opencode-acp 1.14.3 → 1.14.4

package/README.md

@@ -528,6 +528,14 @@ For the complete list with root cause analysis, see the [bug tracker](https://gi
 
 ## Changelog
 
+### v1.14.4 — Tier Detection + E2E Tests + Debug Notification + Nudge Loop Fix (PRs #215, #214, #217, #218)
+
+**Problem**: Four issues accumulated since v1.14.3. (1) **Tier misclassification** (PR #215): `applyCompressionState` determined the compression tier from `consumedBlockIds` — any consumed block bumped the tier up. This misclassified T1 compressions that incidentally overlapped existing T1 blocks as T2 (6 of 18 T2-labeled blocks in a real session were misclassified). (2) **E2E test gaps** (PR #214): E2E tests disabled all protection (`preserveRecentMessages: 0`), CI ran only 4/6 scenarios, and `verify.ts` only checked `blockCount` — the protection mechanism fixed 3× in v1.14.x was completely untested in E2E. (3) **Debug notifications invisible** (PR #217): With `debug: true`, compression notifications went to toast only — users couldn't see them in the chat session for debugging. (4) **Nudge injection loop** (PR #218, issue #216): `applyAnchoredNudges` fired before `nothingToCompress` was computed, so the nudge text ("compress now") was injected even when there was nothing to compress — the model saw a nudge with an empty recommendation list, tried random compressions, failed, and looped. Additionally, `messageHasCompress` only recognized `status === "completed"`, so failed compressions never reset the halved nudge threshold, keeping the loop alive.
+
+**Fix**: (1) PR #215 — Tier is now determined by `selection.startReference.kind` / `endReference.kind`: message boundaries → T1, block boundaries → T2+ (max consumed tier + 1). When T1 consumes an old T1 block, the old block is still deactivated (superseded), but the new block correctly gets `tier=1`. 7 new regression tests. (2) PR #214 — Added per-scenario config override in `run-e2e.sh`, deepened `verify.ts` with `compressedCount`/`minCompressedCount`/`maxCompressedCount` checks, added 2 new scenarios (07: protection-filtered, 08: nudge-with-protection) using production config (`preserveRecentMessages: 5`, `preserveLastUserMessage: true`), and added scenarios 05/06/07/08 to CI (was 4, now 8). (3) PR #217 — When `config.debug` is on, `sendCompressNotification` now calls `sendIgnoredMessage()` to inject the notification into the chat session (user-visible, model-invisible via `ignored: true`) in addition to the toast. `dropEmptyMessages` (FIX #20) strips ignored-only messages before the next LLM call as defense-in-depth. Debug OFF: toast only (unchanged). (4) PR #218 — Moved `applyAnchoredNudges` after `shouldInject` computation and gated it by `if (shouldInject)`, so no nudge text is injected when there's nothing to compress. Added `messageHasCompressAttempt` (recognizes any compress call regardless of status) and uses it for the early-return / nudge-state clearing, while baseline adjustment still requires `messageHasCompress` (completed only). Failed compressions now clear the halved threshold without falsely adjusting the baseline.
+
+Files: `lib/compress/state.ts` (PR #215), `scripts/e2e/{run-e2e.sh,verify.ts}` + `scripts/e2e/scenarios/{07-protection-filtered,08-nudge-with-protection}.json` + `.github/workflows/ci.yml` (PR #214), `lib/ui/notification.ts` (PR #217), `lib/messages/{query,inject/inject}.ts` (PR #218). Tests: `tests/e2e-tier-compression.test.ts`, `tests/query-pure.test.ts`, `tests/inject.test.ts`. 936 tests pass.
+
 ### v1.14.3 — Soften Protected Zone + Reduce Defaults (PR #212)
 
 **Problem**: `checkProtectedRange` hard-rejected any compress call covering protected recent messages (last N messages + last N tokens). The model got an error and had to retry with a different range or use `dangerous: true`. Additionally, the default `preserveRecentMessages: 20` and `preserveRecentTokens: 20000` (≈40 messages) were too aggressive — protecting nearly half the conversation in autonomous sessions.

