opencode-acp 1.13.9-dev.1 → 1.14.0

package/dist/index.js

@@ -909,6 +909,10 @@ var VALID_CONFIG_KEYS = /* @__PURE__ */ new Set([
   "compress.emergencyThresholdPercent",
   "compress.maxVisibleSegments",
   "compress.keepEmbedMaxChars",
+  "compress.lastSegmentSoftBlock",
+  "compress.preserveRecentMessages",
+  "compress.preserveRecentTokens",
+  "compress.preserveLastUserMessage",
   "gc",
   "gc.algorithm",
   "gc.promotionThreshold",
@@ -1277,6 +1281,48 @@ function validateConfigTypes(config) {
           actual: `${compress.keepEmbedMaxChars}`
         });
       }
+      if (compress.lastSegmentSoftBlock !== void 0 && typeof compress.lastSegmentSoftBlock !== "boolean") {
+        errors.push({
+          key: "compress.lastSegmentSoftBlock",
+          expected: "boolean",
+          actual: typeof compress.lastSegmentSoftBlock
+        });
+      }
+      if (compress.preserveRecentMessages !== void 0 && typeof compress.preserveRecentMessages !== "number") {
+        errors.push({
+          key: "compress.preserveRecentMessages",
+          expected: "number",
+          actual: typeof compress.preserveRecentMessages
+        });
+      }
+      if (typeof compress.preserveRecentMessages === "number" && compress.preserveRecentMessages < 0) {
+        errors.push({
+          key: "compress.preserveRecentMessages",
+          expected: "non-negative number (>= 0)",
+          actual: `${compress.preserveRecentMessages}`
+        });
+      }
+      if (compress.preserveRecentTokens !== void 0 && typeof compress.preserveRecentTokens !== "number") {
+        errors.push({
+          key: "compress.preserveRecentTokens",
+          expected: "number",
+          actual: typeof compress.preserveRecentTokens
+        });
+      }
+      if (typeof compress.preserveRecentTokens === "number" && compress.preserveRecentTokens < 0) {
+        errors.push({
+          key: "compress.preserveRecentTokens",
+          expected: "non-negative number (>= 0)",
+          actual: `${compress.preserveRecentTokens}`
+        });
+      }
+      if (compress.preserveLastUserMessage !== void 0 && typeof compress.preserveLastUserMessage !== "boolean") {
+        errors.push({
+          key: "compress.preserveLastUserMessage",
+          expected: "boolean",
+          actual: typeof compress.preserveLastUserMessage
+        });
+      }
       if (typeof compress.iterationNudgeThreshold === "number" && compress.iterationNudgeThreshold < 1) {
         errors.push({
           key: "compress.iterationNudgeThreshold",
@@ -1601,7 +1647,10 @@ var defaultConfig = {
     emergencyThresholdPercent: "98%",
     maxVisibleSegments: 50,
     keepEmbedMaxChars: 2e3,
-    lastSegmentSoftBlock: true
+    lastSegmentSoftBlock: true,
+    preserveRecentMessages: 20,
+    preserveRecentTokens: 2e4,
+    preserveLastUserMessage: true
   },
   strategies: {
     deduplication: {
@@ -1776,7 +1825,10 @@ function mergeCompress(base, override) {
     emergencyThresholdPercent: override.emergencyThresholdPercent ?? base.emergencyThresholdPercent,
     maxVisibleSegments: override.maxVisibleSegments ?? base.maxVisibleSegments,
     keepEmbedMaxChars: override.keepEmbedMaxChars ?? base.keepEmbedMaxChars,
-    lastSegmentSoftBlock: override.lastSegmentSoftBlock ?? base.lastSegmentSoftBlock
+    lastSegmentSoftBlock: override.lastSegmentSoftBlock ?? base.lastSegmentSoftBlock,
+    preserveRecentMessages: override.preserveRecentMessages ?? base.preserveRecentMessages,
+    preserveRecentTokens: override.preserveRecentTokens ?? base.preserveRecentTokens,
+    preserveLastUserMessage: override.preserveLastUserMessage ?? base.preserveLastUserMessage
   };
 }
 function mergeCommands(base, override) {
@@ -3062,7 +3114,20 @@ ${footer}`;
 function applyCompressionState(state, input, selection, anchorMessageId, blockId, summary, consumedBlockIds, gcConfig) {
   const messagesState = state.prune.messages;
   const consumed = [...new Set(consumedBlockIds.filter((id) => Number.isInteger(id) && id > 0))];
-  const included = [...consumed];
+  const createdAt = Date.now();
+  let minConsumedTier;
+  for (const consumedBlockId of consumed) {
+    const cb = messagesState.blocksById.get(consumedBlockId);
+    if (cb) {
+      const cbTier = cb.tier ?? 1;
+      if (minConsumedTier === void 0 || cbTier < minConsumedTier) {
+        minConsumedTier = cbTier;
+      }
+    }
+  }
+  const effectiveMinTier = minConsumedTier ?? 0;
+  const outputTier = Math.min(3, effectiveMinTier + 1);
+  const targetTierForConsumption = effectiveMinTier;
   const effectiveMessageIds = new Set(selection.messageIds);
   const effectiveToolIds = new Set(selection.toolIds);
   for (const consumedBlockId of consumed) {
@@ -3070,6 +3135,9 @@ function applyCompressionState(state, input, selection, anchorMessageId, blockId
     if (!consumedBlock) {
       continue;
     }
+    if ((consumedBlock.tier ?? 1) !== targetTierForConsumption) {
+      continue;
+    }
     for (const messageId of consumedBlock.effectiveMessageIds) {
       effectiveMessageIds.add(messageId);
     }
@@ -3094,7 +3162,6 @@ function applyCompressionState(state, input, selection, anchorMessageId, blockId
       initiallyActiveToolIds.add(toolId);
     }
   }
-  const createdAt = Date.now();
   const block = {
     blockId,
     runId: input.runId,
@@ -3104,6 +3171,7 @@ function applyCompressionState(state, input, selection, anchorMessageId, blockId
     summaryTokens: input.summaryTokens,
     durationMs: 0,
     mode: input.mode,
+    tier: outputTier,
     topic: input.topic,
     batchTopic: input.batchTopic,
     startId: input.startId,
@@ -3111,8 +3179,11 @@ function applyCompressionState(state, input, selection, anchorMessageId, blockId
     anchorMessageId,
     compressMessageId: input.compressMessageId,
     compressCallId: input.compressCallId,
-    includedBlockIds: included,
-    consumedBlockIds: consumed,
+    includedBlockIds: [...consumed],
+    consumedBlockIds: consumed.filter((id) => {
+      const cb = messagesState.blocksById.get(id);
+      return cb && (cb.tier ?? 1) === targetTierForConsumption;
+    }),
     parentBlockIds: [],
     directMessageIds: [],
     directToolIds: [],
@@ -3141,6 +3212,9 @@ function applyCompressionState(state, input, selection, anchorMessageId, blockId
     if (!consumedBlock || !consumedBlock.active) {
       continue;
     }
+    if ((consumedBlock.tier ?? 1) !== targetTierForConsumption) {
+      continue;
+    }
     consumedBlock.active = false;
     consumedBlock.deactivatedAt = deactivatedAt;
     consumedBlock.deactivatedByBlockId = blockId;
@@ -3166,6 +3240,9 @@ function applyCompressionState(state, input, selection, anchorMessageId, blockId
     if (!consumedBlock) {
       continue;
     }
+    if ((consumedBlock.tier ?? 1) !== targetTierForConsumption) {
+      continue;
+    }
     for (const messageId of consumedBlock.effectiveMessageIds) {
       const entry = messagesState.byMessageId.get(messageId);
       if (!entry) {
@@ -3231,6 +3308,14 @@ function applyCompressionState(state, input, selection, anchorMessageId, blockId
