opencode-acp 1.14.25-pr.350.76 → 1.14.25-pr.351.83

package/README.md

@@ -324,11 +324,12 @@ Each level overrides the previous, so project settings take priority over global
         // Soft upper threshold: above this, ACP keeps injecting strong
         // compression nudges (based on nudgeFrequency), so compression is
         // much more likely. Accepts: number or "X%" of model context window.
-        "maxContextLimit": "55%",
-        // Soft lower threshold for reminder nudges: below this, turn/iteration
-        // reminders are off (compression less likely). At/above this, reminders
-        // are on. Accepts: number or "X%" of model context window.
-        "minContextLimit": "45%",
+        "maxContextLimit": "80%",
+        // Soft lower threshold for turn/iteration reminder nudges: below this,
+        // those reminders are off (compression less likely). At/above this, they
+        // are on. Growth nudges have their own floor: minNudgeContextPercent.
+        // Accepts: number or "X%" of model context window.
+        "minContextLimit": "80%",
         // Optional per-model override for maxContextLimit by providerID/modelID.
         // If present, this wins over the global maxContextLimit.
         // Accepts: number or "X%".

package/dist/index.js

@@ -888,6 +888,7 @@ var VALID_CONFIG_KEYS = /* @__PURE__ */ new Set([
   "compress.minContextLimit",
   "compress.modelMaxLimits",
   "compress.modelMinLimits",
+  "compress.providers",
   "compress.nudgeFrequency",
   "compress.minNudgeContextPercent",
   "compress.nudgeGrowthTokens",
@@ -908,7 +909,6 @@ var VALID_CONFIG_KEYS = /* @__PURE__ */ new Set([
   "compress.preserveRecentMessages",
   "compress.preserveRecentTokens",
   "compress.preserveLastUserMessage",
-  "compress.completionReserveTokens",
   "gc",
   "gc.algorithm",
   "gc.promotionThreshold",
@@ -932,7 +932,7 @@ function getConfigKeyPaths(obj, prefix = "") {
   for (const key of Object.keys(obj)) {
     const fullKey = prefix ? `${prefix}.${key}` : key;
     keys.push(fullKey);
-    if (fullKey === "compress.modelMaxLimits" || fullKey === "compress.modelMinLimits" || fullKey === "messageFilters.filters") {
+    if (fullKey === "compress.modelMaxLimits" || fullKey === "compress.modelMinLimits" || fullKey === "compress.providers" || fullKey === "messageFilters.filters") {
       continue;
     }
     if (obj[key] && typeof obj[key] === "object" && !Array.isArray(obj[key])) {
@@ -1262,20 +1262,6 @@ function validateConfigTypes(config) {
           actual: typeof compress.preserveLastUserMessage
         });
       }
-      if (compress.completionReserveTokens !== void 0 && typeof compress.completionReserveTokens !== "number") {
-        errors.push({
-          key: "compress.completionReserveTokens",
-          expected: "number",
-          actual: typeof compress.completionReserveTokens
-        });
-      }
-      if (typeof compress.completionReserveTokens === "number" && compress.completionReserveTokens < 0) {
-        errors.push({
-          key: "compress.completionReserveTokens",
-          expected: "non-negative number (>= 0)",
-          actual: `${compress.completionReserveTokens}`
-        });
-      }
       if (typeof compress.iterationNudgeThreshold === "number" && compress.iterationNudgeThreshold < 1) {
         errors.push({
           key: "compress.iterationNudgeThreshold",
@@ -1326,6 +1312,85 @@ function validateConfigTypes(config) {
       }
       validateModelLimits("compress.modelMaxLimits", compress.modelMaxLimits);
