Delete repetition detection outright (CL-6995)

CHANGELOG.md

@@ -35,6 +35,15 @@ parallel copies under `docs/` or `scripts/notes/`. At cut time: rename
   independent copies of the same `AbortController` + `setTimeout` race.
 - `runtime-bridge.ts` now re-exports `mapReactorLike` from `stream-event-map.ts`
   instead of wrapping it in an identical local function.
+- Removed degenerate-repetition detection outright: the streamed-text loop
+  detector, the tool-fingerprint period/cycle thrash check, the
+  turns-since-user-message backstop, and the leaf no-progress (identical
+  tool-call) counter. These were pattern-matching heuristics layered on top
+  of the transport/policy line the harness actually needs — provider stream
+  error handling, connection retry/backoff, and the turn budget — and had
+  become a source of false-positive stalls without a clear win rate. The
+  turn budget stop and the director's soft tool-only check-in nudge are
+  unchanged; nothing else in this run/stop chain was touched.
 
 ## [0.2.108] - 2026-08-24
 

src/agent/model-family-policy.ts

@@ -12,10 +12,8 @@ export interface ModelFamilyPolicy {
    * Consecutive tool-only assistant turns (tool calls, no text) before the
    * main chat director injects a one-shot wrap-up nudge. A long tool-only
    * streak is normal orchestration (Linear lookups, code reads, etc.) and
-   * must not by itself stop the session — this is a soft check-in, not a
-   * loop-protection trigger. The real stop signal is a repeating cycle in
-   * the tool-fingerprint history, independent of this threshold — see
-   * detectToolFingerprintThrash in subagent/stop-policy.ts.
+   * must not by itself stop the session — this is a soft check-in, and it
+   * never escalates to a pause on its own.
    */
   toolOnlyTurnNudgeAt: number;
   /** Ephemeral nudge text injected at toolOnlyTurnNudgeAt. */

src/agent/prompts.ts

@@ -153,7 +153,7 @@ export function buildGuidelines(
           "- Prefer the typed spawn contract on every worker: `intent`, `success_criteria` (done-when), `do_not` (scope fence), and `report_focus` so workers finish instead of thrashing. Free-form `prompt` alone is weaker.",
           "- After workers return, merge their Summary/Findings into a coherent answer for the operator; do not paste raw sub-agent dumps.",
           "- Pass `maxTurns` on `task` when a job needs a bounded inference budget (unset is unbounded). On turn-budget salvage, re-dispatch with continuation context and a higher maxTurns only a few times on the same brief — after the re-dispatch cap, change approach instead of bumping turns again.",
-          "- After thrash / no-progress / repetition / never-acted salvage, do not re-dispatch an identical brief (prompt/agent/intent/success_criteria/do_not) — it is refused. Change the brief to force a re-run; maxTurns alone does not unlock it.",
+          "- After a thrash / no-ship / never-acted / never-edited salvage, do not re-dispatch an identical brief (prompt/agent/intent/success_criteria/do_not) — it is refused. Change the brief to force a re-run; maxTurns alone does not unlock it.",
           "- Use manage_tasks for your own coordination checklist; spawning workers is `task`, not manage_tasks.",
           "- If context is compacted automatically, do not stop tasks early due to token fear; persist progress via manage_tasks and worker reports.",
         ]),

src/session/stream-journal.test.ts

@@ -70,11 +70,11 @@ describe("createCycleTextRecorder", () => {
   test("flush writes the buffer with a reason and resets", async () => {
     const recorder = createCycleTextRecorder(() => dir);
     recorder.handleEvent(delta("looping output"));
-    await recorder.flush("repetition");
+    await recorder.flush("cancelled");
 
     const records = await readPartialRecords();
     expect(records).toHaveLength(1);
-    expect(records[0]?.reason).toBe("repetition");
+    expect(records[0]?.reason).toBe("cancelled");
     expect(records[0]?.text).toBe("looping output");
     expect(recorder.text()).toBe("");
   });
@@ -177,7 +177,7 @@ describe("createCycleTextRecorder", () => {
     expect(recorder.text()).toBe("visible reply");
     expect(recorder.thinkingText()).toBe("0/1 1/2 2/3 ");
 