   block.directMessageIds = [...newlyCompressedMessageIds];
   block.directToolIds = [...newlyCompressedToolIds];
   block.compressedTokens = compressedTokens;
+  let effectiveTokens = compressedTokens;
+  for (const consumedBlockId of consumed) {
+    const cb = messagesState.blocksById.get(consumedBlockId);
+    if (cb && (cb.tier ?? 1) === targetTierForConsumption) {
+      effectiveTokens += cb.effectiveCompressedTokens ?? cb.compressedTokens;
+    }
+  }
+  block.effectiveCompressedTokens = effectiveTokens;
   state.stats.pruneTokenCounter += compressedTokens;
   state.stats.totalPruneTokens += state.stats.pruneTokenCounter;
   state.stats.pruneTokenCounter = 0;
@@ -3593,6 +3678,7 @@ function loadPruneMessagesState(persisted) {
         runId: typeof block.runId === "number" && Number.isInteger(block.runId) && block.runId > 0 ? block.runId : blockId,
         active: block.active === true,
         deactivatedByUser: block.deactivatedByUser === true,
+        deactivatedByUserDeep: block.deactivatedByUserDeep === true ? true : void 0,
         compressedTokens: typeof block.compressedTokens === "number" && Number.isFinite(block.compressedTokens) ? Math.max(0, block.compressedTokens) : 0,
         summaryTokens: typeof block.summaryTokens === "number" && Number.isFinite(block.summaryTokens) ? Math.max(0, block.summaryTokens) : typeof block.summary === "string" ? countTokens2(block.summary) : 0,
         durationMs: typeof block.durationMs === "number" && Number.isFinite(block.durationMs) ? Math.max(0, block.durationMs) : 0,
@@ -3616,7 +3702,9 @@ function loadPruneMessagesState(persisted) {
         deactivatedByBlockId: typeof block.deactivatedByBlockId === "number" && Number.isInteger(block.deactivatedByBlockId) ? block.deactivatedByBlockId : void 0,
         summary: typeof block.summary === "string" ? block.summary : "",
         survivedCount: typeof block.survivedCount === "number" && Number.isFinite(block.survivedCount) ? Math.max(0, Math.floor(block.survivedCount)) : 0,
-        generation: block.generation === "young" || block.generation === "old" ? block.generation : void 0
+        generation: block.generation === "young" || block.generation === "old" ? block.generation : void 0,
+        tier: block.tier === 1 || block.tier === 2 || block.tier === 3 ? block.tier : void 0,
+        effectiveCompressedTokens: typeof block.effectiveCompressedTokens === "number" && Number.isFinite(block.effectiveCompressedTokens) ? Math.max(0, block.effectiveCompressedTokens) : void 0
       });
     }
   }
@@ -3675,17 +3763,38 @@ function collectTurnNudgeAnchors(messages) {
   }
   return anchors;
 }
-function getActiveSummaryTokenUsage(state) {
+function getActiveSummaryTokenUsage(state, visibleMessageIds) {
   let total = 0;
   for (const blockId of state.prune.messages.activeBlockIds) {
     const block = state.prune.messages.blocksById.get(blockId);
     if (!block || !block.active) {
       continue;
     }
+    if (visibleMessageIds && block.compressMessageId) {
+      if (!visibleMessageIds.has(block.compressMessageId)) {
+        continue;
+      }
+    }
     total += block.summaryTokens;
   }
   return total;
 }
+function getTierTokenUsage(state) {
+  let tier1Tokens = 0;
+  let tier2Tokens = 0;
+  let tier3Tokens = 0;
+  for (const blockId of state.prune.messages.activeBlockIds) {
+    const block = state.prune.messages.blocksById.get(blockId);
+    if (!block || !block.active) {
+      continue;
+    }
+    const tier = block.tier ?? 1;
+    if (tier === 1) tier1Tokens += block.summaryTokens;
+    else if (tier === 2) tier2Tokens += block.summaryTokens;
+    else tier3Tokens += block.summaryTokens;
+  }
+  return { tier1Tokens, tier2Tokens, tier3Tokens };
+}
 function resetOnCompaction(state) {
   state.toolParameters.clear();
   state.prune.tools = /* @__PURE__ */ new Map();
@@ -3697,6 +3806,8 @@ function resetOnCompaction(state) {
     lastPerMessageNudgeTokens: void 0,
     lastNudgeShownTokens: void 0,
     lastToolOutputNudgeTokens: void 0,
+    lastTier2NudgeTokens: void 0,
+    lastTier3NudgeTokens: void 0,
     shouldInjectThisTurn: void 0,
     compressBaselineSet: false
   };
@@ -3773,6 +3884,8 @@ async function saveSessionState(sessionState, logger, sessionName) {
       lastPerMessageNudgeTokens: sessionState.nudges.lastPerMessageNudgeTokens,
       lastNudgeShownTokens: sessionState.nudges.lastNudgeShownTokens,
       lastToolOutputNudgeTokens: sessionState.nudges.lastToolOutputNudgeTokens,
+      lastTier2NudgeTokens: sessionState.nudges.lastTier2NudgeTokens,
+      lastTier3NudgeTokens: sessionState.nudges.lastTier3NudgeTokens,
       compressBaselineSet: sessionState.nudges.compressBaselineSet
     },
     stats: sessionState.stats,
@@ -4385,6 +4498,8 @@ function createSessionState() {
       lastPerMessageNudgeTokens: void 0,
       lastNudgeShownTokens: void 0,
       lastToolOutputNudgeTokens: void 0,
+      lastTier2NudgeTokens: void 0,
+      lastTier3NudgeTokens: void 0,
       shouldInjectThisTurn: void 0,
       compressBaselineSet: false
     },
@@ -4428,6 +4543,8 @@ function resetSessionState(state) {
     lastPerMessageNudgeTokens: void 0,
     lastNudgeShownTokens: void 0,
     lastToolOutputNudgeTokens: void 0,
+    lastTier2NudgeTokens: void 0,
+    lastTier3NudgeTokens: void 0,
     shouldInjectThisTurn: void 0,
     compressBaselineSet: false
   };
@@ -4484,6 +4601,8 @@ async function ensureSessionInitialized(client, state, sessionId, logger, messag
   state.nudges.lastPerMessageNudgeTokens = persisted.nudges.lastPerMessageNudgeTokens;
   state.nudges.lastNudgeShownTokens = persisted.nudges.lastNudgeShownTokens;
   state.nudges.lastToolOutputNudgeTokens = persisted.nudges.lastToolOutputNudgeTokens;
+  state.nudges.lastTier2NudgeTokens = persisted.nudges.lastTier2NudgeTokens ?? persisted.nudges.lastTierNudgeTokens;
+  state.nudges.lastTier3NudgeTokens = persisted.nudges.lastTier3NudgeTokens;
   state.nudges.compressBaselineSet = persisted.nudges.compressBaselineSet ?? false;
   state.stats = {
     pruneTokenCounter: persisted.stats?.pruneTokenCounter || 0,
@@ -5047,7 +5166,7 @@ async function sendCompressNotification(client, logger, config, state, sessionId
   const logTopics = entries.map((e) => state.prune.messages.blocksById.get(e.blockId)?.topic ?? "?");
   const logCompressedTokens = entries.reduce((sum, e) => {
     const block = state.prune.messages.blocksById.get(e.blockId);
-    return sum + (block?.compressedTokens ?? 0);
+    return sum + (block?.effectiveCompressedTokens ?? block?.compressedTokens ?? 0);
   }, 0);
   const logSummaryTokens = entries.reduce((sum, e) => sum + e.summaryTokens, 0);
   logger.info("Compression completed", {
@@ -5991,7 +6110,7 @@ function evaluatePreCommitQuality(rawMessages, messageIds, messageTokenById, sum
   }
 }
 
-// node_modules/context-compress-algorithms/dist/chunk-ZRHPFN6B.js
+// node_modules/context-compress-algorithms/dist/chunk-EBKMI537.js
 var COMPRESS_PHILOSOPHY = `Compression Philosophy:
 - All compression serves the primary task, but be frugal.
