opencode-acp 1.7.0 → 1.8.0

package/README.md

@@ -419,6 +419,34 @@ For the complete list with root cause analysis, see the [bug tracker](https://gi
 
 ---
 
+## Changelog
+
+### v1.7.0 — Principle-Driven Prompts
+
+**Philosophy**: Replaced verbose context-management guidance with 4 concise principles injected every turn. The model now sees *what matters* (principles) instead of *what to do* (rigid rules).
+
+**Prompt changes**:
+- 4 principles replace CONTEXT PRESSURE LEVELS, 7-item priority list, DO NOT RE-COMPRESS rules
+- Context display simplified: absolute token count only, no percentage
+- `<acp-context>` tag wrapping (backward compatible with `<dcp-context>`)
+
+**Hybrid Tips frequency**:
+- 💡 Light Tips (15-45%): Every turn — non-disruptive reminder
+- ⚠️ Warning Tips (45%+): Key nodes only — first crossing or 10pp growth, prevents over-compression
+
+**Config simplification**:
+- Removed `hardNudgeContextPercent` — merged into `minContextLimit`/`maxContextLimit`
+- Removed `perMessageNudgeGrowthPercent` — light Tips show every turn
+- `maxSummaryLength` default: 200 → 2000
+- `maxSummaryLengthHard` default: 3000 → 4000
+
+**Bug fixes**:
+- Windows path validation: `os.tmpdir()` + `path.relative()` (was hardcoded `/tmp/`)
+- Compress after-detection: reset warning tracking
+- Dead code cleanup: `shouldInjectPerMessageNudge`, no-op template
+
+---
+
 ## License
 
 AGPL-3.0-or-later -- This project is a fork of [@tarquinen/opencode-dcp](https://github.com/Tarquinen/opencode-dynamic-context-pruning). Original copyright belongs to the original author. Modifications and bug fixes by ranxianglei.

package/README.zh-CN.md

@@ -391,6 +391,34 @@ 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
+
+### v1.7.0 — 原则驱动提示
+
+**理念**：用 4 条简洁原则替代冗长的上下文管理指导。模型现在看到的是*重要原则*而非*死板规则*。
+
+**提示变更**：
+- 4 条原则替代 CONTEXT PRESSURE LEVELS、7 项优先级列表、DO NOT RE-COMPRESS 规则
+- 上下文显示简化：仅显示绝对 token 数，不显示百分比
+- `<acp-context>` 标签包裹（向后兼容 `<dcp-context>`）
+
+**混合 Tips 频率**：
+- 💡 轻量提示（15-45%）：每轮显示 — 不打扰
+- ⚠️ 警告提示（45%+）：仅关键节点 — 首次跨越或增长 10pp，防止过度压缩
+
+**配置简化**：
+- 移除 `hardNudgeContextPercent` — 合并到 `minContextLimit`/`maxContextLimit`
+- 移除 `perMessageNudgeGrowthPercent` — 轻量提示每轮显示
+- `maxSummaryLength` 默认值：200 → 2000
+- `maxSummaryLengthHard` 默认值：3000 → 4000
+
+**Bug 修复**：
+- Windows 路径校验：`os.tmpdir()` + `path.relative()`（原硬编码 `/tmp/`）
+- 压缩检测后：重置警告追踪
+- 死代码清理：`shouldInjectPerMessageNudge`、空操作模板
+
+---
+
 ## 许可证
 
 AGPL-3.0-or-later — 本项目是 [@tarquinen/opencode-dcp](https://github.com/Tarquinen/opencode-dynamic-context-pruning) 的分支。原始版权归原始作者所有。修改和错误修复由 ranxianglei 完成。

