@plexor-dev/claude-code-plugin-staging 0.1.0-beta.28 → 0.1.0-beta.29

package/hooks/track-response.js

@@ -232,8 +232,7 @@ async function main() {
     // Phase 1 supervisor UX: concise single-line routing summary
     supervisor.emit(response, plexorMeta);
 
-    // Proactive compact warning: check prompt token count against provider limits
-    emitCompactWarning(response, plexorMeta);
+    // Context warnings now handled by supervisor.emit() (Phase 5) to avoid duplicates
 
     // Issue #701: Track ALL responses, not just when enabled
     // This ensures session stats are always accurate

package/lib/supervisor.js

@@ -24,6 +24,9 @@
  * decisions to the developer without requiring them to parse verbose logs.
  */
 
+const fs = require('fs');
+const path = require('path');
+
 const CYAN = '\x1b[36m';
 const YELLOW = '\x1b[33m';
 const RED = '\x1b[31m';
@@ -32,6 +35,8 @@ const RESET = '\x1b[0m';
 
 const CONTEXT_WARNING_THRESHOLD = 70000;
 const CONTEXT_COMPACT_THRESHOLD = 80000;
+const SESSION_STATE_PATH = path.join(process.env.HOME || '', '.plexor', 'supervisor-session.json');
+const SESSION_TIMEOUT_MS = 30 * 60 * 1000; // 30 min — start fresh if idle
 
 class SupervisorEmitter {
   /**
@@ -46,11 +51,54 @@ class SupervisorEmitter {
       this.enabled = opts.enabled !== false;
     }
 
-    // Phase 5: Session-level state
+    // Phase 5: Session-level state — restored from disk across hook invocations
+    this._loadSessionState();
+  }
+
+  /**
+   * Load persisted session state from disk. Resets if stale (>30 min idle).
+   */
+  _loadSessionState() {
+    try {
+      const raw = fs.readFileSync(SESSION_STATE_PATH, 'utf8');
+      const state = JSON.parse(raw);
+      const lastTs = state.lastTimestamp || 0;
+      if (Date.now() - lastTs > SESSION_TIMEOUT_MS) {
+        // Session expired — start fresh
+        this._resetSessionState();
+        return;
+      }
+      this._turnCount = state.turnCount || 0;
+      this._contextTokens = state.contextTokens || 0; // current context window size
+      this._cumulativeCost = state.cumulativeCost || 0;
+      this._providerHistory = state.providerHistory || [];
+    } catch {
+      this._resetSessionState();
+    }
+  }
+
+  _resetSessionState() {
     this._turnCount = 0;
-    this._cumulativeTokens = 0;
+    this._contextTokens = 0;
     this._cumulativeCost = 0;
-    this._providerHistory = []; // last N entries: { provider, hadToolCalls }
+    this._providerHistory = [];
+  }
+
+  /**
+   * Persist session state to disk for cross-process continuity.
+   */
+  _saveSessionState() {
+    try {
+      const dir = path.dirname(SESSION_STATE_PATH);
+      if (!fs.existsSync(dir)) fs.mkdirSync(dir, { recursive: true });
+      fs.writeFileSync(SESSION_STATE_PATH, JSON.stringify({
+        turnCount: this._turnCount,
+        contextTokens: this._contextTokens,
+        cumulativeCost: this._cumulativeCost,
+        providerHistory: this._providerHistory.slice(-20), // keep last 20
+        lastTimestamp: Date.now(),
+      }));
+    } catch { /* best-effort persistence */ }
   }
 
   /**
@@ -253,11 +301,13 @@ class SupervisorEmitter {
   _accumulateSessionState(response, plexorMeta) {
     this._turnCount++;
 
-    // Cumulative token counter — read prompt_tokens from usage block
+    // Context window size — prompt_tokens IS the current context size (not cumulative).
+    // Each turn's prompt_tokens includes the full conversation context.
     const usage = response?.usage || response?.plexor?.usage || {};
     const promptTokens = Number(usage.prompt_tokens) || 0;
-    const completionTokens = Number(usage.completion_tokens) || 0;
-    this._cumulativeTokens += promptTokens + completionTokens;
+    if (promptTokens > 0) {
+      this._contextTokens = promptTokens; // replace, not accumulate
+    }
 
     // Cumulative cost
     const costUsd = Number(
@@ -290,12 +340,12 @@ class SupervisorEmitter {
    */
   _emitSessionNarration() {
     // Session summary line — every turn
-    const tokenStr = this._formatTokenCount(this._cumulativeTokens);
+    const tokenStr = this._formatTokenCount(this._contextTokens);
     const costStr = this._cumulativeCost < 0.01
       ? `$${this._cumulativeCost.toFixed(4)}`
       : `$${this._cumulativeCost.toFixed(2)}`;
     process.stderr.write(
-      `${MAGENTA}[PLEXOR: Session: ${this._turnCount} turns, ${costStr} cost, ${tokenStr} tokens]${RESET}\n`
+      `${MAGENTA}[PLEXOR: Session: ${this._turnCount} turns, ${costStr} cost, ${tokenStr} context]${RESET}\n`
     );
 
     // Provider reliability digest — every 5th turn
@@ -306,18 +356,21 @@ class SupervisorEmitter {
       }
     }
 
-    // Context warning at 70K tokens
-    if (this._cumulativeTokens >= CONTEXT_COMPACT_THRESHOLD) {
-      const kTokens = Math.round(this._cumulativeTokens / 1000);
+    // Context warning at 70K tokens (uses real context window size, not cumulative)
+    if (this._contextTokens >= CONTEXT_COMPACT_THRESHOLD) {
+      const kTokens = Math.round(this._contextTokens / 1000);
       process.stderr.write(
         `${YELLOW}[PLEXOR: Context at ${kTokens}K \u2014 recommend /compact to prevent errors]${RESET}\n`
       );
-    } else if (this._cumulativeTokens >= CONTEXT_WARNING_THRESHOLD) {
-      const kTokens = Math.round(this._cumulativeTokens / 1000);
+    } else if (this._contextTokens >= CONTEXT_WARNING_THRESHOLD) {
+      const kTokens = Math.round(this._contextTokens / 1000);
       process.stderr.write(
         `${YELLOW}[PLEXOR: Context at ${kTokens}K tokens \u2014 approaching provider limits]${RESET}\n`
       );
     }
+
+    // Persist state for next hook invocation
+    this._saveSessionState();
   }
 
   /**
@@ -358,7 +411,7 @@ class SupervisorEmitter {
   // ---- Phase 5 accessors (for testing) ----
 
   get turnCount() { return this._turnCount; }
-  get cumulativeTokens() { return this._cumulativeTokens; }
+  get cumulativeTokens() { return this._contextTokens; }
   get cumulativeCost() { return this._cumulativeCost; }
 
   // ---- private helpers ----

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@plexor-dev/claude-code-plugin-staging",
-  "version": "0.1.0-beta.28",
+  "version": "0.1.0-beta.29",
   "description": "STAGING - LLM cost optimization plugin for Claude Code (internal testing)",
   "main": "lib/constants.js",
   "bin": {