 - Context capacity is precious. Save context by compressing consumed outputs, not by avoiding tools.
@@ -6036,6 +6155,69 @@ PRIORITY \u2014 when the summary must be compact, preserve in this order:
 5. Lessons learned: what failed and why.
 
 Write dense, scannable bullets \u2014 not narrative prose. If the range spans distinct concerns (request \u2192 findings \u2192 decision), group bullets under short thematic headers so a reader can scan to the part they need. Every line must earn its place. Do not mimic the style of existing summaries in context; follow these rules.`;
+var TIER2_DISTILL_RULES = `TIER 2 COMPRESSION \u2014 DISTILLATION
+
+You are compressing historical summaries (not raw conversation). These summaries have already captured the details. Your job is to DISTILL them: extract only what matters for future work, discard the process.
+
+KEEP \u2014 these are the only things that survive distillation:
+- Decisions and their rationale ("chose X over Y because Z" \u2014 the "because" is load-bearing).
+- Final outcomes: version numbers shipped, PR numbers merged/closed, bugs fixed or deferred.
+- Key lessons: what failed and why ("tried X, failed because Y"). These prevent repeating mistakes.
+- Critical constraints discovered ("must support Node 22", "AGENTS.md forbids as any").
+- Design decisions with architectural impact ("chose compress-as-anchor over synthetic messages because prefix cache").
+- Whether content is OBSOLETE or SUPERSEDED \u2014 mark with one line: "[SUPERSEDED by PR #NNN]" or "[OBSOLETE: deleted in vX.Y.Z]". Do NOT keep the obsolete content's details \u2014 just the marker and reason.
+- Function/class/type names and module paths that are the SUBJECT of the work \u2014 e.g., "fixed filterCompressedRanges in prune.ts", "added SessionStateRegistry in state.ts". Not exact line numbers or full signatures \u2014 just enough to LOCATE the code without searching.
+- Exploration findings: if a block was exploratory with no decision, keep the CONCLUSION in one line ("explored X, not viable because Y"). Do not keep the exploration process.
+
+DROP \u2014 these were useful during the work but are no longer needed:
+- Exact line numbers, diffs, verbose function signatures, full code listings.
+- Build/deploy process details, test execution steps.
+- Review process details (who reviewed, what rounds, test counts).
+- Verbose logs, command output, intermediate debugging steps.
+
+FORMAT:
+- Start each distilled block with a source header line:
+  \`Source: bN+bM+... (XK\u2192YK tok, Zx). [original topic]\`
+  Example: \`Source: b5+b7 (56K+44K\u2192268 tok, 375x). [Tool-result recap + publish]\`
+- 3-5 bullet points per source block, each a self-contained fact.
+- Dense, scannable \u2014 no narrative prose.
+- Start with the outcome, not the process: "v1.13.0 shipped (7 PRs bundled)" not "implemented 7 PRs then reviewed then merged".
+- Cross-block synthesis: if multiple source blocks cover the same topic (same PR, same feature, same bug), MERGE them into a single group of bullets. Do not repeat the same fact from different blocks \u2014 keep it once under the most relevant source header.
+
+SIZE TARGET: 50-150 tokens per source block (excluding the header). If you can't fit it in 150 tokens, you're keeping too much process. If a block has nothing worth keeping (pure noise), output just the header followed by "[no actionable content]".`;
+var TIER3_CONDENSE_RULES = `TIER 3 COMPRESSION \u2014 ULTRA-CONDENSATION
+
+You are compressing distilled summaries (Tier 2) into ultra-condensed facts (Tier 3). The distilled summaries already contain only decisions and outcomes. Your job is to reduce them to bare factual references.
+
+PRIORITY \u2014 when a source block has more facts than the size target allows, keep in this order:
+1. Shipped outcomes (versions released, PRs merged) \u2014 these are permanent record.
+2. Open work (PRs/issues still pending) \u2014 these may need follow-up.
+3. Key decisions with architectural impact ("chose X over Y because Z").
+4. Critical constraints ("must support Node 22").
+Drop everything else. Tier 3 is a lookup index, not a knowledge base.
+
+FORMAT:
+- Start with a source header line:
+  \`Source: bN+bM+... (XK\u2192YK tok, Zx). [original topic]\`
+- Output 1-3 facts per source block. Each fact is a single line: subject + outcome.
+- No explanations, no rationale, no process \u2014 just the fact.
+- Format: "[PR/Issue/Version] \u2014 [outcome in \u22648 words]"
+- Merge related facts from different source blocks if they concern the same topic.
+
+EXAMPLES:
+- "v1.13.0 shipped \u2014 quality gate + GC fix (7 PRs)"
+- "PR #196 merged \u2014 preserve-first-user (supersedes #169)"
+- "Bug 1214 fixed \u2014 compress consumed all user messages"
+- "Chose compress-as-anchor \u2014 prefix cache benefit over synthetic injection"
+- "Constraint: AGENTS.md forbids as any \u2014 never suppress types"
+
+DROP:
+- Multi-sentence context. If a fact needs >1 sentence, it's too detailed for Tier 3.
+- Lessons learned ("tried X, failed because Y") \u2014 drop UNLESS the failure is likely to recur and the block is <30 days old.
+- Design rationale details \u2014 keep the decision, drop the "because" unless it's a critical constraint.
+- Anything marked [OBSOLETE] or [SUPERSEDED] \u2014 drop entirely, note "[N blocks obsolete]" in the summary.
