@serkanalgur/opencode-slim 1.1.0 → 1.3.0

package/README.md

@@ -1,15 +1,22 @@
 # opencode-slim
 
+[![npm version](https://img.shields.io/npm/v/@serkanalgur/opencode-slim.svg)](https://www.npmjs.com/package/@serkanalgur/opencode-slim)
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
+
 Smart context management plugin for OpenCode. Optimizes token usage through semantic compression, cost-aware pruning, and adaptive thresholds.
 
 ## Features
 
+- **TUI Panel** - Rich context usage visualization with status indicators
+- **Enhanced Compress** - Auto/range/topic modes for flexible compression
 - **Semantic Compression** - Groups related tool calls and compresses them intelligently
 - **Cost-Aware Pruning** - Considers token pricing when deciding what to compress
 - **Adaptive Thresholds** - Learns from compression history to optimize timing
 - **Session Persistence** - Saves state across restarts
 - **Deduplication** - Removes repeated tool calls automatically
 - **Error Purging** - Cleans up failed tool call outputs after configurable turns
+- **Topic Extraction** - Identifies and tracks conversation topics
+- **Smart Recommendations** - Provides actionable suggestions for context optimization
 
 ## Installation
 
@@ -17,6 +24,39 @@ Smart context management plugin for OpenCode. Optimizes token usage through sema
 opencode plugin @serkanalgur/opencode-slim@latest --global
 ```
 
+## Usage
+
+### Slash Commands
+
+Slim provides slash commands for easy access:
+
+- `/panel` — Opens the rich TUI panel with context usage, stats, and help
+- `/compress [focus]` — Manually trigger compression with focus description
+
+### TUI Panel
+
+The panel provides a real-time overview of your context usage, including:
+- Token usage vs model limit with visual progress bar
+- Message breakdown by role (user/assistant/tools)
+- Compression history and savings
+- Cost estimation based on your model
+- Smart recommendations for optimization
+
+### Compress Tool
+
+The enhanced compress tool supports multiple modes:
+
+```typescript
+// Auto mode (default) - intelligently selects what to compress
+compress({ focus: "old exploration" })
+
+// Range mode - compress specific message range
+compress({ focus: "completed tasks", mode: "range", start: 0, end: 50 })
+
+// Topic mode - compress messages matching a topic
+compress({ focus: "database work", mode: "topic", topic: "database" })
+```
+
 ## Configuration
 
 Create `~/.config/opencode/slim.jsonc`:
@@ -62,6 +102,16 @@ Create `~/.config/opencode/slim.jsonc`:
 
 ## How It Works
 
+### TUI Panel
+
+The panel provides a real-time overview of your context usage, including:
+- Token usage vs model limit with visual progress bar
+- Message breakdown by role (user/assistant/tools)
+- Compression history and savings
+- Cost estimation based on your model
+- Topic distribution across the conversation
+- Smart recommendations for optimization
+
 ### Semantic Compression
 
 Unlike simple text truncation, slim analyzes the semantic content of messages and groups related tool calls together. This preserves context while removing redundancy.
