsWith("…")).toBe(true);
-    const audit = await (env.DB.prepare("select detail from audit_events where event_type = 'agent.action.close' order by created_at desc limit 1").first<{ detail: string }>());
+    const audit = await env.DB.prepare("select detail, metadata_json from audit_events where event_type = 'agent.action.close' order by created_at desc limit 1").first<{ detail: string; metadata_json: string }>();
     expect(audit?.detail).toBe(outcomes[0]?.detail);
     expect(audit?.detail.length).toBeLessThan(longReason.length);
+    const metadata = JSON.parse(audit?.metadata_json ?? "{}");
+    expect(metadata.closeReasons).toEqual([outcomes[0]?.detail]);
+    expect(metadata.closeReasonCount).toBe(1);
   });
 
   it("does NOT truncate a reason at or under the bound (no stray truncation marker on ordinary-length reasons)", async () => {
@@ -157,6 +176,48 @@ describe("executeAgentMaintenanceActions (#778 gate stack)", () => {
     expect(outcomes[0]?.detail.endsWith("…")).toBe(false);
   });
 
+  it("records every structured close reason in audit metadata instead of only the flattened detail", async () => {
+    const env = createTestEnv({});
+    const closeWithReasons: PlannedAgentAction = {
+      actionClass: "close",
+      requiresApproval: false,
+      reason: "CI is failing (codecov/patch); review blocker; base conflict",
+      closeComment: "closing",
+      closeReasons: ["CI is failing (codecov/patch)", "review blocker", "base conflict"],
+    };
+
+    const outcomes = await executeAgentMaintenanceActions(env, ctx(), [closeWithReasons]);
+    expect(outcomes[0]?.outcome).toBe("completed");
+
+    const audit = await auditFor(env, "close");
+    const metadata = JSON.parse(audit?.metadata_json ?? "{}");
+    expect(metadata.closeReasons).toEqual(["CI is failing (codecov/patch)", "review blocker", "base conflict"]);
+    expect(metadata.closeReasonCount).toBe(3);
+    expect(outcomes[0]?.detail).toBe(closeWithReasons.reason);
+  });
+
+  it("bounds explicit structured close reasons before storing them in audit metadata", async () => {
+    const env = createTestEnv({});
+    const longReason = "review blocker: ".repeat(30);
+    const closeWithLongStructuredReason: PlannedAgentAction = {
+      actionClass: "close",
+      requiresApproval: false,
+      reason: "combined close reason",
+      closeComment: "closing",
+      closeReasons: ["short reason", longReason],
+    };
+
+    const outcomes = await executeAgentMaintenanceActions(env, ctx(), [closeWithLongStructuredReason]);
+    expect(outcomes[0]?.outcome).toBe("completed");
+
+    const audit = await auditFor(env, "close");
+    const metadata = JSON.parse(audit?.metadata_json ?? "{}");
+    expect(metadata.closeReasons[0]).toBe("short reason");
+    expect(metadata.closeReasons[1]).toHaveLength(281);
+    expect(metadata.closeReasons[1].endsWith("…")).toBe(true);
+    expect(metadata.closeReasons[1].length).toBeLessThan(longReason.length);
+  });
+
   it("#label-scoping: a label action's autonomyClass (not the literal actionClass) governs the durable re-check", async () => {
     const env = createTestEnv({});
     // autonomy.label is OFF; autonomy.close is ON — a label authorized via autonomyClass: "close" must still
@@ -1201,9 +1262,12 @@ describe("executeIssueMaintenanceActions (#2270 issue-side actuation)", () => {
     expect(outcomes[0]?.outcome).toBe("completed");
     expect(outcomes[0]?.detail.length).toBe(281); // 280 chars + the "…" truncation marker
     expect(outcomes[0]?.detail.endsWith("…")).toBe(true);
-    const audit = await env.DB.prepare("select detail from audit_events where event_type = 'agent.action.close' order by created_at desc limit 1").first<{ detail: string }>();
+    const audit = await env.DB.prepare("select detail, metadata_json from audit_events where event_type = 'agent.action.close' order by created_at desc limit 1").first<{ detail: string; metadata_json: string }>();
     expect(audit?.detail).toBe(outcomes[0]?.detail);
     expect(audit?.detail.length).toBeLessThan(longReason.length);
+    const metadata = JSON.parse(audit?.metadata_json ?? "{}");
+    expect(metadata.closeReasons).toEqual([outcomes[0]?.detail]);
+    expect(metadata.closeReasonCount).toBe(1);
   });
 
   it("does NOT truncate a reason at or under the bound (no stray truncation marker on ordinary-length reasons)", async () => {
diff --git a/test/unit/agent-actions.test.ts b/test/unit/agent-actions.test.ts
index 8ad5199fd4..f92d0b08ef 100644
--- a/test/unit/agent-actions.test.ts
+++ b/test/unit/agent-actions.test.ts
@@ -277,6 +277,31 @@ describe("planAgentMaintenanceActions (#778)", () => {
     expect(winnerClose.reason).not.toContain("duplicate of another open PR");
   });
 
