@link-assistant/hive-mind 1.46.9 → 1.47.1

package/CHANGELOG.md

@@ -1,5 +1,22 @@
 # @link-assistant/hive-mind
 
+## 1.47.1
+
+### Patch Changes
+
+- 3bbd66e: Improve Context and tokens usage output format: move percentage before unit label, parenthesize cached tokens in Total line, use consistent X / Y (Z%) format for output tokens when limit is known, and show sub-sessions under model heading instead of globally
+
+## 1.47.0
+
+### Minor Changes
+
+- 7997308: feat: update free models for --tool agent, set qwen3.6-plus-free as default (#1543)
+  - Change default agent model from `minimax-m2.5-free` to `qwen3.6-plus-free` (~1M context window)
+  - Add `qwen3.6-plus-free` (Alibaba Qwen, ~1M context) to free models
+  - Add `nemotron-3-super-free` (NVIDIA hybrid Mamba-Transformer, ~262K context) to free models
+  - Update documentation, tests, and provider priority lists
+  - Syncs with upstream agent PR #234
+
 ## 1.46.9
 
 ### Patch Changes

package/README.md

@@ -445,10 +445,11 @@ Examples:
 /solve https://github.com/owner/repo/issues/123 --model opus --think max
 
 Free Models (with --tool agent):
+/solve https://github.com/owner/repo/issues/123 --tool agent --model qwen3.6-plus-free
+/solve https://github.com/owner/repo/issues/123 --tool agent --model opencode/qwen3.6-plus-free
+/solve https://github.com/owner/repo/issues/123 --tool agent --model nemotron-3-super-free
 /solve https://github.com/owner/repo/issues/123 --tool agent --model minimax-m2.5-free
-/solve https://github.com/owner/repo/issues/123 --tool agent --model opencode/minimax-m2.5-free
 /solve https://github.com/owner/repo/issues/123 --tool agent --model gpt-5-nano
-/solve https://github.com/owner/repo/issues/123 --tool agent --model big-pickle
 
 Free Models via Kilo Gateway (with --tool agent):
 /solve https://github.com/owner/repo/issues/123 --tool agent --model kilo/glm-5-free

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@link-assistant/hive-mind",
-  "version": "1.46.9",
+  "version": "1.47.1",
   "description": "AI-powered issue solver and hive mind for collaborative problem solving",
   "main": "src/hive.mjs",
   "type": "module",

package/src/claude.budget-stats.lib.mjs

@@ -165,16 +165,15 @@ export const displayBudgetStats = async (usage, tokenUsage, log) => {
       const sub = subSessions[i];
       const subPeak = sub.peakContextUsage || 0;
       // Issue #1539: Only use peak per-request context for context window display.
-      // Cumulative totals across all requests can exceed the context limit and produce
-      // impossible percentages (e.g. 250%). When peak is unknown, skip context display.
+      // Issue #1547: Percentage before unit label: X / Y (Z%) input tokens
       const parts = [];
       if (contextLimit && subPeak > 0) {
         const pct = ((subPeak / contextLimit) * 100).toFixed(0);
-        parts.push(`${formatNumber(subPeak)} / ${formatNumber(contextLimit)} input tokens (${pct}%)`);
+        parts.push(`${formatNumber(subPeak)} / ${formatNumber(contextLimit)} (${pct}%) input tokens`);
       }
       if (outputLimit) {
         const outPct = ((sub.outputTokens / outputLimit) * 100).toFixed(0);
-        parts.push(`${formatNumber(sub.outputTokens)} / ${formatNumber(outputLimit)} output tokens (${outPct}%)`);
+        parts.push(`${formatNumber(sub.outputTokens)} / ${formatNumber(outputLimit)} (${outPct}%) output tokens`);
       }
       if (parts.length > 0) {
         await log(`        ${i + 1}. Context window: ${parts.join(', ')}`);
@@ -182,14 +181,15 @@ export const displayBudgetStats = async (usage, tokenUsage, log) => {
     }
   } else if (peakContext > 0) {
     // Single sub-session with known peak: single-line format
+    // Issue #1547: Percentage before unit label
     const parts = [];
     if (contextLimit) {
       const pct = ((peakContext / contextLimit) * 100).toFixed(0);
-      parts.push(`${formatNumber(peakContext)} / ${formatNumber(contextLimit)} input tokens (${pct}%)`);
+      parts.push(`${formatNumber(peakContext)} / ${formatNumber(contextLimit)} (${pct}%) input tokens`);
     }
     if (outputLimit) {
       const outPct = ((usage.outputTokens / outputLimit) * 100).toFixed(0);
-      parts.push(`${formatNumber(usage.outputTokens)} / ${formatNumber(outputLimit)} output tokens (${outPct}%)`);
+      parts.push(`${formatNumber(usage.outputTokens)} / ${formatNumber(outputLimit)} (${outPct}%) output tokens`);
     }
     if (parts.length > 0) {
       await log(`        Context window: ${parts.join(', ')}`);
@@ -199,15 +199,20 @@ export const displayBudgetStats = async (usage, tokenUsage, log) => {
   // Cumulative totals are shown on the Total line below — no duplication needed.
 