@@ -80,19 +130,24 @@ State is saved to disk, so compression history and learning persist across resta
 
 ## Commands
 
-- `/slim` - Show current context stats and compression history
-- `/slim compress` - Manually trigger compression
-- `/slim reset` - Reset adaptive learning
+| Command | Description |
+|---------|-------------|
+| `/panel` | Open the Slim TUI panel with context usage, stats, and help |
+| `/compress [focus]` | Manually trigger compression with focus description |
 
 ## Comparison with DCP
 
 | Feature | slim | DCP |
 |---------|------|-----|
+| TUI Panel | ✅ | ✅ |
+| Slash Commands | ✅ `/panel` `/compress` | ✅ `/dcp` `/dcp-compress` |
+| Manual Compress | ✅ | ✅ |
 | Semantic grouping | ✅ | ❌ |
 | Cost awareness | ✅ | ❌ |
 | Adaptive thresholds | ✅ | ❌ |
 | Session persistence | ✅ | ❌ |
-| File count | ~8 | ~150 |
+| Topic extraction | ✅ | ❌ |
+| File count | ~10 | ~150 |
 | License | MIT | AGPL-3.0 |
 | Config complexity | Low | High |
 

package/dist/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,MAAM,EAAE,MAAM,qBAAqB,CAAA;;;;;AAoSjD,wBAA8C"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,MAAM,EAAE,MAAM,qBAAqB,CAAA;;;;;AAqYjD,wBAA8C"}

package/dist/index.js

@@ -13663,6 +13663,251 @@ function getNudgeMessage(reason, tokenCount, maxTokens) {
   }
 }
 
+// lib/types.ts
+var COST_PROFILES = {
+  // Anthropic
+  "anthropic/claude-sonnet-4-20250514": {
+    inputPricePer1k: 3e-3,
+    outputPricePer1k: 0.015,
+    cacheReadPricePer1k: 3e-4,
+    cacheWritePricePer1k: 375e-5
+  },
+  "anthropic/claude-3-5-sonnet-20241022": {
+    inputPricePer1k: 3e-3,
+    outputPricePer1k: 0.015,
+    cacheReadPricePer1k: 3e-4,
+    cacheWritePricePer1k: 375e-5
+  },
+  // OpenAI
+  "openai/gpt-4o": {
+    inputPricePer1k: 25e-4,
+    outputPricePer1k: 0.01,
+    cacheReadPricePer1k: 125e-5,
+    cacheWritePricePer1k: 25e-4
+  },
+  "openai/gpt-4o-mini": {
+    inputPricePer1k: 15e-5,
+    outputPricePer1k: 6e-4,
+    cacheReadPricePer1k: 75e-6,
+    cacheWritePricePer1k: 15e-5
+  },
+  // Default
+  "default": {
+    inputPricePer1k: 3e-3,
+    outputPricePer1k: 0.015,
+    cacheReadPricePer1k: 3e-4,
+    cacheWritePricePer1k: 375e-5
+  }
+};
+
+// lib/tui.ts
+async function buildPanelData(sessionId, messages, state, config2, modelId) {
+  const modelContextLimit = state.modelContextLimit || 2e5;
+  const maxTokens = resolveTokenLimit(config2.compress.maxContextLimit, modelContextLimit);
+  let currentTokens = 0;
+  const tokensByRole = { user: 0, assistant: 0, tools: 0, system: 0 };
+  let userMessages = 0;
+  let assistantMessages = 0;
+  let toolCalls = 0;
+  let toolResults = 0;
+  for (const msg of messages) {
+    const role = msg.info.role;
+    const text = getMessageText(msg);
+    const toolContent = getToolResultContent(msg);
+    const msgTokens = await countTokens(text + toolContent);
+    currentTokens += msgTokens;
+    if (role === "user") {
+      tokensByRole.user += msgTokens;
+      userMessages++;
+    } else if (role === "assistant") {
+      tokensByRole.assistant += msgTokens;
+      assistantMessages++;
+    }
+    for (const part of msg.parts) {
+      if (part.type === "tool") {
+        const toolPart = part;
+        if (toolPart.state?.status === "completed") {
+          toolResults++;
+        } else {
+          toolCalls++;
+        }
+      }
+    }
+  }
+  tokensByRole.tools = tokensByRole.user + tokensByRole.assistant - tokensByRole.user - tokensByRole.assistant;
+  tokensByRole.system = Math.max(0, currentTokens - tokensByRole.user - tokensByRole.assistant - tokensByRole.tools);
+  const usagePercent = currentTokens / maxTokens * 100;
+  let status = "healthy";
+  if (usagePercent > 90) status = "critical";
+  else if (usagePercent > 70) status = "warning";
+  const compressionCount = state.compressionCount;
+  const averageRatio = state.averageCompressionRatio;
+  let totalTokensSaved = 0;
+  for (const record2 of state.compressionHistory) {
+    if (record2.success) {
+      totalTokensSaved += record2.inputTokens - record2.outputTokens;
+    }
+  }
+  const lastCompression = state.compressionHistory.length > 0 ? state.compressionHistory[state.compressionHistory.length - 1] : null;
+  const profile = COST_PROFILES[modelId || "default"] || COST_PROFILES.default;
+  const estimatedCost = currentTokens / 1e3 * profile.inputPricePer1k;