       validateModelLimits("compress.modelMinLimits", compress.modelMinLimits);
+      const validatePercentField = (key, value) => {
+        if (value === void 0) {
+          return;
+        }
+        if (typeof value !== "number" || !Number.isFinite(value) || value < 0 || value > 100) {
+          errors.push({
+            key,
+            expected: "number (0-100)",
+            actual: JSON.stringify(value)
+          });
+        }
+      };
+      const validateModelOverrides = (prefix, overrides) => {
+        if (overrides === void 0) {
+          return;
+        }
+        if (typeof overrides !== "object" || overrides === null || Array.isArray(overrides)) {
+          errors.push({ key: prefix, expected: "CompressModelOverrides", actual: typeof overrides });
+          return;
+        }
+        const model = overrides;
+        for (const field of Object.keys(model)) {
+          if (field !== "minNudgeContextPercent") {
+            errors.push({
+              key: `${prefix}.${field}`,
+              expected: "known override field (minNudgeContextPercent)",
+              actual: "unknown field"
+            });
+          }
+        }
+        validatePercentField(`${prefix}.minNudgeContextPercent`, model.minNudgeContextPercent);
+      };
+      const validateProviderOverrides = (providers) => {
+        if (providers === void 0) {
+          return;
+        }
+        if (typeof providers !== "object" || providers === null || Array.isArray(providers)) {
+          errors.push({
+            key: "compress.providers",
+            expected: "Record<string, ProviderOverrides>",
+            actual: typeof providers
+          });
+          return;
+        }
+        for (const [providerId, providerValue] of Object.entries(providers)) {
+          const prefix = `compress.providers.${providerId}`;
+          if (typeof providerValue !== "object" || providerValue === null || Array.isArray(providerValue)) {
+            errors.push({ key: prefix, expected: "ProviderOverrides", actual: typeof providerValue });
+            continue;
+          }
+          const provider = providerValue;
+          for (const field of Object.keys(provider)) {
+            if (field !== "minNudgeContextPercent" && field !== "models") {
+              errors.push({
+                key: `${prefix}.${field}`,
+                expected: "known override field (minNudgeContextPercent, models)",
+                actual: "unknown field"
+              });
+            }
+          }
+          validatePercentField(`${prefix}.minNudgeContextPercent`, provider.minNudgeContextPercent);
+          const models = provider.models;
+          if (models === void 0) {
+            continue;
+          }
+          if (typeof models !== "object" || models === null || Array.isArray(models)) {
+            errors.push({
+              key: `${prefix}.models`,
+              expected: "Record<string, ModelOverrides>",
+              actual: typeof models
+            });
+            continue;
+          }
+          for (const [modelId, modelValue] of Object.entries(models)) {
+            validateModelOverrides(`${prefix}.models.${modelId}`, modelValue);
+          }
+        }
+      };
+      validateProviderOverrides(compress.providers);
       const validValues = ["ask", "allow", "deny"];
       if (compress.permission !== void 0 && !validValues.includes(compress.permission)) {
         errors.push({
@@ -1504,7 +1569,7 @@ var defaultConfig = {
     maxContextLimit: "80%",
     minContextLimit: "80%",
     nudgeFrequency: 5,
-    minNudgeContextPercent: 15,