package/README.zh-CN.md

@@ -482,6 +482,14 @@ ACP 在首次启动时自动将配置从 `dcp.jsonc` 迁移到 `acp.jsonc`，将
 
 ## 更新日志
 
+### v1.14.4 — Tier 检测 + E2E 测试 + Debug 通知 + Nudge 循环修复（PRs #215, #214, #217, #218）
+
+**问题**：自 v1.14.3 以来积累的 4 个问题。(1) **Tier 误分类**（PR #215）：`applyCompressionState` 从 `consumedBlockIds` 判断压缩层级——只要消费了任何已有 block 就会提升层级。这导致那些恰好覆盖了已有 T1 block 的 T1 压缩被错误分类为 T2（在一个真实 session 中，18 个标记为 T2 的 block 里有 6 个被误分类）。(2) **E2E 测试缺口**（PR #214）：E2E 测试禁用了所有保护（`preserveRecentMessages: 0`），CI 只运行 4/6 个场景，`verify.ts` 只检查 `blockCount`——v1.14.x 修了 3 次的保护机制在 E2E 里完全没有覆盖。(3) **Debug 通知不可见**（PR #217）：开启 `debug: true` 时，压缩通知只发到 toast——用户无法在 chat session 里看到用于调试的通知。(4) **Nudge 注入循环**（PR #218，issue #216）：`applyAnchoredNudges` 在 `nothingToCompress` 计算之前就触发，导致即使没有可压缩内容，nudge 文本（"compress now"）也被注入——模型看到 nudge 但推荐列表为空，尝试随机压缩、失败、循环。此外 `messageHasCompress` 只识别 `status === "completed"`，失败的压缩不会重置被减半的 nudge 阈值，循环持续。
+
+**修复**：(1) PR #215——层级改由 `selection.startReference.kind` / `endReference.kind` 判断：消息边界 → T1，block 边界 → T2+（消费的最高层级 + 1）。T1 消费旧 T1 block 时，旧 block 仍会被停用（被取代），但新 block 正确得到 `tier=1`。新增 7 个回归测试。(2) PR #214——在 `run-e2e.sh` 中增加按场景的配置覆盖，`verify.ts` 增加 `compressedCount`/`minCompressedCount`/`maxCompressedCount` 检查，新增 2 个场景（07: protection-filtered、08: nudge-with-protection）使用生产配置（`preserveRecentMessages: 5`、`preserveLastUserMessage: true`），CI 场景从 4 个增加到 8 个。(3) PR #217——当 `config.debug` 开启时，`sendCompressNotification` 在 toast 之外额外调用 `sendIgnoredMessage()` 把通知注入到 chat session（用户可见、模型不可见，通过 `ignored: true`）。`dropEmptyMessages`（FIX #20）作为纵深防御在下次 LLM 调用前剥离 ignored-only 消息。Debug 关闭：仅 toast（不变）。(4) PR #218——把 `applyAnchoredNudges` 移到 `shouldInject` 计算之后并用 `if (shouldInject)` 守卫，没有可压缩内容时不注入 nudge 文本。新增 `messageHasCompressAttempt`（识别任何状态的 compress 调用）用于提前返回 / nudge 状态清理，而 baseline 调整仍要求 `messageHasCompress`（仅 completed）。失败的压缩现在会清除被减半的阈值，但不会错误地调整 baseline。
+
+文件：`lib/compress/state.ts`（PR #215），`scripts/e2e/{run-e2e.sh,verify.ts}` + `scripts/e2e/scenarios/{07-protection-filtered,08-nudge-with-protection}.json` + `.github/workflows/ci.yml`（PR #214），`lib/ui/notification.ts`（PR #217），`lib/messages/{query,inject/inject}.ts`（PR #218）。测试：`tests/e2e-tier-compression.test.ts`、`tests/query-pure.test.ts`、`tests/inject.test.ts`。936 个测试通过。
+
 ### v1.14.3 — 软化保护区 + 减小默认值（PR #212）
 
 **问题**：`checkProtectedRange` 硬拒绝任何覆盖保护区最近消息的压缩调用（最后 N 条消息 + 最后 N tokens）。模型收到错误后必须换范围重试或使用 `dangerous: true`。此外，默认 `preserveRecentMessages: 20` 和 `preserveRecentTokens: 20000`（约 40 条消息）过于激进 —— 在自主会话中保护了近一半的对话。

