fix(sidebar): drop tokens row from Current session — show messages + age only

The token count in the session block was reading the local Cursor JSONL
transcript through a chars/3.5 heuristic. Two problems:

  - The local transcript only contains compact message envelopes — the
    expanded tool results and file reads that dominate Cursor's real
    context budget never land in that file. Our number was 3.8k when
    Cursor's own Context panel showed 43.5k for the same chat.
  - There is no public API to read Cursor's authoritative counter
    (researched the Cursor docs API, forum threads, GitHub, Reddit —
    nothing exposes the per-chat context breakdown to third-party
    extensions, and Cursor staff's only comment is a vague "more
    visibility coming").

User picked Variant B from the menu of options: drop the token row
entirely, keep the two signals we CAN measure accurately — Messages
(count) and Started (age). The session warning now fires on either
>=50 messages OR >=2h elapsed, which roughly maps to "approaching
Cursor's ~50% auto-summarize threshold" without claiming a token
number we can't see. ActiveSession.tokens is still computed for any
future use, just not rendered.

#!axme pr=none repo=AxmeAI/axme-code

Author: George-iam

extension/src/sidebar-webview.ts

@@ -33,12 +33,10 @@ import { log } from "./log.js";
  */
 const SESSION_POLL_MS = 3_000;
 
-/** Threshold above which we warn the user to close the session. Chosen
- * to match the upper end of Cursor's reported auto-summarize trigger
- * (~50–60% of context window for 200k models) so the user has a chance
- * to close cleanly via our handoff flow BEFORE Cursor's lossy condense
- * fires. */
-const SESSION_WARN_TOKENS = 200_000;
+// (Old SESSION_WARN_TOKENS dropped — we no longer display a token number
+//  in the session block. See "Live session block" comment in the render
+//  code for the rationale. Threshold is now message-count + duration in
+//  the render code itself: >50 messages OR >2h triggers the warning.)
 
 export interface SidebarState {
   /** Is the workspace initialised (`.axme-code/` exists)? */
@@ -55,10 +53,10 @@ export interface SidebarState {
   hooksOk: boolean;
   /** Are we running in Cursor (vs other host)? */
   isCursor: boolean;
-  /** Live snapshot of the active chat session — tokens, messages, age. */
+  /** Live snapshot of the active chat session — messages, age (tokens are
+   *  collected but no longer displayed; see "Live session block" comment
+   *  in the render code for the rationale). */
   session: ActiveSession | null;
-  /** Warn threshold (passed to webview so it can hide its own UI). */
-  warnTokens: number;
 }
 
 export type SidebarMessage =
@@ -85,7 +83,7 @@ export class AxmeSidebarProvider implements vscode.WebviewViewProvider {
 
   constructor(
     private readonly context: vscode.ExtensionContext,
-    private readonly initialState: Omit<SidebarState, "counts" | "backlog" | "auditorKeyConfigured" | "session" | "warnTokens">,
+    private readonly initialState: Omit<SidebarState, "counts" | "backlog" | "auditorKeyConfigured" | "session">,
   ) {}
 
   attach(workspaceRoot: string | undefined, binary: string): void {
@@ -152,7 +150,6 @@ export class AxmeSidebarProvider implements vscode.WebviewViewProvider {
       counts,
       backlog,
       hooksOk: hooksAreInstalled(),
-      warnTokens: SESSION_WARN_TOKENS,
       ...this.pendingState,
     });
     this.pendingState = {};
@@ -441,7 +438,7 @@ select, input[type=text], input[type=password] {
 
 const SIDEBAR_JS = `
 const vscode = acquireVsCodeApi();
-let S = { setupDone: false, counts: { memories:0, decisions:0, safety:0, backlog:0, questions:0 }, backlog: [], auditorMode: "cooperative", auditorKeyConfigured: false, hooksOk: false, isCursor: true, session: null, warnTokens: 200000 };
+let S = { setupDone: false, counts: { memories:0, decisions:0, safety:0, backlog:0, questions:0 }, backlog: [], auditorMode: "cooperative", auditorKeyConfigured: false, hooksOk: false, isCursor: true, session: null };
 
 function formatDuration(ms) {
   if (ms <= 0) return "just now";
@@ -453,11 +450,8 @@ function formatDuration(ms) {
   const rm = m % 60;
   return h + "h " + (rm ? rm + "m" : "");
 }
-function formatTokens(n) {
-  if (n < 1000) return n + "";
-  if (n < 100000) return (n / 1000).toFixed(1).replace(/\\.0$/, "") + "k";
-  return Math.round(n / 1000) + "k";
-}
+// formatTokens was used by the old Tokens row in the session block.
+// Variant B dropped that row — function removed to keep dead code out.
 
 function send(msg) { vscode.postMessage(msg); }
 function cmd(id)  { send({ type: "command", commandId: id }); }
@@ -554,22 +548,32 @@ function render() {
   });
 
   // Live session block — driven by readActiveSession on the host side.
+  // We deliberately do NOT show a token count: our only data source is the
+  // local Cursor JSONL transcript which contains compact message envelopes,
+  // not the expanded tool results that dominate Cursor's actual context
+  // budget. Showing "3.8k" while Cursor's own Context panel shows 43.5k
+  // confused users more than it helped. Instead we show two signals we CAN
+  // measure honestly: messages (count) and duration (age).
+  //
+  // Warning threshold uses both. >50 messages OR >2h is a reasonable
+  // proxy for "you're heading toward Cursor's ~50–60% auto-summarize
+  // trigger" without lying about a token number we can't see.
   const sess = S.session;
   let sessionHtml = '<p class="muted">No active chat detected. Tools will record activity when an MCP call lands.</p>';
   if (sess && sess.hasData) {
     const startedMs = Date.parse(sess.startedAt);
     const ageMs = Number.isFinite(startedMs) ? Date.now() - startedMs : 0;
-    const overWarn = sess.tokens >= S.warnTokens;
+    const ageHours = ageMs / 3_600_000;
+    const overWarn = sess.messages >= 50 || ageHours >= 2;
     sessionHtml = \`
       <div class="row"><span class="k">Started</span><span class="v">\${formatDuration(ageMs)} ago</span></div>
-      <div class="row"><span class="k">Tokens</span><span class="v">\${formatTokens(sess.tokens)}</span></div>
       <div class="row"><span class="k">Messages</span><span class="v">\${sess.messages}</span></div>
       \${overWarn ? \`
         <div class="warning-banner">
-          Approaching Cursor's auto-summarize threshold. Cursor will compress
-          your conversation around here and quality often degrades after that.
-          Ask the agent in the chat to <b>close the session</b> — it will
-          extract memories / decisions / safety inline and give you a
+          Session is getting long (\${sess.messages} messages, \${formatDuration(ageMs)}).
+          Cursor auto-summarizes around the ~50% context mark and quality often
+          regresses after that. Ask the agent to <b>close the session</b> —
+          it will extract memories / decisions / safety inline and give you a
           startup prompt for a fresh chat with zero context loss.
         </div>\` : ""}
     \`;