+  const costSaved = totalTokensSaved / 1e3 * profile.inputPricePer1k;
+  const topicMap = /* @__PURE__ */ new Map();
+  for (const msg of messages) {
+    const text = getMessageText(msg);
+    const topics2 = extractTopics(text);
+    for (const topic of topics2) {
+      const existing = topicMap.get(topic) || { count: 0, tokens: 0 };
+      existing.count++;
+      existing.tokens += await countTokens(text);
+      topicMap.set(topic, existing);
+    }
+  }
+  const topics = Array.from(topicMap.entries()).map(([topic, data]) => ({ topic, ...data })).sort((a, b) => b.tokens - a.tokens).slice(0, 10);
+  const recommendations = generateRecommendations(
+    usagePercent,
+    compressionCount,
+    averageRatio,
+    messages.length,
+    config2
+  );
+  return {
+    sessionId,
+    timestamp: Date.now(),
+    currentTokens,
+    maxTokens,
+    usagePercent,
+    status,
+    messageCount: messages.length,
+    userMessages,
+    assistantMessages,
+    toolCalls,
+    toolResults,
+    tokensByRole,
+    compressionCount,
+    averageRatio,
+    totalTokensSaved,
+    lastCompression,
+    estimatedCost,
+    costSaved,
+    model: modelId || "unknown",
+    topics,
+    recommendations
+  };
+}
+var TOPIC_KEYWORDS = {
+  "authentication": ["auth", "login", "password", "token", "session", "jwt"],
+  "database": ["database", "db", "sql", "query", "migration", "schema"],
+  "api": ["api", "endpoint", "route", "request", "response", "http"],
+  "testing": ["test", "spec", "assert", "expect", "describe", "jest"],
+  "configuration": ["config", "settings", "env", "environment", "variable"],
+  "deployment": ["deploy", "docker", "kubernetes", "ci", "cd", "pipeline"],
+  "ui": ["ui", "component", "render", "display", "style", "css"],
+  "error": ["error", "exception", "catch", "throw", "debug", "fix"],
+  "performance": ["performance", "optimize", "cache", "speed", "slow"],
+  "security": ["security", "encrypt", "decrypt", "hash", "sanitize"]
+};
+function extractTopics(text) {
+  const lower = text.toLowerCase();
+  const topics = [];
+  for (const [topic, keywords] of Object.entries(TOPIC_KEYWORDS)) {
+    if (keywords.some((kw) => lower.includes(kw))) {
+      topics.push(topic);
+    }
+  }
+  return topics.length > 0 ? topics : ["general"];
+}
+function generateRecommendations(usagePercent, compressionCount, averageRatio, messageCount, config2) {
+  const recs = [];
+  if (usagePercent > 80) {
+    recs.push("Context usage is high. Consider compressing older messages.");
+  }
+  if (usagePercent > 90) {
+    recs.push("Context nearly full! Run compress immediately to avoid truncation.");
+  }
+  if (compressionCount === 0 && messageCount > 20) {
+    recs.push("No compressions yet with many messages. Consider running compress.");
+  }
+  if (averageRatio < 0.3 && compressionCount > 0) {
+    recs.push("Compression ratio is low. Summaries may be too verbose.");
+  }
+  if (messageCount > 50 && usagePercent < 50) {
+    recs.push("Many messages but low usage. Deduplication may help further.");
+  }
+  if (recs.length === 0) {
+    recs.push("Context is healthy. No action needed.");
+  }
+  return recs;
+}
+function renderPanel(data) {
+  const lines = [];
+  lines.push("\u250C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510");
+  lines.push("\u2502                    SLIM CONTEXT PANEL                       \u2502");
+  lines.push("\u251C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2524");
+  const statusIcon = data.status === "healthy" ? "\u{1F7E2}" : data.status === "warning" ? "\u{1F7E1}" : "\u{1F534}";
+  lines.push(`\u2502 Status: ${statusIcon} ${data.status.toUpperCase().padEnd(10)} \u2502`);
+  lines.push("");
+  const barLength = 30;
+  const filledLength = Math.round(data.usagePercent / 100 * barLength);
+  const emptyLength = barLength - filledLength;
+  const bar = "\u2588".repeat(filledLength) + "\u2591".repeat(emptyLength);
+  lines.push(`\u2502 Context: [${bar}] ${data.usagePercent.toFixed(1)}%`);
+  lines.push(`\u2502          ${formatTokens(data.currentTokens)} / ${formatTokens(data.maxTokens)} tokens`);
+  lines.push("");
+  lines.push("\u2502 Messages:");
+  lines.push(`\u2502   User: ${data.userMessages}  Assistant: ${data.assistantMessages}`);