package/dist/index.js

@@ -894,13 +894,12 @@ var VALID_CONFIG_KEYS = /* @__PURE__ */ new Set([
   "compress.modelMaxLimits",
   "compress.modelMinLimits",
   "compress.nudgeFrequency",
-  "compress.perMessageNudgeGrowthPercent",
+  "compress.minNudgeContextPercent",
   "compress.iterationNudgeThreshold",
   "compress.nudgeForce",
   "compress.protectedTools",
   "compress.protectTags",
   "compress.protectUserMessages",
-  "compress.maxSummaryLength",
   "compress.maxSummaryLengthHard",
   "compress.minCompressRange",
   "gc",
@@ -1121,13 +1120,6 @@ function validateConfigTypes(config) {
           actual: `${compress.nudgeFrequency} (will be clamped to 1)`
         });
       }
-      if (compress.perMessageNudgeGrowthPercent !== void 0 && typeof compress.perMessageNudgeGrowthPercent !== "number") {
-        errors.push({
-          key: "compress.perMessageNudgeGrowthPercent",
-          expected: "number",
-          actual: typeof compress.perMessageNudgeGrowthPercent
-        });
-      }
       if (compress.iterationNudgeThreshold !== void 0 && typeof compress.iterationNudgeThreshold !== "number") {
         errors.push({
           key: "compress.iterationNudgeThreshold",
@@ -1163,20 +1155,6 @@ function validateConfigTypes(config) {
           actual: typeof compress.protectUserMessages
         });
       }
-      if (compress.maxSummaryLength !== void 0 && typeof compress.maxSummaryLength !== "number") {
-        errors.push({
-          key: "compress.maxSummaryLength",
-          expected: "number",
-          actual: typeof compress.maxSummaryLength
-        });
-      }
-      if (typeof compress.maxSummaryLength === "number" && compress.maxSummaryLength < 1) {
-        errors.push({
-          key: "compress.maxSummaryLength",
-          expected: "positive number (>= 1)",
-          actual: `${compress.maxSummaryLength}`
-        });
-      }
       if (compress.maxSummaryLengthHard !== void 0 && typeof compress.maxSummaryLengthHard !== "number") {
         errors.push({
           key: "compress.maxSummaryLengthHard",
@@ -1191,13 +1169,6 @@ function validateConfigTypes(config) {
           actual: `${compress.maxSummaryLengthHard}`
         });
       }
-      if (typeof compress.maxSummaryLength === "number" && typeof compress.maxSummaryLengthHard === "number" && compress.maxSummaryLengthHard < compress.maxSummaryLength) {
-        errors.push({
-          key: "compress.maxSummaryLengthHard",
-          expected: `>= maxSummaryLength (${compress.maxSummaryLength})`,
-          actual: `${compress.maxSummaryLengthHard}`
-        });
-      }
       if (compress.minCompressRange !== void 0 && typeof compress.minCompressRange !== "number") {
         errors.push({
           key: "compress.minCompressRange",
@@ -1519,14 +1490,13 @@ var defaultConfig = {
     maxContextLimit: "55%",
     minContextLimit: "45%",
     nudgeFrequency: 5,
-    perMessageNudgeGrowthPercent: 3,
+    minNudgeContextPercent: 15,
     iterationNudgeThreshold: 15,
     nudgeForce: "soft",
     protectedTools: [...COMPRESS_DEFAULT_PROTECTED_TOOLS],
     protectTags: false,
     protectUserMessages: false,
-    maxSummaryLength: 200,
-    maxSummaryLengthHard: 3e3,
+    maxSummaryLengthHard: 4e3,
     minCompressRange: 2e3
   },
   strategies: {
@@ -1674,13 +1644,12 @@ function mergeCompress(base, override) {
     modelMaxLimits: override.modelMaxLimits ?? base.modelMaxLimits,
     modelMinLimits: override.modelMinLimits ?? base.modelMinLimits,
     nudgeFrequency: override.nudgeFrequency ?? base.nudgeFrequency,
-    perMessageNudgeGrowthPercent: override.perMessageNudgeGrowthPercent ?? base.perMessageNudgeGrowthPercent,
+    minNudgeContextPercent: override.minNudgeContextPercent ?? base.minNudgeContextPercent,
     iterationNudgeThreshold: override.iterationNudgeThreshold ?? base.iterationNudgeThreshold,
     nudgeForce: override.nudgeForce ?? base.nudgeForce,
     protectedTools: [.../* @__PURE__ */ new Set([...base.protectedTools, ...override.protectedTools ?? []])],
     protectTags: override.protectTags ?? base.protectTags,
     protectUserMessages: override.protectUserMessages ?? base.protectUserMessages,
-    maxSummaryLength: override.maxSummaryLength ?? base.maxSummaryLength,
     maxSummaryLengthHard: override.maxSummaryLengthHard ?? base.maxSummaryLengthHard,
     minCompressRange: override.minCompressRange ?? base.minCompressRange
   };
@@ -4596,7 +4565,7 @@ ${output}`);
 }
 