+  it("keeps every close cause as a structured closeReasons list for historical audit accuracy", () => {
+    const plan = planAgentMaintenanceActions(
+      input({
+        conclusion: "failure",
+        autonomy: { close: "auto" },
+        blockerTitles: ["AI reviewers found a blocker", "Security scanner found a blocker"],
+        ciState: "failed",
+        failingCheckNames: ["codecov/patch", "validate"],
+        pr: { labels: [], linkedDuplicateCount: 2, mergeableState: "dirty", slopRisk: 80 },
+      }),
+    );
+    const close = plan.find((a) => a.actionClass === "close");
+
+    expect(close?.closeReasons).toEqual([
+      "CI is failing (codecov/patch, validate)",
+      "conflicts with the base branch — resolve and open a fresh PR",
+      "AI reviewers found a blocker",
+      "Security scanner found a blocker",
+      "slop score 80 ≥ 60",
+      "duplicate of another open PR",
+    ]);
+    expect(close?.reason).toBe(close?.closeReasons?.join("; "));
+    for (const reason of close?.closeReasons ?? []) expect(close?.closeComment).toContain(reason);
+  });
+
   it("never plans both merge and close", () => {
     const plan = planAgentMaintenanceActions(input({ conclusion: "success", autonomy: { merge: "auto", close: "auto" }, pr: { labels: [], mergeableState: "clean", reviewDecision: "APPROVED", slopRisk: 95 } }));
     const cls = classes(plan);
@@ -715,6 +740,7 @@ describe("planAgentMaintenanceActions (#778)", () => {
       const close = planAgentMaintenanceActions(input({ conclusion: "success", autonomy: { close: "auto" }, ciState: "passed", linkedIssueHardRule: violation, pr: { labels: [], mergeableState: "clean", reviewDecision: "APPROVED" } })).find((a) => a.actionClass === "close");
       expect(close).toBeTruthy();
       expect(close?.reason).toBe(violation.reason);
+      expect(close?.closeReasons).toEqual([violation.reason]);
       // the cited reason is surfaced in the close comment too
       expect(close?.closeComment).toContain(violation.reason);
     });
@@ -1160,6 +1186,7 @@ describe("contributor blacklist short-circuit (#1425)", () => {
     // guard always has the close's outcome already recorded by the time it evaluates the label.
     expect(classes(plan)).toEqual(["close", "label"]); // short-circuit: no approve/merge despite a SUCCESS gate
     expect(plan[0]).toMatchObject({ actionClass: "close", closeKind: "blacklist" });
+    expect(plan[0]?.closeReasons).toEqual(["blacklisted contributor"]);
     expect(plan[1]).toMatchObject({ actionClass: "label", label: DEFAULT_BLACKLIST_LABEL, labelOp: "add", closeKind: "blacklist" });
     expect(plan[0]?.closeComment).not.toContain("plagiarism");
     expect(plan[0]?.closeComment).toContain("blocked from contributing");
@@ -1243,6 +1270,7 @@ describe("per-contributor open-item cap short-circuit (#2270)", () => {
     // close is pushed BEFORE its coupled label (#label-close-split-brain) — see the blacklist section above.
     expect(classes(plan)).toEqual(["close", "label"]); // short-circuit: no approve/merge despite a SUCCESS gate
     expect(plan[0]).toMatchObject({ actionClass: "close", closeKind: "contributor_cap" });
+    expect(plan[0]?.closeReasons).toEqual(["over the per-contributor open-item cap"]);
     expect(plan[1]).toMatchObject({ actionClass: "label", label: DEFAULT_CONTRIBUTOR_CAP_LABEL, labelOp: "add", closeKind: "contributor_cap" });
   });
 
