@mtayfur/opencode-cache-view 0.0.4 → 1.0.2

package/README.md

@@ -28,7 +28,7 @@ customize-opencode       3.1K tok
 
 - **Cache:** `Hit` is `cache.read / (input + cache.read + cache.write)` for the latest regular LLM step. `Session Hit` is the aggregate OpenCode session ratio, including compaction calls. `Miss` combines uncached input and cache writes.
 - **Estimated Tokens:** Uses one character-based estimator for the visible system, user, tool, reasoning, and output content from the latest completed compaction summary, its preserved tail, and subsequent messages loaded in the TUI. Provider system prompts, tool schemas, and media tokens are not included.
-- **Speed:** During streaming, `TPS` shows a `~`-prefixed estimate from visible text and reasoning deltas over the latest five seconds. Hidden reasoning with no deltas uses the current turn's cumulative rate when available, otherwise the previous completed turn, or remains `… tok/s` when neither exists. Completed model steps accumulate `(output + reasoning)` tokens and full step duration across tool calls, excluding tool execution time. The cumulative rate remains `~`-prefixed while the turn is incomplete; the final non-tool-call step replaces it with the exact weighted turn throughput. `TTFT` retains the estimated delay to the first visible reasoning or text block, and `Trend` shows the latest eight completed-turn TPS values.
+- **Speed:** During streaming, `TPS` shows a `~`-prefixed estimate from visible text and reasoning deltas over the latest five seconds. Hidden reasoning with no deltas uses the previous completed turn or remains `… tok/s` when none exists. Completed tool-call steps and tool execution time are excluded. The rate remains `~`-prefixed while the turn is incomplete; the final non-tool-call assistant step replaces it with exact `(output + reasoning) / duration` throughput. `TTFT` retains the estimated delay to the first visible reasoning or text block, and `Trend` shows the latest eight completed-turn TPS values.
 - **Loaded Skills:** Shows active, non-compacted skill outputs and their estimated token counts. When a skill is loaded more than once, the latest load is shown.
 
 Rows and the cache-hit bar adapt to the available sidebar width. The arrow next to the hit rate compares the latest two LLM steps. Sections can be expanded or collapsed with the mouse.
@@ -88,9 +88,11 @@ Metric calculations are isolated from the TUI rendering layer. `read.ts` snapsho
 
 ## Development
 
+From the repository root:
+
 ```sh
 bun install --frozen-lockfile
-bun run typecheck
-bun run build
-npm pack --dry-run
+bun run --filter @mtayfur/opencode-cache-view typecheck
+bun run --filter @mtayfur/opencode-cache-view build
+npm pack ./packages/cache-view --dry-run
 ```