 // lib/compress/message.ts
-function buildSchema(maxSummaryLength) {
+function buildSchema() {
   return {
     topic: tool2.schema.string().describe(
       "Short label (3-5 words) for the overall batch - e.g., 'Closed Research Notes'"
@@ -4606,10 +4575,11 @@ function buildSchema(maxSummaryLength) {
         messageId: tool2.schema.string().describe("Raw message ID to compress (e.g. m00001)"),
         topic: tool2.schema.string().describe("Short label (3-5 words) for this one message summary"),
         summary: tool2.schema.string().describe(
-          `Complete technical summary replacing that one message. Aim for <=${maxSummaryLength} chars; exceed only when strictly necessary to preserve critical detail (file paths, decisions, signatures, exact values). Never pad.`
+          "Complete technical summary replacing that one message. Keep only essential details (conclusions, file paths, decisions, exact values, etc.)."
         )
       })
-    ).describe("Batch of individual message summaries to create in one tool call")
+    ).describe("Batch of individual message summaries to create in one tool call"),
+    summaryMaxChars: tool2.schema.number().optional().describe("Override max summary length (default max: 4000 chars). Use when content is important and needs more detail \u2014 don't lose critical info just to fit the limit.")
   };
 }
 function createCompressMessageTool(ctx) {
@@ -4617,15 +4587,18 @@ function createCompressMessageTool(ctx) {
   const runtimePrompts = ctx.prompts.getRuntimePrompts();
   return tool2({
     description: runtimePrompts.compressMessage + MESSAGE_FORMAT_EXTENSION,
-    args: buildSchema(ctx.config.compress.maxSummaryLength),
+    args: buildSchema(),
     async execute(args, toolCtx) {
       const input = args;
       validateArgs(input);
-      const maxSummaryLengthHard = ctx.config.compress.maxSummaryLengthHard;
+      const maxLen = args.summaryMaxChars ?? ctx.config.compress.maxSummaryLengthHard;
       for (const entry of input.content) {
-        if (entry.summary.length > maxSummaryLengthHard) {
+        if (entry.summary.length > maxLen) {
           throw new Error(
-            `Summary too long (${entry.summary.length} chars; limit ${maxSummaryLengthHard}). Rewrite to under ${maxSummaryLengthHard} chars \u2014 keep only the most essential details (conclusions, file paths, decisions, exact values) and drop verbose narration or raw dumps.`
+            `Summary too long (${entry.summary.length} chars, max ${maxLen}).
+1. If this summary is nearly the same size as the original content, it may not be worth compressing \u2014 skip it.
+2. Strip noise (failed attempts, verbose outputs) but keep project-critical details (file paths, decisions, exact values).
+3. For important content needing detail, pass summaryMaxChars to increase the limit \u2014 don't lose critical info just to fit. Example: add "summaryMaxChars": 6000 to the tool call args.`
           );
         }
       }
@@ -4870,7 +4843,7 @@ function appendMissingBlockSummaries(summary, _missingBlockIds, _summaryByBlockI
 }
 
 // lib/compress/range.ts
-function buildSchema2(maxSummaryLength) {
+function buildSchema2() {
   return {
     topic: tool3.schema.string().describe("Short label (3-5 words) for display - e.g., 'Auth System Exploration'"),
     content: tool3.schema.array(
@@ -4880,12 +4853,13 @@ function buildSchema2(maxSummaryLength) {
         ),
         endId: tool3.schema.string().describe("Message or block ID marking the end of range (e.g. m00012, b5)"),
         summary: tool3.schema.string().describe(
-          `Complete technical summary replacing all content in range. Aim for <=${maxSummaryLength} chars; exceed only when strictly necessary to preserve critical detail (file paths, decisions, signatures, exact values). Never pad.`
+          "Complete technical summary replacing all content in range. Keep only essential details (conclusions, file paths, decisions, exact values, etc.)."
         )
       })
     ).describe(
       "One or more ranges to compress, each with start/end boundaries and a summary"
-    )
+    ),
+    summaryMaxChars: tool3.schema.number().optional().describe("Override max summary length (default max: 4000 chars). Use when content is important and needs more detail \u2014 don't lose critical info just to fit the limit.")
   };
 }
 function createCompressRangeTool(ctx) {
@@ -4893,15 +4867,18 @@ function createCompressRangeTool(ctx) {
   const runtimePrompts = ctx.prompts.getRuntimePrompts();
   return tool3({
     description: runtimePrompts.compressRange + RANGE_FORMAT_EXTENSION,
-    args: buildSchema2(ctx.config.compress.maxSummaryLength),
+    args: buildSchema2(),
     async execute(args, toolCtx) {
       const input = args;
       validateArgs2(input);
-      const maxSummaryLengthHard = ctx.config.compress.maxSummaryLengthHard;
+      const maxLen = args.summaryMaxChars ?? ctx.config.compress.maxSummaryLengthHard;
       for (const entry of input.content) {
-        if (entry.summary.length > maxSummaryLengthHard) {
+        if (entry.summary.length > maxLen) {
           throw new Error(
-            `Summary too long (${entry.summary.length} chars; limit ${maxSummaryLengthHard}). Rewrite to under ${maxSummaryLengthHard} chars \u2014 keep only the most essential details (conclusions, file paths, decisions, exact values) and drop verbose narration or raw dumps.`
+            `Summary too long (${entry.summary.length} chars, max ${maxLen}).
+1. If this summary is nearly the same size as the original content, it may not be worth compressing \u2014 skip it.
+2. Strip noise (failed attempts, verbose outputs) but keep project-critical details (file paths, decisions, exact values).
+3. For important content needing detail, pass summaryMaxChars to increase the limit \u2014 don't lose critical info just to fit. Example: add "summaryMaxChars": 6000 to the tool call args.`
           );
         }
       }
@@ -5067,10 +5044,10 @@ var MERGED_SUMMARY_FOOTER = `
 [End ACP compressed context summary]
 