@@ -1337,6 +1365,7 @@ describe("review-nag cooldown short-circuit (#2463)", () => {
     // close is pushed BEFORE its coupled label (#label-close-split-brain) — see the blacklist section above.
     expect(classes(plan)).toEqual(["close", "label"]); // short-circuit: no approve/merge despite a SUCCESS gate
     expect(plan[0]).toMatchObject({ actionClass: "close", closeKind: "review_nag" });
+    expect(plan[0]?.closeReasons).toEqual(["review-nag cooldown"]);
     expect(plan[1]).toMatchObject({ actionClass: "label", label: DEFAULT_REVIEW_NAG_LABEL, labelOp: "add", closeKind: "review_nag" });
     expect(plan[0]?.closeComment).toContain("chatty-contributor");
     expect(plan[0]?.closeComment).toContain("4");
diff --git a/test/unit/agent-approval-queue.test.ts b/test/unit/agent-approval-queue.test.ts
index b97d88c07d..5ceba0297c 100644
--- a/test/unit/agent-approval-queue.test.ts
+++ b/test/unit/agent-approval-queue.test.ts
@@ -1211,7 +1211,7 @@ describe("agent approval queue (#779)", () => {
     expect(actionParams({ actionClass: "label", autonomyClass: "review_state_label", requiresApproval: false, reason: "x", label: "L" })).toEqual({ autonomyClass: "review_state_label", label: "L" });
     expect(actionParams({ actionClass: "request_changes", requiresApproval: false, reason: "x", reviewBody: "B" })).toEqual({ reviewBody: "B" });
     expect(actionParams({ actionClass: "merge", requiresApproval: false, reason: "x", mergeMethod: "rebase" })).toEqual({ mergeMethod: "rebase" });
-    expect(actionParams({ actionClass: "close", requiresApproval: false, reason: "x", closeComment: "C" })).toEqual({ closeComment: "C" });
+    expect(actionParams({ actionClass: "close", requiresApproval: false, reason: "x", closeComment: "C", closeReasons: ["x"] })).toEqual({ closeComment: "C", closeReasons: ["x"] });
     // closeKind must round-trip through staging — without it the close-precision breaker could never match a
     // staged close as heuristic on accept (#2127).
     expect(actionParams({ actionClass: "close", requiresApproval: false, reason: "x", closeComment: "C", closeKind: "heuristic", closeRequiresCiState: "failed" })).toEqual({ closeComment: "C", closeKind: "heuristic", closeRequiresCiState: "failed" });
@@ -1232,6 +1232,12 @@ describe("agent approval queue (#779)", () => {
   it("pendingActionToPlanned clears requiresApproval and defaults the reason", () => {
     expect(pendingActionToPlanned({ actionClass: "merge", params: { mergeMethod: "squash" } })).toMatchObject({ actionClass: "merge", requiresApproval: false, reason: "maintainer-approved", mergeMethod: "squash" });
     expect(pendingActionToPlanned({ actionClass: "label", params: { label: "L" }, reason: "explicit" }).reason).toBe("explicit");
+    expect(pendingActionToPlanned({ actionClass: "close", params: { closeReasons: ["ci failed", "blocker"] }, reason: "ci failed; blocker" })).toMatchObject({
+      actionClass: "close",
+      requiresApproval: false,
+      reason: "ci failed; blocker",
+      closeReasons: ["ci failed", "blocker"],
+    });
   });
 
   it("countPendingAgentActions respects both the repo filter and the status filter", async () => {
diff --git a/test/unit/agent-execution.test.ts b/test/unit/agent-execution.test.ts
index 5bb25bf2ba..531ad7fa1c 100644
--- a/test/unit/agent-execution.test.ts
+++ b/test/unit/agent-execution.test.ts
@@ -74,6 +74,52 @@ describe("buildAgentActionAudit", () => {
     expect(audit.actor).toBeNull();
     expect(audit.detail).toBeNull();
   });
+
+  it("records structured close reasons only for close-action audit metadata", () => {
+    const closeAudit = buildAgentActionAudit({
+      actionClass: "close",
+      autonomyLevel: "auto",
+      mode: "live",
+      outcome: "completed",
+      repoFullName: "owner/repo",
+      reason: "ci failed; blocker",
+      closeReasons: ["ci failed", "blocker"],
+    });
+    expect(closeAudit.metadata).toMatchObject({ closeReasons: ["ci failed", "blocker"], closeReasonCount: 2 });
+
+    const mergeAudit = buildAgentActionAudit({
+      actionClass: "merge",
+      autonomyLevel: "auto",
+      mode: "live",
+      outcome: "completed",
+      repoFullName: "owner/repo",
+      reason: "clean",
+      closeReasons: ["must not attach"],
+    });
+    expect(mergeAudit.metadata).not.toHaveProperty("closeReasons");
+    expect(mergeAudit.metadata).not.toHaveProperty("closeReasonCount");
+
+    const legacyCloseAudit = buildAgentActionAudit({
+      actionClass: "close",
+      autonomyLevel: "auto",
+      mode: "live",
+      outcome: "completed",
+      repoFullName: "owner/repo",
+      reason: "legacy flattened reason",
+    });
+    expect(legacyCloseAudit.metadata).not.toHaveProperty("closeReasons");
+
+    const emptyCloseAudit = buildAgentActionAudit({
+      actionClass: "close",
+      autonomyLevel: "auto",
+      mode: "live",
+      outcome: "completed",
+      repoFullName: "owner/repo",
+      reason: "empty reason list",
+      closeReasons: [],
+    });
+    expect(emptyCloseAudit.metadata).not.toHaveProperty("closeReasons");
+  });
 });
 
 describe("agent write-permission readiness (#775)", () => {