+    minNudgeContextPercent: 5,
     iterationNudgeThreshold: 15,
     nudgeForce: "soft",
     protectedTools: [...COMPRESS_DEFAULT_PROTECTED_TOOLS],
@@ -1626,6 +1691,39 @@ function loadConfigFile(configPath) {
     return { data: null, parseError: error.message || "Failed to parse config" };
   }
 }
+function stripUndefined(value) {
+  const result = {};
+  for (const [key, entry] of Object.entries(value)) {
+    if (entry !== void 0) {
+      result[key] = entry;
+    }
+  }
+  return result;
+}
+function mergeModelOverrides(base, override) {
+  if (base === void 0) return override;
+  if (override === void 0) return base;
+  const merged = { ...base };
+  for (const [modelId, model] of Object.entries(override)) {
+    merged[modelId] = { ...base[modelId], ...stripUndefined(model) };
+  }
+  return merged;
+}
+function mergeProviderOverrides(base, override) {
+  if (base === void 0) return override;
+  if (override === void 0) return base;
+  const merged = { ...base };
+  for (const [providerId, provider] of Object.entries(override)) {
+    const baseProvider = base[providerId];
+    const mergedModels = mergeModelOverrides(baseProvider?.models, provider.models);
+    merged[providerId] = {
+      ...baseProvider,
+      ...stripUndefined(provider),
+      ...mergedModels !== void 0 ? { models: mergedModels } : {}
+    };
+  }
+  return merged;
+}
 function mergeCompress(base, override) {
   if (!override) {
     return base;
@@ -1638,6 +1736,7 @@ function mergeCompress(base, override) {
     minContextLimit: override.minContextLimit ?? base.minContextLimit,
     modelMaxLimits: override.modelMaxLimits ?? base.modelMaxLimits,
     modelMinLimits: override.modelMinLimits ?? base.modelMinLimits,
+    providers: mergeProviderOverrides(base.providers, override.providers),
     nudgeFrequency: override.nudgeFrequency ?? base.nudgeFrequency,
     minNudgeContextPercent: override.minNudgeContextPercent ?? base.minNudgeContextPercent,
     nudgeGrowthTokens: override.nudgeGrowthTokens,
@@ -1657,8 +1756,7 @@ function mergeCompress(base, override) {
     lastSegmentSoftBlock: override.lastSegmentSoftBlock ?? base.lastSegmentSoftBlock,
     preserveRecentMessages: override.preserveRecentMessages ?? base.preserveRecentMessages,
     preserveRecentTokens: override.preserveRecentTokens ?? base.preserveRecentTokens,
-    preserveLastUserMessage: override.preserveLastUserMessage ?? base.preserveLastUserMessage,
-    completionReserveTokens: override.completionReserveTokens ?? base.completionReserveTokens
+    preserveLastUserMessage: override.preserveLastUserMessage ?? base.preserveLastUserMessage
   };
 }
 function mergeCommands(base, override) {
@@ -1689,6 +1787,22 @@ function deepCloneConfig(config) {
       ...config.compress,
       modelMaxLimits: { ...config.compress.modelMaxLimits },
       modelMinLimits: { ...config.compress.modelMinLimits },
+      providers: config.compress.providers ? Object.fromEntries(
+        Object.entries(config.compress.providers).map(([providerId, provider]) => [
+          providerId,
+          {
+            ...provider,
+            ...provider.models ? {
+              models: Object.fromEntries(
+                Object.entries(provider.models).map(([modelId, model]) => [
+                  modelId,
+                  { ...model }
+                ])
+              )
+            } : {}
+          }
+        ])
+      ) : void 0,
       protectedTools: [...config.compress.protectedTools]
     },
     gc: {
@@ -4317,8 +4431,7 @@ function createSessionState() {