+
+SIZE TARGET: 30-60 tokens per source block (including header). For a batch of N source blocks, total output \u2248 N \xD7 40 tokens. If a source block has only one trivial fact, output just the header + one line.`;
 
 // lib/compress/quality-gate/rejection.ts
 function formatMetric(result, name) {
@@ -6180,17 +6362,6 @@ async function finalizeSession(ctx, toolCtx, rawMessages, entries, batchTopic) {
     contextTokensBefore
   );
 }
-function getLastVisibleMessageId(rawMessages, state) {
-  for (let i = rawMessages.length - 1; i >= 0; i--) {
-    const msg = rawMessages[i];
-    const id = msg?.info?.id;
-    if (!id || typeof id !== "string") continue;
-    if (isSyntheticMessage(msg)) continue;
-    if (state.prune.messages.byMessageId.has(id)) continue;
-    return id;
-  }
-  return null;
-}
 function checkPhantomBlock(state, plans) {
   for (let i = 0; i < plans.length; i++) {
     const plan = plans[i];
@@ -6208,6 +6379,16 @@ function checkPhantomBlock(state, plans) {
       return !entry || entry.activeBlockIds.length === 0;
     });
     if (!hasNew) {
+      if (plan.consumedBlockIds.length >= 2) {
+        const tiers = new Set(
+          plan.consumedBlockIds.map(
+            (id) => state.prune.messages.blocksById.get(id)?.tier ?? 1
+          )
+        );
+        if (tiers.size === 1) {
+          continue;
+        }
+      }
       return new Error(
         `Compression range ${i + 1} contains only already-compressed messages (0 new direct messages, 0 tokens saved). Nothing to compress \u2014 pick a range with visible, uncompressed content. Use \`acp_status({scope:"uncompressed"})\` to see which ranges are still compressible.`
       );
@@ -6215,16 +6396,71 @@ function checkPhantomBlock(state, plans) {
   }
   return null;
 }
-function checkLastSegmentDangerous(ctx, allPlanMessageIds, rawMessages, dangerous) {
+function computeProtectedRawIds(rawMessages, state, compress) {
+  const preserveN = compress.preserveRecentMessages ?? 20;
+  const preserveTokens = compress.preserveRecentTokens ?? 2e4;
+  const preserveLastUser = compress.preserveLastUserMessage ?? true;
+  const result = /* @__PURE__ */ new Set();
+  const visible = [];
+  for (const msg of rawMessages) {
+    const id = msg?.info?.id;
+    if (!id || typeof id !== "string") continue;
+    if (isSyntheticMessage(msg)) continue;
+    if (isIgnoredUserMessage(msg)) continue;
+    if (state.prune.messages.byMessageId.has(id)) continue;
+    let tokens = 0;
+    for (const part of msg.parts || []) {
+      if (part.type === "text" && typeof part.text === "string") {
+        tokens += Math.round(part.text.length / 4);
+      } else if (part.type !== "text" && part.type !== "reasoning") {
+        tokens += Math.round(JSON.stringify(part).length / 4);
+      }
+    }
+    visible.push({ id, tokens, isUser: msg.info.role === "user" });
+  }
+  if (preserveN > 0) {
+    for (const m of visible.slice(-preserveN)) {
+      result.add(m.id);
+    }
+  }
+  if (preserveTokens > 0) {
+    let tokenAccum = 0;
+    for (let i = visible.length - 1; i >= 0 && tokenAccum < preserveTokens; i--) {
+      result.add(visible[i].id);
+      tokenAccum += visible[i].tokens;
+    }
+  }
+  if (preserveLastUser) {
+    for (let i = visible.length - 1; i >= 0; i--) {
+      if (visible[i].isUser) {
+        result.add(visible[i].id);
+        break;
+      }
+    }
+  }
+  return result;
+}
+function checkProtectedRange(ctx, allPlanMessageIds, rawMessages, dangerous) {
   if (ctx.config.compress.lastSegmentSoftBlock === false) return null;
-  const lastVisibleId = getLastVisibleMessageId(rawMessages, ctx.state);
-  if (!lastVisibleId) return null;
-  const coversLast = allPlanMessageIds.some((ids) => ids.includes(lastVisibleId));
-  if (!coversLast) return null;
+  const protectedIds = computeProtectedRawIds(rawMessages, ctx.state, ctx.config.compress);
+  if (protectedIds.size === 0) return null;
+  const coveredProtected = [];
+  for (const ids of allPlanMessageIds) {
+    for (const id of ids) {
+      if (protectedIds.has(id)) {
+        coveredProtected.push(id);
+      }
+    }
+  }
+  if (coveredProtected.length === 0) return null;
   if (dangerous) return null;
+  const sample = coveredProtected.slice(0, 3).join(", ");
+  const nMsgs = ctx.config.compress.preserveRecentMessages ?? 20;
+  const nToks = ctx.config.compress.preserveRecentTokens ?? 2e4;
   return new Error(
-    `This range includes the most recent message (${lastVisibleId}), which is likely still needed for the current task step.
+    `This range includes ${coveredProtected.length} protected recent message(s) (${sample}), which are likely still needed for the current task step.
 
+Protected zone: last ${nMsgs} messages + last ${nToks >= 1e3 ? `${nToks / 1e3}K` : nToks} tokens + most recent user message.
 If you are certain this content is genuinely consumed and must be compressed, re-issue the call with \`dangerous: true\`.
 Otherwise, compress older ranges that do not include the tail of the conversation.`
   );
@@ -6423,7 +6659,7 @@ function createCompressMessageTool(factoryCtx) {
         }
       }
       const dangerous = args.dangerous === true;
-      const lastSegmentError = checkLastSegmentDangerous(
+      const lastSegmentError = checkProtectedRange(
         ctx,
         plans.map((p) => p.selection.messageIds),
         rawMessages,
@@ -6643,7 +6879,7 @@ function createCompressRangeTool(factoryCtx) {
         }
       }
       const dangerous = args.dangerous === true;
-      const lastSegmentError = checkLastSegmentDangerous(
+      const lastSegmentError = checkProtectedRange(
         ctx,
         filteredPlans.map((p) => p.selection.messageIds),
         rawMessages,
@@ -6915,10 +7151,9 @@ var syncCompressionBlocks = (state, logger, messages) => {
   messagesState.activeBlockIds.clear();
   messagesState.activeByAnchorMessageId.clear();
   const now = Date.now();
-  const missingOriginBlockIds = [];
   const orderedBlocks = Array.from(messagesState.blocksById.values()).sort(sortBlocksByCreation);
   for (const block of orderedBlocks) {
-    if (block.deactivatedByUser) {
+    if (block.deactivatedByUser || block.deactivatedByUserDeep) {
       block.active = false;
       if (block.deactivatedAt === void 0) {
         block.deactivatedAt = now;
@@ -6926,12 +7161,6 @@ var syncCompressionBlocks = (state, logger, messages) => {
       block.deactivatedByBlockId = void 0;
       continue;
     }
-    if (typeof block.anchorMessageId === "string" && block.anchorMessageId.length > 0 && !messageIds.has(block.anchorMessageId)) {
-      block.active = false;
-      block.deactivatedAt = now;
-      block.deactivatedByBlockId = void 0;
-      continue;
-    }
     for (const consumedBlockId of block.consumedBlockIds) {
       if (!messagesState.activeBlockIds.has(consumedBlockId)) {
         continue;
@@ -6976,9 +7205,8 @@ var syncCompressionBlocks = (state, logger, messages) => {
       reactivatedCount++;
     }
   }
-  if (missingOriginBlockIds.length > 0 || deactivatedCount > 0 || reactivatedCount > 0) {
+  if (deactivatedCount > 0 || reactivatedCount > 0) {
     logger.info("Synced compress block state", {
-      missingOriginCount: missingOriginBlockIds.length,
       deactivatedCount,
       reactivatedCount
     });
@@ -7389,7 +7617,7 @@ function listPriorityRefsBeforeIndex(messages, priorities, anchorIndex, priority
   return refs;
 }
 
-// node_modules/context-compress-algorithms/dist/chunk-UX2UW5FY.js
+// node_modules/context-compress-algorithms/dist/chunk-BZYW3CH5.js
 var NUDGE_GROWTH_FLOOR = 6e3;
 var NUDGE_GROWTH_CAP = 5e4;
 var NUDGE_GROWTH_RATIO = 0.05;
@@ -7522,7 +7750,10 @@ function resolveContextTokenLimit(config, state, providerId, modelId, threshold)