 `;
-var DCP_BLOCK_ID_TAG_REGEX = /(<dcp-message-id(?=[\s>])[^>]*>)b\d+(<\/dcp-message-id>)/g;
-var DCP_MESSAGE_REF_TAG_REGEX = /<dcp-message-id>m\d+<\/dcp-message-id>/g;
-var DCP_PAIRED_TAG_REGEX = /<dcp[^>]*>[\s\S]*?<\/dcp[^>]*>/gi;
-var DCP_UNPAIRED_TAG_REGEX = /<\/?dcp[^>]*>/gi;
+var DCP_BLOCK_ID_TAG_REGEX = /(<dcp-message-id(?=[\s>])[^>]*>)b\d+(<\/(?:dcp|acp)-message-id>)/g;
+var DCP_MESSAGE_REF_TAG_REGEX = /<dcp-message-id>m\d+<\/(?:dcp|acp)-message-id>/g;
+var DCP_PAIRED_TAG_REGEX = /<dcp[^>]*>[\s\S]*?<\/(?:dcp|acp)[^>]*>/gi;
+var DCP_UNPAIRED_TAG_REGEX = /<\/?(?:dcp|acp)[^>]*>/gi;
 var generateStableId = (prefix, seed) => {
   const hash = createHash("sha256").update(seed).digest("hex").slice(0, SUMMARY_ID_HASH_LENGTH);
   return `${prefix}_${hash}`;
@@ -5528,7 +5505,11 @@ var syncCompressPermissionState = (state, config, hostPermissions, messages) =>
 // lib/prompts/extensions/nudge.ts
 function buildCompressedBlockGuidance(state, gcConfig, context) {
   const activeBlockIds = Array.from(state.prune.messages.activeBlockIds).filter((id) => Number.isInteger(id) && id > 0).sort((a, b) => a - b);
-  const refs = activeBlockIds.map((id) => `b${id}`);
+  const refs = activeBlockIds.map((id) => {
+    const block = state.prune.messages.blocksById.get(id);
+    const tokens = block?.summaryTokens ?? 0;
+    return `b${id}${tokens > 0 ? ` (${tokens}t)` : ""}`;
+  });
   const blockCount = refs.length;
   let blockList;
   if (blockCount <= 20) {
@@ -5539,12 +5520,10 @@ function buildCompressedBlockGuidance(state, gcConfig, context) {
   }
   const includeHint = context?.includeHint ?? true;
   const lines = [
-    "Compressed block context:",
-    `- Active compressed blocks: ${blockCount} (${blockList})`,
-    "- System auto-detects blocks in range \u2014 no need to manually list (bN) placeholders. Just write a short prose summary."
+    `- Compressed blocks: ${blockCount} (${blockList})`
   ];
   if (includeHint) {
-    lines.push("- \u{1F4A1} When you've finished using tool outputs, compress them \u2014 you can decompress later if needed. Lean context improves accuracy.");
+    lines.push("- \u{1F4A1} Tools: compress, decompress, search_context.");
   }
   if (blockCount > 50) {
     const oldBlockIds = activeBlockIds.slice(0, Math.max(0, blockCount - 20));
@@ -5574,8 +5553,6 @@ function buildCompressedBlockGuidance(state, gcConfig, context) {
         lines.push(...targets);
         lines.push(`  System auto-detects blocks in range \u2014 no need to manually list (bN) placeholders. Just write a short prose summary.`);
       }
-    } else {
-      lines.push(`- \u{1F500} You have ${blockCount} blocks \u2014 use compress to consolidate adjacent same-topic blocks.`);
     }
   }
   const usageRatio = context?.currentTokens && context?.modelContextLimit ? context.currentTokens / context.modelContextLimit : 0;
@@ -5911,41 +5888,15 @@ function applyMessageModeAnchoredNudge(anchorMessageIds, messages, baseNudgeText
     injectAnchoredNudge(message, nudgeText);
   }
 }
-function resolveThresholdPercent(threshold, modelContextLimit) {
-  if (threshold === void 0) return void 0;
-  if (typeof threshold === "number") {
-    if (!modelContextLimit) return void 0;
-    return threshold / modelContextLimit * 100;
-  }
-  const parsed = parseFloat(threshold);
-  return isNaN(parsed) ? void 0 : parsed;
-}
-function buildContextUsageGuidance(config, currentTokens, modelContextLimit, minimal = false) {
+function buildContextUsageGuidance(config, currentTokens, modelContextLimit) {
   if (currentTokens === void 0 || modelContextLimit === void 0 || modelContextLimit === 0) {
     return "";
   }
-  const pct = currentTokens / modelContextLimit * 100;
-  const percentage = pct.toFixed(1);
   const formatK = (n) => n >= 1e3 ? `${(n / 1e3).toFixed(1)}K` : String(n);
-  const minPct = resolveThresholdPercent(config.compress.minContextLimit, modelContextLimit) ?? 45;
-  const maxPct = resolveThresholdPercent(config.compress.maxContextLimit, modelContextLimit) ?? 55;
-  const base = `Context usage: ${formatK(currentTokens)} / ${formatK(modelContextLimit)} tokens (${percentage}%).`;
-  if (minimal) {
-    return `
-
-${base}`;
-  }
-  let guidance;
-  if (pct < minPct) {
-    guidance = " \u{1F4A1} Be frugal with context. If any visible tool output exceeds 5000 characters and you've finished reading it, compress it into a summary now \u2014 don't keep large outputs 'just in case'. You can decompress later if needed.";
-  } else if (pct < maxPct) {
-    guidance = " \u26A0\uFE0F Context is growing \u2014 compress completed sections and high-token waste now.";
-  } else {
-    guidance = " \u{1F525} Context is high \u2014 compress aggressively, preserve only what is essential.";
-  }
   return `
 