   // Cumulative totals — single line
+  // Issue #1547: Parenthesized cached format and consistent output format
   const totalInputNonCached = usage.inputTokens + usage.cacheCreationTokens;
   const cachedTokens = usage.cacheReadTokens;
-  let totalLine = `${formatNumber(totalInputNonCached)}`;
-  if (cachedTokens > 0) totalLine += ` + ${formatNumber(cachedTokens)} cached`;
-  totalLine += ` input tokens, ${formatNumber(usage.outputTokens)} output tokens`;
-  // Issue #1539: When peakContextUsage is unknown, embed output percentage in Total line
+  let totalLine;
+  if (cachedTokens > 0) {
+    totalLine = `(${formatNumber(totalInputNonCached)} + ${formatNumber(cachedTokens)} cached) input tokens`;
+  } else {
+    totalLine = `${formatNumber(totalInputNonCached)} input tokens`;
+  }
   if (peakContext === 0 && outputLimit) {
     const outPct = ((usage.outputTokens / outputLimit) * 100).toFixed(0);
-    totalLine += ` (${outPct}% of ${formatNumber(outputLimit)} output limit)`;
+    totalLine += `, ${formatNumber(usage.outputTokens)} / ${formatNumber(outputLimit)} (${outPct}%) output tokens`;
+  } else {
+    totalLine += `, ${formatNumber(usage.outputTokens)} output tokens`;
   }
   await log(`        Total: ${totalLine}`);
 };
@@ -325,12 +330,12 @@ const formatContextOutputLine = (peakContext, contextLimit, outputTokens, output
     // context window metrics and produce impossible percentages (e.g. 250%).
     if (peakContext > 0) {
       const pct = ((peakContext / contextLimit) * 100).toFixed(0);
-      parts.push(`${formatTokensCompact(peakContext)} / ${formatTokensCompact(contextLimit)} input tokens (${pct}%)`);
+      parts.push(`${formatTokensCompact(peakContext)} / ${formatTokensCompact(contextLimit)} (${pct}%) input tokens`);
     }
   }
   if (outputLimit) {
     const outPct = ((outputTokens / outputLimit) * 100).toFixed(0);
-    parts.push(`${formatTokensCompact(outputTokens)} / ${formatTokensCompact(outputLimit)} output tokens (${outPct}%)`);
+    parts.push(`${formatTokensCompact(outputTokens)} / ${formatTokensCompact(outputLimit)} (${outPct}%) output tokens`);
   }
   if (parts.length === 0) return '';
   return `\n${prefix}Context window: ${parts.join(', ')}`;
@@ -364,15 +369,6 @@ export const buildBudgetStatsString = tokenUsage => {
     const subSessions = tokenUsage.subSessions || [];
     const hasMultipleSubSessions = subSessions.length > 1;
 
-    if (isMultiModel && hasMultipleSubSessions) {
-      // Issue #1508: For multi-model sessions, show global sub-sessions once (not per-model),
-      // since sub-sessions track compactification boundaries which are session-wide.
-      // Per-model context/output limits are shown below under each model heading.
-      const primaryModelId = modelIds[0];
-      const primaryUsage = tokenUsage.modelUsage[primaryModelId];
-      stats += formatSubSessionsList(subSessions, primaryUsage.modelInfo?.limit?.context, primaryUsage.modelInfo?.limit?.output);
-    }
-
     for (const modelId of modelIds) {
       const usage = tokenUsage.modelUsage[modelId];
       const modelName = usage.modelName || modelId;
@@ -383,8 +379,9 @@ export const buildBudgetStatsString = tokenUsage => {
 
       const peakContext = usage.peakContextUsage || 0;
 
-      if (!isMultiModel && hasMultipleSubSessions) {
-        // Single-model + multiple sub-sessions: show numbered sub-sessions under that model
+      if (hasMultipleSubSessions && (!isMultiModel || modelId === modelIds[0])) {
+        // Issue #1547: Show sub-sessions under the primary model heading (not globally).
+        // For single-model sessions, show under that model. For multi-model, under the first model.
         stats += formatSubSessionsList(subSessions, contextLimit, outputLimit);
       } else if (peakContext > 0) {
         // Issue #1526: Single line format for context window + output tokens
@@ -394,18 +391,22 @@ export const buildBudgetStatsString = tokenUsage => {
       // Cumulative totals are shown on the Total line below — no duplication needed.
 