   return parseLimitValue(globalLimit);
 }
 function isContextOverLimits(config, state, providerId, modelId, messages) {
-  const summaryTokenExtension = config.compress.summaryBuffer ? getActiveSummaryTokenUsage(state) : 0;
+  const summaryTokenExtension = config.compress.summaryBuffer ? getActiveSummaryTokenUsage(
+    state,
+    new Set(messages.map((m) => m.info.id))
+  ) : 0;
   const resolvedMaxContextLimit = resolveContextTokenLimit(
     config,
     state,
@@ -8145,6 +8376,57 @@ ${lines2.join("\n")}`;
   return `Compressible ranges (oldest first):
 ${lines.join("\n")}`;
 }
+function computeProtectedRefs(messages, state, compress) {
+  if (compress.lastSegmentSoftBlock === false) return /* @__PURE__ */ new Set();
+  const preserveN = compress.preserveRecentMessages ?? 20;
+  const preserveTokens = compress.preserveRecentTokens ?? 2e4;
+  const preserveLastUser = compress.preserveLastUserMessage ?? true;
+  const result = /* @__PURE__ */ new Set();
+  const visible = [];
+  for (const msg of messages) {
+    if (isSyntheticMessage(msg)) continue;
+    if (isIgnoredUserMessage(msg)) continue;
+    const ref = state.messageIds.byRawId.get(msg.info.id);
+    if (!ref) continue;
+    if (state.prune.messages.byMessageId.has(msg.info.id)) continue;
+    let tokens = 0;
+    for (const part of msg.parts || []) {
+      if (part.type === "text" && typeof part.text === "string") {
+        tokens += Math.round(part.text.length / 4);
+      } else if (part.type !== "text" && part.type !== "reasoning") {
+        tokens += Math.round(JSON.stringify(part).length / 4);
+      }
+    }
+    visible.push({ ref, tokens, isUser: msg.info.role === "user" });
+  }
+  if (preserveN > 0) {
+    for (const m of visible.slice(-preserveN)) {
+      result.add(m.ref);
+    }
+  }
+  if (preserveTokens > 0) {
+    let tokenAccum = 0;
+    for (let i = visible.length - 1; i >= 0 && tokenAccum < preserveTokens; i--) {
+      result.add(visible[i].ref);
+      tokenAccum += visible[i].tokens;
+    }
+  }
+  if (preserveLastUser) {
+    for (let i = visible.length - 1; i >= 0; i--) {
+      if (visible[i].isUser) {
+        result.add(visible[i].ref);
+        break;
+      }
+    }
+  }
+  return result;
+}
+function excludeProtectedRanges(ranges, protectedRefs) {
+  if (protectedRefs.size === 0) return ranges;
+  return ranges.filter(
+    (r) => !protectedRefs.has(r.startRef) && !protectedRefs.has(r.endRef)
+  );
+}
 
 // lib/messages/inject/inject.ts
 var ACP_SUFFIX_SEED = "acp-dynamic-guidance";
@@ -8184,6 +8466,8 @@ var injectCompressNudges = (state, config, logger, messages, prompts, compressio
     state.nudges.iterationNudgeAnchors.clear();
     state.nudges.lastNudgeShownTokens = void 0;
     state.nudges.lastToolOutputNudgeTokens = void 0;
+    state.nudges.lastTier2NudgeTokens = void 0;
+    state.nudges.lastTier3NudgeTokens = void 0;
     if (wasNudgeTriggered && !state.nudges.compressBaselineSet) {
       const baseline = state.nudges.lastPerMessageNudgeTokens;
       const postCompress = currentTokens;
@@ -8322,8 +8606,10 @@ var injectCompressNudges = (state, config, logger, messages, prompts, compressio
     config.compress.protectedTools,
     config.protectedFilePatterns
   );
+  const protectedRefs = computeProtectedRefs(messages, state, config.compress);
+  const unprotectedCompressible = excludeProtectedRanges(contextRanges.compressible, protectedRefs);
   const recommendedRanges = filterRecommendedRanges(
-    contextRanges.compressible,
+    unprotectedCompressible,
     contextRanges.protected,
     { modelContextLimit, growthRatio: 0.05, logger }
   );
@@ -8350,11 +8636,92 @@ var injectCompressNudges = (state, config, logger, messages, prompts, compressio
   const filterSuppressed = contextRanges.compressible.length > 0 && !hasRecommendations;
   const allProtected = contextRanges.compressible.length === 0 && contextRanges.protected.length > 0;
   const nothingToCompress = filterSuppressed || allProtected;
-  const shouldInject = nudgeAllowed && (!nothingToCompress || emergencyOverride);
+  let shouldInject = nudgeAllowed && (!nothingToCompress || emergencyOverride);
   if (nudgeAllowed && nothingToCompress && !emergencyOverride && currentTokens !== void 0) {
     state.nudges.lastPerMessageNudgeTokens = currentTokens;
     state.nudges.lastNudgeShownTokens = void 0;
   }
+  if (suffixMessage && !shouldInject) {
+    const tierUsage = getTierTokenUsage(state);
+    const tierChecks = [
+      { triggerTier: 2, targetTier: 1, tokens: tierUsage.tier1Tokens, lastNudge: state.nudges.lastTier2NudgeTokens },
+      { triggerTier: 3, targetTier: 2, tokens: tierUsage.tier2Tokens, lastNudge: state.nudges.lastTier3NudgeTokens }
+    ];
+    for (const tc of tierChecks) {
+      if (tc.tokens < nudgeGrowthTokens) continue;
+      const cadenceMet = tc.lastNudge === void 0 || currentTokens !== void 0 && currentTokens - tc.lastNudge >= growthFloor;
+      if (!cadenceMet) continue;
+      let candidates = [...state.prune.messages.activeBlockIds].map((id) => state.prune.messages.blocksById.get(id)).filter((b) => b !== void 0 && b.active && (b.tier ?? 1) === tc.targetTier).sort((a, b) => a.blockId - b.blockId);
+      if (candidates.length >= 2) {
+        const firstId = candidates[0].blockId;
+        const lastId = candidates[candidates.length - 1].blockId;
+        const nonTargetInIdRange = new Set(
+          [...state.prune.messages.activeBlockIds].map((id) => state.prune.messages.blocksById.get(id)).filter(
+            (b) => b !== void 0 && b.active && (b.tier ?? 1) !== tc.targetTier && b.blockId > firstId && b.blockId < lastId
+          ).map((b) => b.blockId)
+        );
+        if (nonTargetInIdRange.size > 0) {
+          let bestStart = 0;
+          let bestLen = 1;
+          let curStart = 0;
+          for (let i = 1; i < candidates.length; i++) {
+            const prevId = candidates[i - 1].blockId;
+            const currId = candidates[i].blockId;
+            let hasGap = false;
+            for (const nid of nonTargetInIdRange) {
+              if (nid > prevId && nid < currId) {
+                hasGap = true;
+                break;
+              }
+            }
+            if (hasGap) {
+              const curLen = i - curStart;
+              if (curLen > bestLen) {
+                bestLen = curLen;
+                bestStart = curStart;
+              }
+              curStart = i;
+            }
+          }
+          const finalLen = candidates.length - curStart;
+          if (finalLen > bestLen) {
+            bestLen = finalLen;
+            bestStart = curStart;
+          }
+          candidates = candidates.slice(bestStart, bestStart + bestLen);
+        }
+      }
+      if (candidates.length < 2) continue;
+      const rules = tc.triggerTier === 2 ? TIER2_DISTILL_RULES : TIER3_CONDENSE_RULES;
+      const candidateTokens = candidates.reduce((s, b) => s + b.summaryTokens, 0);
+      const firstBlock = candidates[0];
+      const lastBlock = candidates[candidates.length - 1];
+      const fmt = (n) => n >= 1e3 ? `${(n / 1e3).toFixed(1)}K` : String(n);
+      const sourceTier = tc.triggerTier === 2 ? "Tier 1" : "Tier 2";
+      const action = tc.triggerTier === 2 ? "Distill" : "Condense";
+      const blockList = candidates.slice(0, 10).map((b) => `b${b.blockId} (age=${b.survivedCount}, ${fmt(b.summaryTokens)}tok): "${b.topic}"`).join("\n");
+      const extraCount = candidates.length > 10 ? `
+...and ${candidates.length - 10} more` : "";
+      const tierText = `
+
+[Tier ${tc.triggerTier} Trigger] ${sourceTier} summaries accumulated (${fmt(candidateTokens)} tokens across ${candidates.length} blocks). ${action} them to free context.