-${base}${guidance}`;
+Context: ${formatK(currentTokens)} tokens.
+All compression serves the primary task, but be frugal. Context capacity is precious \u2014 compress waste promptly. Save context by compressing consumed outputs, not by avoiding tools. Compress by need, not by percentage.`;
 }
 function applyAnchoredNudges(state, config, messages, prompts, compressionPriorities, currentTokens, modelContextLimit, suffixMessage) {
   const contextUsageInfo = buildContextUsageGuidance(config, currentTokens, modelContextLimit);
@@ -6039,18 +5990,6 @@ function createSuffixMessage(messages) {
   messages.push(synthetic);
   return synthetic;
 }
-function shouldInjectPerMessageNudge(state, config, currentTokens, modelContextLimit) {
-  const turn = state.currentTurn ?? 0;
-  const lastTurn = state.nudges.lastPerMessageNudgeTurn ?? 0;
-  const turnsSinceLast = turn - lastTurn;
-  const tokens = currentTokens ?? 0;
-  const lastTokens = state.nudges.lastPerMessageNudgeTokens ?? 0;
-  const tokenGrowth = tokens - lastTokens;
-  const tokenGrowthPercent = modelContextLimit ? tokenGrowth / modelContextLimit * 100 : 0;
-  const frequency = config.compress.nudgeFrequency ?? 5;
-  const growthThreshold = config.compress.perMessageNudgeGrowthPercent ?? 3;
-  return turnsSinceLast >= frequency || tokenGrowthPercent >= growthThreshold;
-}
 var injectCompressNudges = (state, config, logger, messages, prompts, compressionPriorities) => {
   if (compressPermission(state, config) === "deny") {
     return;
@@ -6064,6 +6003,7 @@ var injectCompressNudges = (state, config, logger, messages, prompts, compressio
     state.nudges.contextLimitAnchors.clear();
     state.nudges.turnNudgeAnchors.clear();
     state.nudges.iterationNudgeAnchors.clear();
+    state.nudges.lastPerMessageNudgeTokens = 0;
     void saveSessionState(state, logger);
     return;
   }
@@ -6135,8 +6075,24 @@ var injectCompressNudges = (state, config, logger, messages, prompts, compressio
   }
   const suffixMessage = createSuffixMessage(messages);
   applyAnchoredNudges(state, config, messages, prompts, compressionPriorities, currentTokens, modelContextLimit, suffixMessage);
-  const shouldNudge = shouldInjectPerMessageNudge(state, config, currentTokens, modelContextLimit);
-  injectContextUsage(suffixMessage, config, currentTokens, modelContextLimit, !shouldNudge);
+  const contextPct = modelContextLimit && currentTokens ? currentTokens / modelContextLimit * 100 : 0;
+  const minPercent = config.compress?.minNudgeContextPercent ?? 15;
+  injectContextUsage(suffixMessage, config, currentTokens, modelContextLimit);
+  let tipsText = null;
+  if (overMaxLimit || overMinLimit) {
+    const lastWarnPct = state.nudges.lastPerMessageNudgeTokens && modelContextLimit ? state.nudges.lastPerMessageNudgeTokens / modelContextLimit * 100 : 0;
+    const growthSinceWarn = contextPct - lastWarnPct;
+    if (lastWarnPct === 0 || growthSinceWarn >= 10) {
+      tipsText = overMaxLimit ? '\n\n\u26A0\uFE0F Context limit reached \u2014 compress now. Prioritize consumed tool outputs.\n\n{ "topic": "...", "content": [{ "startId": "<ID>", "endId": "<ID>", "summary": "..." }] }\n\nOnly use IDs from visible messages above. Compress older work first.' : "\n\n\u26A0\uFE0F Context is growing \u2014 consider compressing older work. Tools: compress, decompress, search_context.";
+      state.nudges.lastPerMessageNudgeTokens = currentTokens ?? 0;
+      state.nudges.lastPerMessageNudgeTurn = state.currentTurn ?? 0;
+    }
+  } else if (contextPct >= minPercent) {
+    tipsText = "\n\n\u{1F4A1} Tools: compress, decompress, search_context.";
+    if (state.nudges.lastPerMessageNudgeTokens) {
+      state.nudges.lastPerMessageNudgeTokens = 0;
+    }
+  }
   if (config.compress.mode !== "message") {
     const visibleMessageIds = new Set(
       messages.map((message) => message.info.id)
@@ -6144,26 +6100,29 @@ var injectCompressNudges = (state, config, logger, messages, prompts, compressio
     const blockGuidance = buildCompressedBlockGuidance(state, config.gc, {
       currentTokens,
       modelContextLimit,
-      includeHint: shouldNudge,
+      includeHint: tipsText !== null,
       visibleMessageIds
     });
     if (blockGuidance.trim() && suffixMessage) {
       appendToLastTextPart(suffixMessage, "\n\n" + blockGuidance);
     }
   }
-  if (shouldNudge) {
-    state.nudges.lastPerMessageNudgeTurn = state.currentTurn ?? 0;
-    state.nudges.lastPerMessageNudgeTokens = currentTokens ?? 0;
+  if (tipsText && suffixMessage) {
+    appendToLastTextPart(suffixMessage, tipsText);
   }
   injectVisibleIdRange(state, messages, suffixMessage);
+  if (suffixMessage) {
+    appendToLastTextPart(suffixMessage, "\n</acp-context>");
+  }
   if (anchorsChanged) {
     void saveSessionState(state, logger);
   }
 };
-function injectContextUsage(target, config, currentTokens, modelContextLimit, minimal = false) {
+function injectContextUsage(target, config, currentTokens, modelContextLimit) {
   if (!target) return;
-  const usageTag = buildContextUsageGuidance(config, currentTokens, modelContextLimit, minimal);
-  if (!usageTag) return;
+  const rawUsage = buildContextUsageGuidance(config, currentTokens, modelContextLimit);
+  if (!rawUsage) return;
+  const usageTag = rawUsage;
   for (const part of target.parts) {
     if (part.type === "text") {
       appendToTextPart(part, usageTag);
@@ -6187,7 +6146,7 @@ function injectVisibleIdRange(state, messages, target) {
   const last = visibleRefs[visibleRefs.length - 1];
   const rangeTag = `
 