-    await recorder.flush("repetition");
+    await recorder.flush("cancelled");
     const records = await readPartialRecords();
     expect(records[0]?.text).toBe("visible reply");
     expect(records[0]?.thinkingText).toBe("0/1 1/2 2/3 ");
@@ -189,11 +189,11 @@ describe("createCycleTextRecorder", () => {
     // must still be diagnosable from thinkingText alone.
     const recorder = createCycleTextRecorder(() => dir);
     recorder.handleEvent(thinkingDelta("0/1 1/2 2/3 3/4 4/5 "));
-    const snapshot = await recorder.dispose("repetition");
+    const snapshot = await recorder.dispose("cancelled");
 
     expect(snapshot).toBe("");
     const records = await readPartialRecords();
-    expect(records[0]?.reason).toBe("repetition");
+    expect(records[0]?.reason).toBe("cancelled");
     expect(records[0]?.text).toBe("");
     expect(records[0]?.thinkingText).toBe("0/1 1/2 2/3 3/4 4/5 ");
   });

src/session/stream-journal.ts

@@ -33,7 +33,6 @@ export function appendCycleText(
 }
 
 export type PartialFlushReason =
-  | "repetition"
   | "deadline"
   | "cancelled"
   | "interrupted"

src/session/summarizer.ts

@@ -12,7 +12,6 @@ import { createDefaultDependencies } from "@intx/inference/providers";
 import { getLogger } from "@intx/log";
 import type { ConversationTurn, InferenceSource } from "@intx/types/runtime";
 import { LOG_NAMESPACE_ROOT } from "../branding.js";
-import { detectRepetition } from "../subagent/repetition.js";
 import { buildTurnSummary } from "./compactor.js";
 
 const logger = getLogger([LOG_NAMESPACE_ROOT, "session", "summarizer"]);
@@ -75,13 +74,7 @@ export function condenseTurns(turns: ConversationTurn[]): string {
         if (turn.role === "user") {
           userMessages.push(block.text.slice(0, 400));
         } else if (turn.role === "assistant" && block.text.length > 0) {
-          // Compaction often fires mid-degeneration, when the tail of the
-          // history is the model looping one phrase. Seeding the summary from
-          // those turns hands the looped text to the summarizer verbatim, so
-          // repetition-flagged turns are dropped from the excerpt entirely.
-          if (detectRepetition(block.text) === null) {
-            assistantSnippets.push(block.text.slice(0, 300));
-          }
+          assistantSnippets.push(block.text.slice(0, 300));
         }
       }
       if (block.type === "tool_call") {

src/subagent/brief-dispatch.ts

@@ -15,8 +15,7 @@ import type { TaskIntent } from "./report.js";
 import type { ForcedStopReason } from "./stop-policy.js";
 
 /** Salvage classes that must not be re-dispatched with an identical brief. */
-export type HardBlockSalvage =
-  "no-ship" | "no-progress" | "repetition" | "never-acted" | "never-edited";
+export type HardBlockSalvage = "no-ship" | "never-acted" | "never-edited";
 
 // Every forced-stop reason a leaf can report maps 1:1 onto a salvage kind
 // the parent ledger cares about.
@@ -44,13 +43,7 @@ export interface BriefDispatchRecord {
  */
 export const TURN_BUDGET_STOP_AFTER_DISPATCHES = 3;
 
-const HARD_BLOCK_SALVAGES = new Set<BriefSalvageKind>([
-  "no-ship",
-  "no-progress",
-  "repetition",
-  "never-acted",
-  "never-edited",
-]);
+const HARD_BLOCK_SALVAGES = new Set<BriefSalvageKind>(["no-ship", "never-acted", "never-edited"]);
 
 export function isHardBlockSalvage(kind: BriefSalvageKind): kind is HardBlockSalvage {
   return HARD_BLOCK_SALVAGES.has(kind);