     modelProviderID: void 0,
     modelID: void 0,
     systemPromptTokens: void 0,
-    qualityGateRetryPending: false,
-    noContextLimitWarned: false
+    qualityGateRetryPending: false
   };
 }
 function resetSessionState(state) {
@@ -4360,7 +4473,6 @@ function resetSessionState(state) {
   state.modelID = void 0;
   state.systemPromptTokens = void 0;
   state.qualityGateRetryPending = false;
-  state.noContextLimitWarned = false;
 }
 async function ensureSessionInitialized(client, state, sessionId, logger, messages, config) {
   if (state.sessionId === sessionId) {
@@ -6975,6 +7087,26 @@ function computeShouldNudge2(params) {
   return policy.computeShouldNudge(params);
 }
 var DEFAULT_NUDGE_GROWTH_TOKENS = 5e4;
+var DEFAULT_MIN_NUDGE_CONTEXT_PERCENT = 5;
+function resolveMinNudgeContextPercent(config, providerId, modelId) {
+  const modelOverrides = providerId !== void 0 && modelId !== void 0 ? config.compress?.providers?.[providerId]?.models?.[modelId] : void 0;
+  if (modelOverrides?.minNudgeContextPercent !== void 0) {
+    return modelOverrides.minNudgeContextPercent;
+  }
+  const providerOverrides = providerId !== void 0 ? config.compress?.providers?.[providerId] : void 0;
+  if (providerOverrides?.minNudgeContextPercent !== void 0) {
+    return providerOverrides.minNudgeContextPercent;
+  }
+  return config.compress?.minNudgeContextPercent;
+}
+function resolveMinNudgeFloorTokens(config, modelContextLimit, providerId, modelId) {
+  if (modelContextLimit === void 0) {
+    return void 0;
+  }
+  const percent = resolveMinNudgeContextPercent(config, providerId, modelId) ?? DEFAULT_MIN_NUDGE_CONTEXT_PERCENT;
+  const clamped = Math.min(100, Math.max(0, percent));
+  return Math.round(clamped / 100 * modelContextLimit);
+}
 function addAnchor(anchorMessageIds, anchorMessageId, anchorMessageIndex, messages, interval) {
   if (anchorMessageIndex < 0) {
     return false;
@@ -7657,11 +7789,18 @@ var injectCompressNudges = (state, config, logger, messages, prompts, compressio
     overMinLimit,
     overMaxLimit,
     lastNudgeTokens: growthReference,
-    minNudgeContextPercent: config.compress?.minNudgeContextPercent ?? 15,
+    minNudgeContextPercent: resolveMinNudgeContextPercent(config, providerId, modelId) ?? DEFAULT_MIN_NUDGE_CONTEXT_PERCENT,
     nudgeGrowthTokens: effectiveThreshold
   });
   const growthSinceBaseline = currentTokens !== void 0 && growthReference !== void 0 ? currentTokens - growthReference : void 0;
-  const nudgeAllowed = emergencyOverride || decision.shouldNudge && growthSinceBaseline !== void 0 && growthSinceBaseline >= growthFloor;
+  const minNudgeFloorTokens = resolveMinNudgeFloorTokens(
+    config,
+    modelContextLimit,
+    providerId,
+    modelId
+  );
+  const overMinNudgeFloor = minNudgeFloorTokens === void 0 || currentTokens === void 0 || currentTokens >= minNudgeFloorTokens;
+  const nudgeAllowed = emergencyOverride || decision.shouldNudge && (overMaxLimit || overMinNudgeFloor) && growthSinceBaseline !== void 0 && growthSinceBaseline >= growthFloor;
   const effectiveTipsVariant = emergencyOverride ? "maxLimit" : decision.tipsVariant;
   if (state.nudges.lastPerMessageNudgeTokens === void 0 && currentTokens !== void 0) {
     state.nudges.lastPerMessageNudgeTokens = currentTokens;
@@ -9128,7 +9267,7 @@ 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 ? "1.14.25-pr.350.76" : "dev";
+var LOG_VERSION = true ? "1.14.25-pr.351.83" : "dev";
 var LEVEL_RANK = {
   debug: 10,
   info: 20,
@@ -10013,153 +10152,6 @@ function truncateLargeToolOutputs(state, config, logger, messages) {