+
+Target blocks (oldest first):
+${blockList}${extraCount}
+
+Compress range: \`content: [{ startId: "b${firstBlock.blockId}", endId: "b${lastBlock.blockId}", summary: "..." }]\`
+
+${rules}`;
+      appendToLastTextPart(suffixMessage, tierText);
+      shouldInject = true;
+      if (tc.triggerTier === 2) {
+        state.nudges.lastTier2NudgeTokens = currentTokens;
+      } else {
+        state.nudges.lastTier3NudgeTokens = currentTokens;
+      }
+      break;
+    }
+  }
   state.nudges.shouldInjectThisTurn = shouldInject;
   let tipsText = null;
   if (shouldInject) {
@@ -8559,7 +8926,10 @@ function buildTarget(blocks) {
     displayId: first.blockId,
     runId: first.runId,
     topic: grouped ? first.batchTopic || first.topic : first.topic,
-    compressedTokens: ordered.reduce((total, block) => total + block.compressedTokens, 0),
+    compressedTokens: ordered.reduce(
+      (total, block) => total + (block.effectiveCompressedTokens ?? block.compressedTokens),
+      0
+    ),
     durationMs: ordered.reduce((total, block) => Math.max(total, block.durationMs), 0),
     grouped,
     blocks: ordered
@@ -8727,17 +9097,25 @@ function snapshotActiveMessages(messagesState) {
   }
   return activeMessages;
 }
-function deactivateCompressionTarget(messagesState, target) {
+function deactivateCompressionTarget(messagesState, target, options) {
   const deactivatedAt = Date.now();
   for (const block of target.blocks) {
     block.active = false;
     block.deactivatedByUser = true;
     block.deactivatedAt = deactivatedAt;
     block.deactivatedByBlockId = void 0;
-    for (const consumedId of block.consumedBlockIds) {
-      const consumedBlock = messagesState.blocksById.get(consumedId);
-      if (consumedBlock) {
-        consumedBlock.deactivatedByUser = true;
+    if (options?.full) {
+      const visited = /* @__PURE__ */ new Set();
+      const queue = [...block.consumedBlockIds];
+      while (queue.length > 0) {
+        const consumedId = queue.shift();
+        if (visited.has(consumedId)) continue;
+        visited.add(consumedId);
+        const consumedBlock = messagesState.blocksById.get(consumedId);
+        if (consumedBlock) {
+          consumedBlock.deactivatedByUserDeep = true;
+          queue.push(...consumedBlock.consumedBlockIds);
+        }
       }
     }
   }
@@ -8953,14 +9331,18 @@ ARGUMENTS:
 IMPORTANT:
 - Decompressing inflates context. Check context usage before decompressing.
 - Message-mode blocks from the same batch (same runId) are restored together.
-- After decompression, the restored messages will appear in full in your next context window.
+- TIER-AWARE: by default, decompressing a multi-tier block restores the PREVIOUS tier's
+  summaries (e.g., decompress T2 \u2192 T1 summaries visible, not raw messages). Use full:true
+  to restore all the way to original messages (can be very expensive for T2/T3 blocks).
+- After decompression, the restored content will appear in full in your next context window.
 - Do NOT call this tool in parallel with compress \u2014 their state mutations may conflict.`;
 function buildSchema3() {
   return {
     blockId: tool4.schema.string().optional().describe('Block reference to decompress (e.g., "b0", "b2"). Mutually exclusive with startId/endId.'),
     startId: tool4.schema.string().optional().describe('Range start: message ref (e.g., "m00150") or block ref (e.g., "b2"). Used with endId.'),
     endId: tool4.schema.string().optional().describe('Range end: message ref (e.g., "m00200") or block ref (e.g., "b5"). Used with startId.'),
-    toFile: tool4.schema.string().optional().describe("If provided, writes restored content to this file path instead of inflating context. Block stays compressed. Path must be under /tmp or ~/.cache/opencode/. Example: '/tmp/block52.txt'")
+    toFile: tool4.schema.string().optional().describe("If provided, writes restored content to this file path instead of inflating context. Block stays compressed. Path must be under /tmp or ~/.cache/opencode/. Example: '/tmp/block52.txt'"),
+    full: tool4.schema.boolean().optional().describe("If true, restores ALL content down to original messages (multi-level decompress). Default: false \u2014 restores one tier up (e.g., decompressing a T2 block restores T1 summaries, not raw messages). Use full:true only when you need the exact original content and have context budget for it.")