package/dist/index.js

@@ -410,7 +410,7 @@ function createSpeedTracker(api) {
   const starts = new Map;
   const latest = new Map;
   const trends = new Map;
-  const turnTotals = new Map;
+  const pendingToolTurns = new Set;
   const streamParts = new Map;
   const streamSamples = new Map;
   const clearStream = (messageID) => {
@@ -446,28 +446,27 @@ function createSpeedTracker(api) {
     if (!start || start.sessionID !== event.properties.sessionID)
       return;
     clearStream(value.messageID);
-    const duration = event.properties.time - start.time;
+    const duration = start.firstTokenTime === undefined ? 0 : event.properties.time - start.firstTokenTime;
     const tokens = value.tokens.output + value.tokens.reasoning;
-    if (duration <= 0 || tokens <= 0)
-      return;
-    const previous = turnTotals.get(start.sessionID) ?? { tokens: 0, duration: 0 };
-    const turn = {
-      tokens: previous.tokens + tokens,
-      duration: previous.duration + duration
-    };
-    const speed = turn.tokens / turn.duration * 1000;
-    latest.set(start.sessionID, speed);
     if (value.reason === "tool-calls") {
-      turnTotals.set(start.sessionID, turn);
+      pendingToolTurns.add(start.sessionID);
       return;
     }
-    turnTotals.delete(start.sessionID);
+    pendingToolTurns.delete(start.sessionID);
+    if (duration <= 0 || tokens <= 0)
+      return;
+    const speed = tokens / duration * 1000;
+    latest.set(start.sessionID, speed);
     trends.set(start.sessionID, [...trends.get(start.sessionID) ?? [], speed].slice(-8));
   });
   const delta = api.event.on("message.part.delta", (event) => {
-    if (event.properties.field !== "text" || !starts.has(event.properties.messageID))
+    if (event.properties.field !== "text")
+      return;
+    const start = starts.get(event.properties.messageID);
+    if (!start)
       return;
     const now = Date.now();
+    start.firstTokenTime ??= now;
     const previous = streamParts.get(event.properties.partID) ?? {
       messageID: event.properties.messageID,
       units: 0
@@ -489,7 +488,7 @@ function createSpeedTracker(api) {
     starts.delete(event.properties.info.id);
     clearStream(event.properties.info.id);
     if (event.properties.info.error !== undefined)
-      turnTotals.delete(event.properties.sessionID);
+      pendingToolTurns.delete(event.properties.sessionID);
   });
   const messageRemoved = api.event.on("message.removed", (event) => {
     starts.delete(event.properties.messageID);
@@ -498,7 +497,7 @@ function createSpeedTracker(api) {
   const sessionDeleted = api.event.on("session.deleted", (event) => {
     latest.delete(event.properties.sessionID);
     trends.delete(event.properties.sessionID);
-    turnTotals.delete(event.properties.sessionID);
+    pendingToolTurns.delete(event.properties.sessionID);
     for (const [messageID, start] of starts) {
       if (start.sessionID === event.properties.sessionID) {
         starts.delete(messageID);
@@ -519,7 +518,7 @@ function createSpeedTracker(api) {
       const active = activeMessageID !== undefined;
       const live = activeMessageID === undefined ? undefined : streamingSpeed(activeMessageID, Date.now());
       const fallback = latest.get(sessionID);
-      const provisional = active || turnTotals.has(sessionID);
+      const provisional = active || pendingToolTurns.has(sessionID);
       return {
         active,
         value: active ? live ?? fallback : fallback,
@@ -536,7 +535,7 @@ function createSpeedTracker(api) {
       starts.clear();
       latest.clear();
       trends.clear();
-      turnTotals.clear();
+      pendingToolTurns.clear();
       streamParts.clear();
       streamSamples.clear();
     }

package/package.json

@@ -1,16 +1,17 @@
 {
   "$schema": "https://json.schemastore.org/package.json",
   "name": "@mtayfur/opencode-cache-view",
-  "version": "0.0.4",
+  "version": "1.0.2",
   "description": "Minimal OpenCode TUI sidebar for cache hit, estimated tokens, and generation speed.",
   "repository": {
     "type": "git",
-    "url": "git+https://github.com/mtayfur/opencode-cache-view.git"
+    "url": "git+https://github.com/mtayfur/opencode-plugins.git",
+    "directory": "packages/cache-view"
   },
   "bugs": {
-    "url": "https://github.com/mtayfur/opencode-cache-view/issues"
+    "url": "https://github.com/mtayfur/opencode-plugins/issues"
   },
-  "homepage": "https://github.com/mtayfur/opencode-cache-view#readme",
+  "homepage": "https://github.com/mtayfur/opencode-plugins/tree/main/packages/cache-view#readme",
   "files": [
     "dist"
   ],
@@ -38,8 +39,8 @@
     "@opentui/solid": ">=0.2.0"
   },
   "devDependencies": {
-    "@opencode-ai/plugin": "^1.18.16",
-    "@opencode-ai/sdk": "^1.18.16",
+    "@opencode-ai/plugin": "1.18.16",
+    "@opencode-ai/sdk": "1.18.16",
     "@opentui/core": "^0.5.1",
     "@opentui/solid": "^0.5.1",
     "jsonc-parser": "^3.3.1",