   }
 }
 
-// lib/messages/enforce-budget.ts
-var DEFAULT_COMPLETION_RESERVE_TOKENS = 32768;
-var TRUNCATION_MARKER2 = "[truncated for context space";
-var KEEP_PREFIX_CHARS2 = 2e3;
-var KEEP_SUFFIX_CHARS2 = 2e3;
-var PROTECT_RECENT_MESSAGES2 = 3;
-var MIN_CLEAR_TOKENS = 200;
-function resolveContextWindow(state) {
-  if (typeof state.modelContextLimit === "number" && state.modelContextLimit > 0) {
-    return state.modelContextLimit;
-  }
-  return void 0;
-}
-function estimateWireTokens(state, messages) {
-  const base = getCurrentTokenUsage(state, messages);
-  if (base > 0) {
-    let baseAssistant = -1;
-    for (let i = messages.length - 1; i >= 0; i--) {
-      if (messages[i].info.role !== "assistant") continue;
-      const tokens = messages[i].info.tokens;
-      if ((tokens?.input || 0) <= 0 && (tokens?.output || 0) <= 0) continue;
-      baseAssistant = i;
-      break;
-    }
-    if (baseAssistant >= 0) {
-      let additions = 0;
-      for (let i = baseAssistant + 1; i < messages.length; i++) {
-        additions += countAllMessageTokens(messages[i]);
-      }
-      return base + additions;
-    }
-  }
-  let total = 0;
-  for (const m of messages) total += countAllMessageTokens(m);
-  return total + (state.systemPromptTokens ?? 0);
-}
-function enforceContextBudget(state, config, logger, messages) {
-  const window = resolveContextWindow(state);
-  if (window === void 0) return void 0;
-  const configuredReserve = config.compress?.completionReserveTokens;
-  const reserve = typeof configuredReserve === "number" && Number.isFinite(configuredReserve) && configuredReserve >= 0 ? configuredReserve : DEFAULT_COMPLETION_RESERVE_TOKENS;
-  const budget = window - reserve;
-  if (budget <= 0) return void 0;
-  const estimatedTokens = estimateWireTokens(state, messages);
-  if (estimatedTokens <= budget) {
-    return {
-      applied: false,
-      window,
-      reserve,
-      budget,
-      estimatedTokens,
-      finalEstimate: estimatedTokens,
-      truncatedCount: 0,
-      clearedCount: 0
-    };
-  }
-  const protectedIndex = messages.length - PROTECT_RECENT_MESSAGES2;
-  const protectedTools = new Set(config.compress?.protectedTools ?? []);
-  const candidates = [];
-  for (let mi = 0; mi < protectedIndex; mi++) {
-    if (mi === 0 && messages[mi].info.role === "user") continue;
-    const msg = messages[mi];
-    const parts = Array.isArray(msg.parts) ? msg.parts : [];
-    for (const part of parts) {
-      if (part?.type !== "tool") continue;
-      if (part.state?.status !== "completed") continue;
-      if (part.tool === "compress") continue;
-      if (protectedTools.has(part.tool)) continue;
-      const content = extractCompletedToolOutput(part);
-      if (content === void 0) continue;
-      if (content === COMPACTED_TOOL_OUTPUT_PLACEHOLDER) continue;
-      const tokens = countTokens2(content);
-      if (tokens <= 0) continue;
-      candidates.push({ part, content, tokens, index: mi });
-    }
-  }
-  let saved = 0;
-  let truncatedCount = 0;
-  let clearedCount = 0;
-  const truncatable = candidates.filter(
-    (c) => !c.content.includes(TRUNCATION_MARKER2) && c.content.length > KEEP_PREFIX_CHARS2 + KEEP_SUFFIX_CHARS2
-  ).sort((a, b) => b.tokens - a.tokens);
-  for (const c of truncatable) {
-    if (estimatedTokens - saved <= budget) break;
-    const prefix = c.content.slice(0, KEEP_PREFIX_CHARS2);
-    const suffix = c.content.slice(-KEEP_SUFFIX_CHARS2);
-    const truncated = prefix + `
-
-...${TRUNCATION_MARKER2} \u2014 original ~${c.tokens} tokens]...