package/dist/index.js

@@ -2085,6 +2085,16 @@ var messageHasCompress = (message) => {
     (part) => part.type === "tool" && part.tool === "compress" && part.state?.status === "completed"
   );
 };
+var messageHasCompressAttempt = (message) => {
+  if (!isMessageWithInfo(message)) {
+    return false;
+  }
+  if (message.info.role !== "assistant") {
+    return false;
+  }
+  const parts = Array.isArray(message.parts) ? message.parts : [];
+  return parts.some((part) => part.type === "tool" && part.tool === "compress");
+};
 var isIgnoredUserMessage = (message) => {
   if (!isMessageWithInfo(message)) {
     return false;
@@ -3200,6 +3210,7 @@ function applyCompressionState(state, input, selection, anchorMessageId, blockId
   const messagesState = state.prune.messages;
   const consumed = [...new Set(consumedBlockIds.filter((id) => Number.isInteger(id) && id > 0))];
   const createdAt = Date.now();
+  const isBlockBoundary = selection.startReference?.kind === "compressed-block" || selection.endReference?.kind === "compressed-block";
   let minConsumedTier;
   for (const consumedBlockId of consumed) {
     const cb = messagesState.blocksById.get(consumedBlockId);
@@ -3210,9 +3221,8 @@ function applyCompressionState(state, input, selection, anchorMessageId, blockId
       }
     }
   }
-  const effectiveMinTier = minConsumedTier ?? 0;
-  const outputTier = Math.min(3, effectiveMinTier + 1);
-  const targetTierForConsumption = effectiveMinTier;
+  const outputTier = isBlockBoundary ? Math.min(3, (minConsumedTier ?? 1) + 1) : 1;
+  const targetTierForConsumption = isBlockBoundary ? minConsumedTier ?? 1 : 1;
   const effectiveMessageIds = new Set(selection.messageIds);
   const effectiveToolIds = new Set(selection.toolIds);
   for (const consumedBlockId of consumed) {
@@ -5369,6 +5379,10 @@ ${progressBar}`;
   }
   let toastMessage = message;
   toastMessage = config.pruneNotification === "minimal" ? toastMessage : truncateToastBody(toastMessage);
+  if (config.debug) {
+    const chatMessage = config.pruneNotification === "minimal" ? message : truncateToastBody(message);
+    await sendIgnoredMessage(client, sessionId, chatMessage, params, logger);
+  }
   await client.tui.showToast({
     body: {
       title: "ACP: Compress Notification",
@@ -8492,7 +8506,7 @@ var injectCompressNudges = (state, config, logger, messages, prompts, compressio
     (m) => m.info.role === "user" && !isIgnoredUserMessage(m)
   );
   const currentTurnStart = lastUserIdx >= 0 ? lastUserIdx + 1 : 0;
-  const currentTurnHasCompress = messages.slice(currentTurnStart).some((m) => m.info.role === "assistant" && messageHasCompress(m));
+  const currentTurnHasCompress = messages.slice(currentTurnStart).some((m) => m.info.role === "assistant" && messageHasCompressAttempt(m));
   if (currentTurnHasCompress) {
     const wasNudgeTriggered = state.nudges.lastNudgeShownTokens !== void 0;
     state.nudges.contextLimitAnchors.clear();
@@ -8502,7 +8516,8 @@ var injectCompressNudges = (state, config, logger, messages, prompts, compressio
     state.nudges.lastToolOutputNudgeTokens = void 0;
     state.nudges.lastTier2NudgeTokens = void 0;
     state.nudges.lastTier3NudgeTokens = void 0;
-    if (wasNudgeTriggered && !state.nudges.compressBaselineSet) {
+    const currentTurnHasSuccessfulCompress = messages.slice(currentTurnStart).some((m) => m.info.role === "assistant" && messageHasCompress(m));
+    if (currentTurnHasSuccessfulCompress && wasNudgeTriggered && !state.nudges.compressBaselineSet) {
       const baseline = state.nudges.lastPerMessageNudgeTokens;
       const postCompress = currentTokens;
       const preCompress = preCompressTokens;
@@ -8621,9 +8636,6 @@ var injectCompressNudges = (state, config, logger, messages, prompts, compressio
   const growthSinceBaseline = currentTokens !== void 0 && growthReference !== void 0 ? currentTokens - growthReference : void 0;
   const nudgeAllowed = emergencyOverride || decision.shouldNudge && growthSinceBaseline !== void 0 && growthSinceBaseline >= growthFloor;
   const effectiveTipsVariant = emergencyOverride ? "maxLimit" : decision.tipsVariant;
-  if (nudgeAllowed) {
-    applyAnchoredNudges(state, config, messages, prompts, compressionPriorities, currentTokens, modelContextLimit, suffixMessage);
-  }
   if (state.nudges.lastPerMessageNudgeTokens === void 0 && currentTokens !== void 0) {
     state.nudges.lastPerMessageNudgeTokens = currentTokens;
     baselineReEstablished = true;
@@ -8672,10 +8684,18 @@ var injectCompressNudges = (state, config, logger, messages, prompts, compressio
   const allProtected = contextRanges.compressible.length === 0 && contextRanges.protected.length > 0;
   const allInProtectedZone = protectedRefs.size > 0 && unprotectedCompressible.length === 0;
   const nothingToCompress = filterSuppressed || allProtected || allInProtectedZone;
-  let shouldInject = nudgeAllowed && (!nothingToCompress || emergencyOverride);
+  const shouldInjectNudge = nudgeAllowed && (!nothingToCompress || emergencyOverride);
+  let shouldInject = shouldInjectNudge;
   if (nudgeAllowed && nothingToCompress && !emergencyOverride) {
     state.nudges.lastNudgeShownTokens = void 0;
   }
+  if (shouldInjectNudge) {
+    applyAnchoredNudges(state, config, messages, prompts, compressionPriorities, currentTokens, modelContextLimit, suffixMessage);
+  }
+  if (state.nudges.lastPerMessageNudgeTokens === void 0 && currentTokens !== void 0) {
+    state.nudges.lastPerMessageNudgeTokens = currentTokens;
+    baselineReEstablished = true;
+  }
   if (suffixMessage && !shouldInject) {
     const tierUsage = getTierTokenUsage(state);
     const tierChecks = [
@@ -10062,7 +10082,7 @@ import { writeFile as writeFile2, mkdir as mkdir2 } from "fs/promises";
 import { join as join3 } from "path";
 import { existsSync as existsSync3 } from "fs";
 import { homedir as homedir3 } from "os";
-var LOG_VERSION = true ? "1.14.3" : "dev";
+var LOG_VERSION = true ? "1.14.4" : "dev";
 var Logger = class {
   logDir;
   enabled;