       // Cumulative totals per model: input tokens + cached shown separately
-      // Issue #1526: Shorter format — single "Total:" line
+      // Issue #1547: Parenthesized cached format: (X + Y cached) input tokens
       const totalInputNonCached = usage.inputTokens + usage.cacheCreationTokens;
       const cachedTokens = usage.cacheReadTokens;
-      let totalLine = `${formatTokensCompact(totalInputNonCached)}`;
-      if (cachedTokens > 0) totalLine += ` + ${formatTokensCompact(cachedTokens)} cached`;
-      totalLine += ` input tokens, ${formatTokensCompact(usage.outputTokens)} output tokens`;
+      let totalLine;
+      if (cachedTokens > 0) {
+        totalLine = `(${formatTokensCompact(totalInputNonCached)} + ${formatTokensCompact(cachedTokens)} cached) input tokens`;
+      } else {
+        totalLine = `${formatTokensCompact(totalInputNonCached)} input tokens`;
+      }
 
-      // Issue #1539: When peakContextUsage is unknown (no per-request data), embed
-      // output token percentage in the Total line so no data is lost.
+      // Issue #1547: Consistent output format — use X / Y (Z%) output tokens when limit known
       if (peakContext === 0 && outputLimit) {
         const outPct = ((usage.outputTokens / outputLimit) * 100).toFixed(0);
-        totalLine += ` (${outPct}% of ${formatTokensCompact(outputLimit)} output limit)`;
+        totalLine += `, ${formatTokensCompact(usage.outputTokens)} / ${formatTokensCompact(outputLimit)} (${outPct}%) output tokens`;
+      } else {
+        totalLine += `, ${formatTokensCompact(usage.outputTokens)} output tokens`;
       }
 
       // Issue #1508: Show per-model cost when available

package/src/models/index.mjs

@@ -47,6 +47,7 @@ export const claudeModels = {
 
 // Agent models (OpenCode API and Kilo Gateway via agent CLI)
 // Issue #1300: Updated free models to match agent PR #191
+// Issue #1543: Added qwen3.6-plus-free (new default) and nemotron-3-super-free per agent PR #234
 export const agentModels = {
   // OpenCode Zen free models (current)
   grok: 'opencode/grok-code',
@@ -54,7 +55,9 @@ export const agentModels = {
   'grok-code-fast-1': 'opencode/grok-code',
   'big-pickle': 'opencode/big-pickle',
   'gpt-5-nano': 'opencode/gpt-5-nano',
-  'minimax-m2.5-free': 'opencode/minimax-m2.5-free', // New: upgraded from M2.1 (Issue #1391: now default)
+  'minimax-m2.5-free': 'opencode/minimax-m2.5-free', // Upgraded from M2.1 (Issue #1391)
+  'qwen3.6-plus-free': 'opencode/qwen3.6-plus-free', // New: ~1M context, default (Issue #1543)
+  'nemotron-3-super-free': 'opencode/nemotron-3-super-free', // New: NVIDIA hybrid Mamba-Transformer (Issue #1543)
   // Kilo Gateway free models (Issue #1282, updated in #1300)
   // Short names for Kilo-exclusive models (Issue #1300)
   'glm-5-free': 'kilo/glm-5-free', // Kilo-exclusive
@@ -112,7 +115,7 @@ export const codexModels = {
 // Default model for each tool (Issue #1473: centralized to avoid scattered hardcoded defaults)
 export const defaultModels = {
   claude: 'sonnet',
-  agent: 'minimax-m2.5-free',
+  agent: 'qwen3.6-plus-free', // Issue #1543: changed from minimax-m2.5-free per agent PR #234
   opencode: 'grok-code-fast-1',
   codex: 'gpt-5',
 };
@@ -140,6 +143,8 @@ export const freeToBaseModelMap = {
   'glm-4.7-free': 'glm-4.7',
   'minimax-m2.1-free': 'minimax-m2.1',
   'minimax-m2.5-free': 'minimax-m2.5',
+  'qwen3.6-plus-free': 'qwen3.6-plus', // Issue #1543
+  'nemotron-3-super-free': 'nemotron-3-super', // Issue #1543
   'glm-5-free': 'glm-5',
   'glm-4.5-air-free': 'glm-4.5-air',
   'deepseek-r1-free': 'deepseek-r1',
@@ -187,6 +192,8 @@ export const AGENT_MODELS = {
   'opencode/big-pickle': 'opencode/big-pickle',
   'opencode/gpt-5-nano': 'opencode/gpt-5-nano',
   'opencode/minimax-m2.5-free': 'opencode/minimax-m2.5-free',
+  'opencode/qwen3.6-plus-free': 'opencode/qwen3.6-plus-free', // Issue #1543
+  'opencode/nemotron-3-super-free': 'opencode/nemotron-3-super-free', // Issue #1543
   'opencode/kimi-k2.5-free': 'opencode/kimi-k2.5-free', // Deprecated
   'opencode/glm-4.7-free': 'opencode/glm-4.7-free', // Deprecated
   'opencode/minimax-m2.1-free': 'opencode/minimax-m2.1-free', // Deprecated
@@ -297,7 +304,7 @@ export const primaryModelNames = {
   claude: ['opus', 'sonnet', 'haiku', 'opusplan'],
   opencode: ['grok', 'gpt4o'],
   codex: ['gpt5', 'gpt5-codex', 'o3'],
-  agent: ['minimax-m2.5-free', 'big-pickle', 'gpt-5-nano', 'glm-5-free', 'deepseek-r1-free'],
+  agent: ['qwen3.6-plus-free', 'nemotron-3-super-free', 'minimax-m2.5-free', 'big-pickle', 'gpt-5-nano', 'glm-5-free', 'deepseek-r1-free'],
 };
 
 /**