-
-` + suffix;
-    if (truncated.length >= c.content.length) continue;
-    c.part.state.output = truncated;
-    saved += c.tokens - countTokens2(truncated);
-    truncatedCount++;
-  }
-  if (estimatedTokens - saved > budget) {
-    const clearable = candidates.filter((c) => {
-      const out = extractCompletedToolOutput(c.part);
-      return out !== void 0 && out !== COMPACTED_TOOL_OUTPUT_PLACEHOLDER && countTokens2(out) > MIN_CLEAR_TOKENS;
-    }).sort((a, b) => a.index - b.index);
-    for (const c of clearable) {
-      if (estimatedTokens - saved <= budget) break;
-      const current = extractCompletedToolOutput(c.part);
-      if (current === void 0 || current === COMPACTED_TOOL_OUTPUT_PLACEHOLDER) continue;
-      c.part.state.output = COMPACTED_TOOL_OUTPUT_PLACEHOLDER;
-      saved += countTokens2(current) - countTokens2(COMPACTED_TOOL_OUTPUT_PLACEHOLDER);
-      clearedCount++;
-    }
-  }
-  const finalEstimate = Math.max(0, estimatedTokens - saved);
-  if (truncatedCount > 0 || clearedCount > 0) {
-    logger.warn("Context budget guard: pruned tool outputs to fit the request", {
-      session: state.sessionId,
-      estimatedTokens: Math.round(estimatedTokens),
-      budget,
-      window,
-      reserve,
-      truncatedCount,
-      clearedCount,
-      estimatedSavedTokens: Math.round(saved),
-      finalEstimate: Math.round(finalEstimate)
-    });
-  }
-  if (finalEstimate > budget) {
-    logger.warn(
-      "Context budget guard: still over budget after pruning all compressible tool outputs; the request may be rejected by the model",
-      {
-        session: state.sessionId,
-        finalEstimate: Math.round(finalEstimate),
-        budget,
-        window
-      }
-    );
-  }
-  return {
-    applied: truncatedCount > 0 || clearedCount > 0,
-    window,
-    reserve,
-    budget,
-    estimatedTokens,
-    finalEstimate,
-    truncatedCount,
-    clearedCount
-  };
-}
-
 // lib/commands/context.ts
 function analyzeTokens(state, messages) {
   const breakdown = {
@@ -11261,16 +11253,6 @@ function createChatMessageTransformHandler(client, registry4, logger, config, pr
         });
       }
       await updatePerTurnState(state, logger, messages);
-      if (state.modelContextLimit === void 0 && !state.noContextLimitWarned && requestModel?.providerID && requestModel?.modelID && registry4.resolveModelLimit(requestModel.providerID, requestModel.modelID) === void 0) {
-        state.noContextLimitWarned = true;
-        logger.warn(
-          'Model reports no context window and the catalog has no entry for it; all percentage thresholds (min/max/emergency, GC) and the context-budget guard are disabled. Set the model limit in opencode.json (e.g. "limit": {"context": 262144, "output": 16384}) to enable them (also fixes the 32000 max_tokens fallback); an absolute compress.maxContextLimit in acp.jsonc only enables proactive nudges, not the guard.',
-          {
-            session: state.sessionId,
-            model: `${requestModel.providerID}/${requestModel.modelID}`
-          }
-        );
-      }
     }
     syncCompressPermissionState(state, config, hostPermissions, output.messages);
     if (state.isSubAgent && !config.allowSubAgents) {
@@ -11301,7 +11283,6 @@ function createChatMessageTransformHandler(client, registry4, logger, config, pr
     const prePruneTokens = getCurrentTokenUsage(state, output.messages);
     prune(state, logger, config, output.messages);
     truncateLargeToolOutputs(state, config, logger, output.messages);
-    enforceContextBudget(state, config, logger, output.messages);
     hideConsumedCompressCalls(state, output.messages);
     assignMessageRefs(state, output.messages);
     const compressionPriorities = buildPriorityMap(config, state, output.messages);
@@ -11727,7 +11708,7 @@ var server = (async (ctx) => {
   }
   const logger = new Logger(config.debug, config.debug ? "debug" : config.logLevel);
   logger.info("ACP plugin initialized", {
-    version: true ? "1.14.25-pr.350.76" : "dev",
+    version: true ? "1.14.25-pr.351.83" : "dev",
     workspace: ctx.directory,
     logLevel: logger.level,
     debug: config.debug,