-[Visible message IDs: ${first} to ${last} (${visibleRefs.length} messages). Only use IDs in this range for compress.]`;
+[Visible messages: ${first} to ${last} (${visibleRefs.length} messages)]`;
   for (const part of target.parts) {
     if (part.type === "text") {
       appendToTextPart(part, rangeTag);
@@ -6611,7 +6570,8 @@ IMPORTANT:
 - Do NOT call this tool in parallel with compress \u2014 their state mutations may conflict.`;
 function buildSchema3() {
   return {
-    blockId: tool4.schema.string().describe('Block reference to decompress (e.g., "b0", "b2")')
+    blockId: tool4.schema.string().describe('Block reference to decompress (e.g., "b0", "b2")'),
+    toFile: tool4.schema.string().optional().describe("If provided, writes restored content to this file path instead of inflating context. Block stays compressed. Use read tool to access specific parts. Example: '/tmp/block52.txt'")
   };
 }
 function createDecompressTool(ctx) {
@@ -6640,6 +6600,40 @@ function createDecompressTool(ctx) {
         }
         return `Error: Block ${target.displayId} is not active. It may have already been decompressed.`;
       }
+      if (args.toFile) {
+        const targetPath = args.toFile;
+        const os = await import("os");
+        const path = await import("path");
+        const allowedDirs = [
+          os.tmpdir() + "/",
+          path.join(os.homedir(), ".cache", "opencode") + "/"
+        ];
+        const resolved = path.resolve(targetPath);
+        const isAllowed = allowedDirs.some((dir) => {
+          const rel = path.relative(dir, resolved);
+          return rel === "" || !rel.startsWith("..") && !path.isAbsolute(rel);
+        });
+        if (!isAllowed) {
+          return `Error: toFile path must be under ${os.tmpdir()} or ~/.cache/opencode/. Got: ${targetPath}`;
+        }
+        const block = activeBlocks[0];
+        const msgIds = new Set(block.effectiveMessageIds ?? []);
+        const blockMessages = rawMessages.filter((m) => {
+          const id = m.id ?? m.messageId ?? "";
+          return msgIds.has(id);
+        });
+        const lines2 = blockMessages.map((m) => {
+          const msg = m;
+          const role = msg.role || msg.type || "unknown";
+          const content = typeof msg.content === "string" ? msg.content : typeof msg.text === "string" ? msg.text : JSON.stringify(msg.content || msg.text || "");
+          return `[${role}]
+${content}`;
+        });
+        const { writeFile: writeFile3 } = await import("fs/promises");
+        const fileContent = lines2.length > 0 ? lines2.join("\n\n---\n\n") : block.summary ?? "(no content available)";
+        await writeFile3(args.toFile, fileContent, "utf-8");
+        return `Block b${target.displayId} content (${blockMessages.length} messages, ${fileContent.length} chars) written to ${args.toFile}. Block stays compressed \u2014 context unchanged. Use read tool to access specific parts.`;
+      }
       const activeMessagesBefore = snapshotActiveMessages(messagesState);
       const activeBlockIdsBefore = new Set(messagesState.activeBlockIds);
       deactivateCompressionTarget(messagesState, target);
@@ -6903,71 +6897,19 @@ You operate in a context-constrained environment. Context management helps prese
 