+  lines.push(`\u2502   Tool calls: ${data.toolCalls}  Results: ${data.toolResults}`);
+  lines.push("");
+  lines.push("\u2502 Token Distribution:");
+  lines.push(`\u2502   User: ${formatTokens(data.tokensByRole.user)}`);
+  lines.push(`\u2502   Assistant: ${formatTokens(data.tokensByRole.assistant)}`);
+  lines.push("");
+  lines.push("\u2502 Compression Stats:");
+  lines.push(`\u2502   Count: ${data.compressionCount}`);
+  lines.push(`\u2502   Avg ratio: ${(data.averageRatio * 100).toFixed(1)}%`);
+  lines.push(`\u2502   Tokens saved: ${formatTokens(data.totalTokensSaved)}`);
+  if (data.lastCompression) {
+    const ago = Date.now() - data.lastCompression.timestamp;
+    lines.push(`\u2502   Last: ${formatTimeAgo(ago)} ago`);
+  }
+  lines.push("");
+  lines.push("\u2502 Cost Estimate:");
+  lines.push(`\u2502   Current: $${data.estimatedCost.toFixed(4)}`);
+  lines.push(`\u2502   Saved: $${data.costSaved.toFixed(4)}`);
+  lines.push(`\u2502   Model: ${data.model}`);
+  lines.push("");
+  if (data.topics.length > 0) {
+    lines.push("\u2502 Top Topics:");
+    for (const topic of data.topics.slice(0, 5)) {
+      lines.push(`\u2502   ${topic.topic}: ${topic.count} msgs (${formatTokens(topic.tokens)})`);
+    }
+    lines.push("");
+  }
+  lines.push("\u2502 Recommendations:");
+  for (const rec of data.recommendations) {
+    lines.push(`\u2502   \u2022 ${rec}`);
+  }
+  lines.push("\u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518");
+  return lines.join("\n");
+}
+function formatTokens(tokens) {
+  if (tokens >= 1e6) {
+    return `${(tokens / 1e6).toFixed(1)}M`;
+  }
+  if (tokens >= 1e3) {
+    return `${(tokens / 1e3).toFixed(1)}K`;
+  }
+  return String(tokens);
+}
+function formatTimeAgo(ms) {
+  const seconds = Math.floor(ms / 1e3);
+  if (seconds < 60) return `${seconds}s`;
+  const minutes = Math.floor(seconds / 60);
+  if (minutes < 60) return `${minutes}m`;
+  const hours = Math.floor(minutes / 60);
+  return `${hours}h`;
+}
+
 // index.ts
 var sessionStates = /* @__PURE__ */ new Map();
 var sessionConfigs = /* @__PURE__ */ new Map();
@@ -13682,7 +13927,12 @@ var server = async (ctx) => {
   const compressTool = tool({
     description: getCompressToolDescription(),
     args: {
-      focus: external_exports.string().describe("Description of what content should be compressed")
+      focus: external_exports.string().describe("What to compress (e.g., 'old exploration', 'completed tasks')"),
+      mode: external_exports.enum(["auto", "range", "topic"]).default("auto").describe("Compression mode"),
+      start: external_exports.number().optional().describe("Start message index (for range mode)"),
+      end: external_exports.number().optional().describe("End message index (for range mode)"),
+      topic: external_exports.string().optional().describe("Topic to compress (for topic mode)"),
+      keepRecent: external_exports.number().default(5).describe("Number of recent messages to always keep")
     },
     async execute(args, context) {
       const config2 = getConfig(context.sessionID);
@@ -13699,17 +13949,42 @@ var server = async (ctx) => {
           info: m.info,
           parts: m.parts
         }));
+        let targetIndices = [];
         let inputTokens = 0;
-        for (const msg of messageWithParts) {
-          const text = getMessageText(msg) + getToolResultContent(msg);
-          inputTokens += await countTokens(text);
+        if (args.mode === "range" && args.start !== void 0 && args.end !== void 0) {
+          const start = Math.max(0, args.start);
+          const end = Math.min(messageWithParts.length, args.end);
+          for (let i = start; i < end; i++) {
+            targetIndices.push(i);
+            const text = getMessageText(messageWithParts[i]) + getToolResultContent(messageWithParts[i]);
+            inputTokens += await countTokens(text);
+          }
+        } else if (args.mode === "topic" && args.topic) {
+          const topicLower = args.topic.toLowerCase();
+          for (let i = 0; i < messageWithParts.length - args.keepRecent; i++) {
+            const msg = messageWithParts[i];
+            const text = getMessageText(msg) + getToolResultContent(msg);
+            if (text.toLowerCase().includes(topicLower)) {
+              targetIndices.push(i);
+              inputTokens += await countTokens(text);