   };
 }
 function extractMessageId(m) {
@@ -9030,7 +9412,7 @@ function createDecompressTool(factoryCtx) {
       const activeMessagesBefore = snapshotActiveMessages(messagesState);
       const activeBlockIdsBefore = new Set(messagesState.activeBlockIds);
       for (const target of targets) {
-        deactivateCompressionTarget(messagesState, target);
+        deactivateCompressionTarget(messagesState, target, { full: args.full === true });
       }
       syncCompressionBlocks(ctx.state, ctx.logger, rawMessages);
       const { restoredMessageCount, restoredTokens } = computeRestoredMessages(
@@ -9109,13 +9491,54 @@ function formatIdRange(block) {
   const count = block.effectiveMessageIds?.length || 0;
   return count > 0 ? `${count} msg${count !== 1 ? "s" : ""}` : "\u2014";
 }
+function getEffectiveCompressedTokens(block, blocksById, visited = /* @__PURE__ */ new Set()) {
+  if (block.effectiveCompressedTokens !== void 0) {
+    return block.effectiveCompressedTokens;
+  }
+  if (visited.has(block.blockId)) return 0;
+  visited.add(block.blockId);
+  let total = block.compressedTokens || 0;
+  for (const consumedId of block.consumedBlockIds || []) {
+    const consumed = blocksById.get(consumedId);
+    if (consumed) {
+      total += getEffectiveCompressedTokens(consumed, blocksById, visited);
+    }
+  }
+  return total;
+}
+function tierLabel(block) {
+  const tier = block.tier ?? 1;
+  return `T${tier}`;
+}
+function tierBreakdown(blocks) {
+  const tierTokens = {};
+  for (const b of blocks) {
+    const t = b.tier ?? 1;
+    tierTokens[t] = (tierTokens[t] || 0) + (b.summaryTokens || 0);
+  }
+  const tiers = Object.keys(tierTokens).map(Number);
+  if (tiers.length <= 1 && (!tierTokens[2] || tierTokens[2] === 0) && (!tierTokens[3] || tierTokens[3] === 0)) {
+    return null;
+  }
+  const parts = [];
+  for (const t of [1, 2, 3]) {
+    if (tierTokens[t]) {
+      parts.push(`T${t}: ${formatTokens(tierTokens[t])}`);
+    }
+  }
+  return parts.join(" | ");
+}
 function collectVisibleMessages(rawMessages, ctx) {
   const pruneMap = ctx.state.prune.messages.byMessageId;
   const byRawId = ctx.state.messageIds.byRawId;
   const result = [];
   let summaryTokens = 0;
+  const visibleMessageIds = new Set(rawMessages.map((m) => m.info.id));
   const activeBlocks = Array.from(ctx.state.prune.messages.activeBlockIds).map((id) => ctx.state.prune.messages.blocksById.get(id)).filter((b) => b !== void 0 && b.active);
   for (const block of activeBlocks) {
+    if (block.compressMessageId && !visibleMessageIds.has(block.compressMessageId)) {
+      continue;
+    }
     summaryTokens += block.summaryTokens || 0;
   }
   rawMessages.forEach((msg, idx) => {
@@ -9176,19 +9599,32 @@ function renderOverview(visibleMessages, summaryTokens, blocks, fetchFailed, raw
     lines.push("COMPRESSED BLOCKS");
     lines.push("  No compressed blocks.");
   } else {
+    const blocksById = ctx.state.prune.messages.blocksById;
     const totalSummary = blocks.reduce((s, b) => s + (b.summaryTokens || 0), 0);
-    const totalCompressed = blocks.reduce((s, b) => s + (b.compressedTokens || 0), 0);
-    lines.push(
-      `COMPRESSED BLOCKS \u2014 ${blocks.length} active (${formatTokens(totalSummary)} summary, ${formatTokens(totalCompressed)} original)`
+    const totalEffective = blocks.reduce(
+      (s, b) => s + getEffectiveCompressedTokens(b, blocksById),
+      0
     );
+    const header = `COMPRESSED BLOCKS \u2014 ${blocks.length} active (${formatTokens(totalSummary)} summary, ${formatTokens(totalEffective)} original)`;
+    lines.push(header);
+    const breakdown = tierBreakdown(blocks);
+    if (breakdown) {
+      lines.push(`  Tier usage: ${breakdown}`);
+    }
     lines.push("");
-    const sorted = [...blocks].sort((a, b) => b.createdAt - a.createdAt);
+    const sorted = [...blocks].sort((a, b) => {
+      const effA = getEffectiveCompressedTokens(a, blocksById);
+      const effB = getEffectiveCompressedTokens(b, blocksById);
+      return effB - effA || b.createdAt - a.createdAt;
+    });
     for (const b of sorted.slice(0, 30)) {
       const ageStr = formatAge(b.createdAt);
       const range = formatIdRange(b);
       const topic = b.topic || "(no topic)";
+      const tier = tierLabel(b);
+      const effTokens = getEffectiveCompressedTokens(b, blocksById);
       lines.push(
-        `  b${b.blockId}  ${formatTokens(b.compressedTokens)}\u2192${formatTokens(b.summaryTokens)}  ${ageStr}  ${range}  "${topic}"`
+        `  b${b.blockId} (${tier})  ${formatTokens(effTokens)}\u2192${formatTokens(b.summaryTokens)}  ${ageStr}  ${range}  "${topic}"`
       );
     }
   }
@@ -9292,7 +9728,7 @@ function renderUncompressedDrilldown(visibleMessages, toolFilter, sort, limit) {
   }
   return lines;
 }
-function renderCompressedDrilldown(blocks, sort, limit) {
+function renderCompressedDrilldown(blocks, sort, limit, blocksById) {
   const lines = [];
   let sorted = [...blocks];
   if (sort === "time") {
@@ -9300,13 +9736,22 @@ function renderCompressedDrilldown(blocks, sort, limit) {
   } else if (sort === "age") {
     sorted.sort((a, b) => (b.survivedCount || 0) - (a.survivedCount || 0));
   } else {
-    sorted.sort((a, b) => (b.compressedTokens || 0) - (a.compressedTokens || 0));
+    sorted.sort(
+      (a, b) => getEffectiveCompressedTokens(b, blocksById) - getEffectiveCompressedTokens(a, blocksById) || b.createdAt - a.createdAt
+    );
   }
   const totalSummary = sorted.reduce((s, b) => s + (b.summaryTokens || 0), 0);
-  const totalCompressed = sorted.reduce((s, b) => s + (b.compressedTokens || 0), 0);
+  const totalEffective = sorted.reduce(
+    (s, b) => s + getEffectiveCompressedTokens(b, blocksById),
+    0
+  );
   lines.push(
-    `COMPRESSED \u2014 ${sorted.length} blocks | ${formatTokens(totalCompressed)} original \u2192 ${formatTokens(totalSummary)} summary`
+    `COMPRESSED \u2014 ${sorted.length} blocks | ${formatTokens(totalEffective)} original \u2192 ${formatTokens(totalSummary)} summary`
   );
+  const breakdown = tierBreakdown(sorted);
+  if (breakdown) {
+    lines.push(`Tier usage: ${breakdown}`);
+  }
   lines.push(`Sorted by ${sort === "time" ? "time" : sort === "age" ? "age" : "size"}`);
   lines.push("");
   const shown = sorted.slice(0, limit);
@@ -9314,10 +9759,12 @@ function renderCompressedDrilldown(blocks, sort, limit) {
     const survived = b.survivedCount ?? 0;
     const gen = b.generation ?? "young";
     const effCount = b.effectiveMessageIds?.length ?? 0;
-    const consumed = b.consumedBlockIds && b.consumedBlockIds.length > 0 ? ` nested=[${b.consumedBlockIds.map((n) => `b${n}`).join(",")}]` : "";
+    const consumed = b.includedBlockIds && b.includedBlockIds.length > 0 ? ` nested=[${b.includedBlockIds.map((n) => `b${n}`).join(",")}]` : "";
     const topic = b.topic || "(no topic)";
+    const tier = tierLabel(b);
+    const effTokens = getEffectiveCompressedTokens(b, blocksById);
     lines.push(
-      `  b${b.blockId}  ${formatTokens(b.compressedTokens)}\u2192${formatTokens(b.summaryTokens)}  ${formatAge(b.createdAt)}  ${formatIdRange(b)}  age=${survived} ${gen} eff=${effCount}${consumed}`
+      `  b${b.blockId} (${tier})  ${formatTokens(effTokens)}\u2192${formatTokens(b.summaryTokens)}  ${formatAge(b.createdAt)}  ${formatIdRange(b)}  age=${survived} ${gen} eff=${effCount}${consumed}`
     );
     lines.push(`    "${topic}"`);
   }
@@ -9358,7 +9805,14 @@ function createAcpStatusTool(factoryCtx) {