 The tools you have for context management are \`compress\`, \`decompress\`, and \`search_context\`. \`compress\` replaces older conversation content with technical summaries you produce. \`decompress\` restores previously compressed content when you need exact details. \`search_context\` searches compressed block summaries (and visible messages) to locate relevant content before you decompress.
 
-\`<dcp-message-id>\` and \`<dcp-system-reminder>\` tags are environment-injected metadata. Do not output them.
+\`<acp-context>\` tags wrap ACP (Agent Context Pruning) system metadata \u2014 context management information injected each turn. This is system data, not user input. You may also see \`<dcp-message-id>\` and \`<dcp-system-reminder>\` tags \u2014 these are equivalent (DCP was the previous name for ACP).
 
 COMPRESSION PHILOSOPHY
 
-Compression replaces raw conversation content with dense summaries. When used correctly, it keeps your context sharp and focused. When used carelessly, it destroys information you need.
-
-The key principle: compress selectively to keep context lean \u2014 but never compress content you're actively using for an ongoing task. Compression is for COMPLETED work, not work in progress. Before compressing, ask: "Will I need this in the next few turns?" If yes or unsure, keep it. Large tool outputs (shell, diffs, logs) can be compressed into summaries after the task using them is done \u2014 you can decompress later if needed. Extract and keep what matters: user intent, key decisions, file paths, and important findings \u2014 even if buried in large messages. Compress everything else, including verbose parts of user messages, large code dumps, and long discussions.
-
-Target the largest UNCOMPRESSED content first. Savings scale with original size \u2014 compressing a 5000-token tool output frees far more than re-shrinking an already-summarized 300-token block.
-
-CONTEXT PRESSURE LEVELS
-
-- Normal: After completing a task or sub-task, compress its tool outputs (agent results, verbose commands, large tool outputs) into summaries. Do NOT compress content you're actively using for an ongoing task \u2014 wait until the task is complete. You can decompress later if needed.
-- Elevated: Context is growing \u2014 compress completed sections and high-token waste now.
-- Critical: Compress aggressively now \u2014 preserve only what is essential for the current task.
-
-WHAT TO COMPRESS FIRST (high value, low risk)
-
-- Agent/subagent review and consultation results: Prime compression targets when context pressure rises \u2014 the surrounding reasoning and tool-call chatter is typically the largest block of uncompressed content. Note: if the agent tool is in your protected list, its output is auto-preserved in the summary, so the savings come from the surrounding conversation, not the agent output itself. Compress once you have fully consumed the results (all recommended actions applied or recorded in files). Recover via \`decompress\` while the block is still active. Re-invoking the agent is a last resort \u2014 it is a fresh run, not a cache hit.
-- Verbose command output (build/test runs, git diff/log/status, publish logs, directory listings): Once you have read the result, compress. Keep only the verdict \u2014 pass/fail status, commit hash, version number, or count. For failures, keep the specific error messages and file/line references needed to act on them. The full output is reproducible by re-running the command.
-- Exploration that led nowhere (failed approaches, dead-end searches): Compress to a one-line note about what was tried and why it failed.
-- Redundant tool results (reading the same file multiple times, repeated status checks, exhausted search results): Keep only the most recent result.
-- Intermediate steps of completed multi-step tasks: Once the task is done, compress the process. Keep only the final outcome.
-- Resolved discussion threads (clarification rounds, negotiated requirements, design debate that reached a decision): Once a conclusion is recorded, compress the back-and-forth. Keep the decision and its rationale.
-- Large file contents that have already been used and are no longer needed: Compress to a summary of key functions, types, or patterns.
-
-DO NOT RE-COMPRESS (low value, diminishing returns)
-
-- Already-compressed block summaries: Re-compressing a summary into a shorter summary saves negligible tokens. If a block needs better detail, use \`decompress\` to restore it, then compress the original content properly. Exception: if a block-aging warning flags specific block IDs as facing GC truncation, re-summarize exactly those flagged blocks into a fresh range \u2014 this preserves detail that GC would otherwise destroy.
-- Short messages (1-3 sentences): The compression overhead (block metadata, summary structure) may exceed the tokens saved.
-- Content whose immediate use is complete \u2014 the task it supported is done and no open todo/plan references it. If still in active use, let it stay.
-- User instructions and requirements: These must remain visible until the task is complete.
-- Tool calls that are still pending or in-progress: Wait until the result is returned and consumed.
-
-WHAT TO COMPRESS CAREFULLY (high risk - verify before compressing)
-
-- Temporary secrets/keys/tokens needed later: Do NOT compress unless recorded elsewhere
-- File paths and directory structures: Keep in summary - losing these wastes tokens rediscovering them
-- Key function/method signatures and APIs: Summarize with exact names and signatures
-- Critical error messages and stack traces: Keep the error type and key detail in summary
-- User preferences and requirements: These must survive compression intact
-- Architectural decisions and rationale: Summarize the decision, not just the conclusion
-
-BEFORE COMPRESSING IMPORTANT CONTENT
-
-Verify the information is persisted in one of:
-- A file you have written or edited
-- An issue, PR, or devlog entry
-- The compression summary itself (include the critical bits explicitly)
-
-If it is not persisted anywhere, either persist it first or include it explicitly in your compression summary.
-
-AFTER COMPRESSING
-
-Generate recovery breadcrumbs in your summary so future-you can reconstruct the context:
-- Reference specific files by path
-- Include key variable names, function signatures, or configuration values
-- Note what was decided and why, not just what was done
-- Example: "Implemented auth check in src/middleware.ts using validateToken() from auth.ts - user table is users not user"
+All compression serves the primary task, but be frugal. Two failure modes to avoid:
+- Over-compression: Compressing too aggressively loses critical details, decisions, and state needed for your task. This directly harms task quality.
+- Under-compression: Failing to compress verbose outputs causes context overflow, reducing accuracy and eventually blocking your work.
 