+            }
+          }
+        } else {
+          const keepRecent = args.keepRecent;
+          for (let i = 0; i < messageWithParts.length - keepRecent; i++) {
+            const msg = messageWithParts[i];
+            const text = getMessageText(msg) + getToolResultContent(msg);
+            const tokens = await countTokens(text);
+            if (tokens < 100) continue;
+            targetIndices.push(i);
+            inputTokens += tokens;
+          }
         }
-        const targets = messageWithParts.length > 10 ? [{ start: 0, end: messageWithParts.length - 5, reason: "user_requested", estimatedTokens: inputTokens }] : [];
-        if (targets.length === 0) {
+        if (targetIndices.length === 0) {
           return "Nothing to compress - context is already efficient";
         }
-        const compressedMessages = messageWithParts.slice(targets[0].start, targets[0].end);
-        const summary = buildCompressionSummary(compressedMessages, args.focus);
+        const targetMessages = targetIndices.map((i) => messageWithParts[i]);
+        const summary = buildCompressionSummary(targetMessages, args.focus);
         const outputTokens = await countTokens(summary);
         const ratio = inputTokens > 0 ? 1 - outputTokens / inputTokens : 0;
         addCompressionRecord(
@@ -13719,19 +13994,20 @@ var server = async (ctx) => {
             inputTokens,
             outputTokens,
             ratio,
-            messageCount: compressedMessages.length,
+            messageCount: targetMessages.length,
             success: true
           },
           config2.adaptive.learningRate
         );
         saveSessionState(state, config2.persistence.directory);
         return {
-          title: `Compressed ${compressedMessages.length} messages`,
+          title: `Compressed ${targetMessages.length} messages`,
           output: summary,
           metadata: {
             inputTokens,
             outputTokens,
             ratio: Math.round(ratio * 100) + "%",
+            mode: args.mode,
             focus: args.focus
           }
         };
@@ -13740,6 +14016,55 @@ var server = async (ctx) => {
       }
     }
   });
+  const panelTool = tool({
+    description: `Display a rich context usage panel showing:
+- Current token usage vs model limit
+- Message breakdown (user/assistant/tools)
+- Token distribution by role
+- Compression history and savings
+- Cost estimate
+- Topic distribution
+- Smart recommendations`,
+    args: {},
+    async execute(_args, context) {
+      const config2 = getConfig(context.sessionID);
+      const state = getState(context.sessionID, config2);
+      try {
+        const response = await ctx.client.session.messages({
+          path: { id: context.sessionID }
+        });
+        if (!response.data || response.error) {
+          return "Failed to fetch messages";
+        }
+        const messageList = response.data;
+        const messageWithParts = messageList.map((m) => ({
+          info: m.info,
+          parts: m.parts
+        }));
+        const modelId = context.model?.id || "unknown";
+        const panelData = await buildPanelData(
+          context.sessionID,
+          messageWithParts,
+          state,
+          config2,
+          modelId
+        );
+        const panel = renderPanel(panelData);
+        return {
+          title: "Context Panel",
+          output: panel,
+          metadata: {
+            usagePercent: panelData.usagePercent,
+            status: panelData.status,
+            currentTokens: panelData.currentTokens,
+            maxTokens: panelData.maxTokens
+          }
+        };
+      } catch (error45) {
+        return `Error generating panel: ${error45 instanceof Error ? error45.message : "Unknown error"}`;
+      }
+    }
+  });
   return {
     config: async (opencodeConfig) => {
       if (!opencodeConfig.permission) {
@@ -13747,9 +14072,11 @@ var server = async (ctx) => {
       }
       ;
       opencodeConfig.permission.compress = globalConfig2.compress.permission;
+      opencodeConfig.permission.panel = "allow";
     },
     tool: {
-      compress: compressTool
+      compress: compressTool,
+      panel: panelTool
     },
     "experimental.chat.system.transform": async (input, output) => {
       const config2 = getConfig(input.sessionID || "");