       const allBlocks = activeIds.map((id) => msgState.blocksById.get(id)).filter((b) => b !== void 0 && b.active);
       const lines = [];
       if (scope === "compressed") {
-        lines.push(...renderCompressedDrilldown(allBlocks, sort, limit));
+        lines.push(
+          ...renderCompressedDrilldown(
+            allBlocks,
+            sort,
+            limit,
+            msgState.blocksById
+          )
+        );
         return lines.join("\n");
       }
       let visibleMsgs = [];
@@ -9517,6 +9971,53 @@ IMPORTANT: This was an automatic context pruning. You MUST continue your previou
   });
 }
 
+// lib/compress/hide-consumed.ts
+function hideConsumedCompressCalls(state, messages) {
+  if (state.prune.messages.blocksById.size === 0) {
+    return 0;
+  }
+  const consumedMessageIds = /* @__PURE__ */ new Set();
+  for (const block of state.prune.messages.blocksById.values()) {
+    if (block.active) continue;
+    if (block.deactivatedByUser) continue;
+    if (block.deactivatedByUserDeep) continue;
+    if (block.deactivatedByBlockId === void 0) continue;
+    if (!block.compressMessageId) continue;
+    consumedMessageIds.add(block.compressMessageId);
+  }
+  if (consumedMessageIds.size === 0) {
+    return 0;
+  }
+  const lastUserIdx = messages.findLastIndex(
+    (m) => m.info.role === "user" && !isIgnoredUserMessage(m)
+  );
+  let hidden = 0;
+  for (let i = 0; i < messages.length; i++) {
+    if (lastUserIdx >= 0 && i >= lastUserIdx) break;
+    const msg = messages[i];
+    if (!consumedMessageIds.has(msg.info.id)) continue;
+    const parts = Array.isArray(msg.parts) ? msg.parts : [];
+    let changed = false;
+    const remaining = parts.filter((p) => {
+      if (p.type === "tool" && p.tool === "compress") {
+        hidden++;
+        changed = true;
+        return false;
+      }
+      return true;
+    });
+    if (changed) {
+      if (remaining.length > 0) {
+        messages[i] = { ...msg, parts: remaining };
+      } else {
+        messages.splice(i, 1);
+        i--;
+      }
+    }
+  }
+  return hidden;
+}
+
 // lib/logger.ts
 import { writeFile as writeFile2, mkdir as mkdir2 } from "fs/promises";
 import { join as join3 } from "path";
@@ -9745,7 +10246,7 @@ TOOLS
 You have five context-management tools:
 
 - \`compress\` \u2014 Replace a contiguous range of older conversation with a single detailed summary you write. Use when content is genuinely consumed (no longer needed for the current task step). Single range: \`compress({ topic: "API exploration", content: [{ startId: "m00150", endId: "m00220", summary: "..." }] })\`. Batch (multiple unrelated ranges, each with its own topic): \`compress({ content: [{ topic: "Auth", startId: "m00150", endId: "m00220", summary: "..." }, { topic: "Deploy", startId: "m00300", endId: "m00350", summary: "..." }] })\`.
-- \`decompress\` \u2014 Restore a previously compressed block's full original content, optionally to a file for large blocks. Use when a summary lacks the exact detail you need. Example: \`decompress({ blockId: "b5" })\` or \`decompress({ blockId: "b5", toFile: "path" })\`.
+- \`decompress\` \u2014 Restore a previously compressed block's content. By default restores one tier up (T2\u2192T1 summaries, not raw messages). Use \`full: true\` to restore all the way to original messages. Use \`toFile\` to write to file instead of inflating context. Example: \`decompress({ blockId: "b5" })\` or \`decompress({ blockId: "b5", toFile: "path" })\` or \`decompress({ blockId: "b5", full: true })\`.
 - \`search_context\` \u2014 Search compressed block summaries (and optionally visible messages) by keyword. Use BEFORE decompressing to find the right block. Example: \`search_context({ query: "auth token refresh" })\`.
 - \`prune\` \u2014 Remove old tool outputs by tool type, keeping only recent calls. Unlike compress (which creates summaries), prune directly strips outputs. Use for disposable outputs like old todowrite states or edit echoes. Example: \`prune({ toolType: "todowrite", keepLatest: 3 })\`.
 - \`acp_status\` \u2014 Context status with compressible ranges. No args = overview + ranges. \`scope:"uncompressed"\` for range view; add \`view:"messages"\` for per-message listing with \`tool\`/\`sort\` filters. \`scope:"compressed"\` for block details.
@@ -9779,6 +10280,16 @@ WHEN NOT TO COMPRESS
 
 ${HOW_TO_COMPRESS_RULES}
 
+MULTI-TIER COMPRESSION
+
+Summaries accumulate as the session grows. When tier-1 summaries pile up, the system injects a [Tier 2 Trigger] prompting you to DISTILL old blocks into a single tier-2 summary. If tier-2 summaries also accumulate, a [Tier 3 Trigger] asks you to CONDENSE them further.
+
+- Tier 1 (default): Full-detail compression of conversation ranges. Uses HOW TO COMPRESS rules above.
+- Tier 2: Distillation of old tier-1 block summaries. Uses TIER 2 DISTILLATION rules (decisions/outcomes only, drop paths/code/process).
+- Tier 3: Ultra-condensation of tier-2 summaries. Uses TIER 3 CONDENSATION rules (bare facts, 1-3 lines per block).
+
+To compress blocks: use block IDs as boundaries: \`compress({ content: [{ startId: "b3", endId: "b15", summary: "..." }] })\`. This deactivates the consumed blocks and creates a new higher-tier block. The system prompt at the trigger tells you which rules to follow.
+
 PERIODIC CONTEXT STATUS
 
 Periodically, as context grows, the system appends a short status line in a synthetic suffix message. It looks like:
@@ -10956,8 +11467,21 @@ async function handleRecompressCommand(ctx) {
   const activeBlockIdsBefore = new Set(messagesState.activeBlockIds);
   for (const block of target.blocks) {
     block.deactivatedByUser = false;
+    block.deactivatedByUserDeep = false;
     block.deactivatedAt = void 0;
     block.deactivatedByBlockId = void 0;
+    const queue = [...block.consumedBlockIds];
+    const visited = /* @__PURE__ */ new Set();
+    while (queue.length > 0) {
+      const consumedId = queue.shift();
+      if (visited.has(consumedId)) continue;
+      visited.add(consumedId);
+      const consumedBlock = messagesState.blocksById.get(consumedId);
+      if (consumedBlock) {
+        consumedBlock.deactivatedByUserDeep = false;
+        queue.push(...consumedBlock.consumedBlockIds);
+      }
+    }
   }
   syncCompressionBlocks(state, logger, messages);
   let recompressedMessageCount = 0;
@@ -11044,8 +11568,9 @@ function formatCompressionTime(ms) {
 async function handleStatsCommand(ctx) {
   const { client, state, logger, sessionId, messages } = ctx;
   const sessionTokens = state.stats.totalPruneTokens;
+  const visibleMessageIds = new Set(messages.map((m) => m.info.id));
   const sessionSummaryTokens = Array.from(state.prune.messages.blocksById.values()).reduce(
-    (total, block) => block.active ? total + block.summaryTokens : total,
+    (total, block) => block.active && visibleMessageIds.has(block.compressMessageId ?? "") ? total + block.summaryTokens : total,
     0
   );
   const sessionDurationMs = getActiveCompressionTargets(state.prune.messages).reduce(
@@ -11623,6 +12148,7 @@ function createChatMessageTransformHandler(client, registry3, logger, config, pr
     }
     const prePruneTokens = getCurrentTokenUsage(state, output.messages);
     prune(state, logger, config, output.messages);
+    hideConsumedCompressCalls(state, output.messages);
     assignMessageRefs(state, output.messages);
     const compressionPriorities = buildPriorityMap(config, state, output.messages);
     prompts.reload();