-If you later realize you need the original details from a compressed block, use \`decompress\` to restore them. You can decompress, read the content, then re-compress if needed.
+Balance is key. Compress selectively to keep context lean. But never compress content you're actively using for an ongoing task. Use \`search_context\` to find compressed content when needed, and \`decompress\` to restore details.
 
-Use \`search_context\` to find relevant compressed content before decompressing \u2014 it returns ranked matches across all active block summaries so you can pick the right block ID without inflating context by trial-and-error decompression.
+BE FRUGAL
 
-Use \`compress\` and \`decompress\` deliberately with quality-first summaries. Prioritize stale content intelligently to maintain a high-signal context window.
+Be frugal with context \u2014 compress obvious waste promptly. Examples include verbose command output (build/test logs, git diff/status, npm install), sub-agent results once consumed, experiment/training logs (keep final metrics only), duplicate file reads, and failed explorations. Any content that is finished serving the task and would not be needed in upcoming turns should be compressed \u2014 not just these examples.
 `;
 
 // lib/prompts/compress-range.ts
@@ -7230,7 +7172,7 @@ var PROMPT_DEFINITIONS = [
 ];
 var HTML_COMMENT_REGEX = /<!--[\s\S]*?-->/g;
 var LEGACY_INLINE_COMMENT_LINE_REGEX = /^[ \t]*\/\/.*?\/\/[ \t]*$/gm;
-var DCP_SYSTEM_REMINDER_TAG_REGEX = /^\s*<dcp-system-reminder\b[^>]*>[\s\S]*<\/dcp-system-reminder>\s*$/i;
+var DCP_SYSTEM_REMINDER_TAG_REGEX = /^\s*<dcp-system-reminder\b[^>]*>[\s\S]*<\/(?:dcp|acp)-system-reminder>\s*$/i;
 var DEFAULTS_README_FILE = "README.md";
 var BUNDLED_EDITABLE_PROMPTS = {
   system: SYSTEM,
@@ -7309,7 +7251,7 @@ function stripConditionalTag(content, tagName) {
 function unwrapDcpTagIfWrapped(content) {
   const trimmed = content.trim();
   if (DCP_SYSTEM_REMINDER_TAG_REGEX.test(trimmed)) {
-    return trimmed.replace(/^\s*<dcp-system-reminder\b[^>]*>\s*/i, "").replace(/\s*<\/dcp-system-reminder>\s*$/i, "").trim();
+    return trimmed.replace(/^\s*<dcp-system-reminder\b[^>]*>\s*/i, "").replace(/\s*<\/(?:dcp|acp)-system-reminder>\s*$/i, "").trim();
   }
   return trimmed;
 }
@@ -7319,7 +7261,7 @@ function normalizeReminderPromptContent(content) {
     return "";
   }
   const startsWrapped = /^\s*<dcp-system-reminder\b[^>]*>/i.test(normalized);
-  const endsWrapped = /<\/dcp-system-reminder>\s*$/i.test(normalized);
+  const endsWrapped = /<\/(?:dcp|acp)-system-reminder>\s*$/i.test(normalized);
   if (startsWrapped !== endsWrapped) {
     return "";
   }