copilot-tracer 1.0.6 → 1.0.7

package/dist/consoleUi.js

@@ -10,7 +10,7 @@ function fmtDuration(ms) {
     return `${(ms / 1000).toFixed(1)}s`;
 }
 function fmtCredits(c) {
-    return `${c.toFixed(2)} cr | $${(c * 0.01).toFixed(4)}`;
+    return `$${(c * 0.01).toFixed(4)}`;
 }
 function statusColor(status, text) {
     if (status === 'running')
@@ -25,7 +25,7 @@ export function renderConsoleTable(entries, summary) {
         head: [
             chalk.cyan('Date / Time'),
             chalk.cyan('Prompt'),
-            chalk.cyan('AI Credits'),
+            chalk.cyan('Est Cost'),
             chalk.cyan('Duration'),
             chalk.cyan('Tokens\nCached|Written|Reason'),
             chalk.cyan('Skills'),

package/dist/otlpReceiver.js

@@ -59,6 +59,31 @@ function getBodyText(body) {
     return undefined;
 }
 const inFlight = new Map(); // traceId → InFlight
+// Anthropic API pricing per 1K tokens (USD), converted to GitHub "AI credit" units
+// (1 credit = $0.01) so aiCredits stays one unit across Copilot and Claude entries.
+// Rates are Anthropic's current published per-model prices; a model id that doesn't
+// match a specific entry falls back to its tier's (opus/sonnet/haiku) latest rate.
+const ANTHROPIC_USD_PER_1K = {
+    'claude-fable-5': { input: 0.010, output: 0.050 },
+    'claude-mythos-5': { input: 0.010, output: 0.050 },
+    'claude-opus-5': { input: 0.005, output: 0.025 },
+    'claude-opus-4-8': { input: 0.005, output: 0.025 },
+    'claude-opus-4-7': { input: 0.005, output: 0.025 },
+    'claude-opus-4-6': { input: 0.005, output: 0.025 },
+    'claude-sonnet-5': { input: 0.002, output: 0.010 },
+    'claude-sonnet-4-6': { input: 0.003, output: 0.015 },
+    'claude-haiku-4-5': { input: 0.001, output: 0.005 },
+    'opus': { input: 0.005, output: 0.025 },
+    'sonnet': { input: 0.003, output: 0.015 },
+    'haiku': { input: 0.001, output: 0.005 },
+    'default': { input: 0.003, output: 0.015 },
+};
+export function calcClaudeCredits(tokens, model) {
+    const key = Object.keys(ANTHROPIC_USD_PER_1K).find(k => (model ?? '').toLowerCase().includes(k)) ?? 'default';
+    const rate = ANTHROPIC_USD_PER_1K[key];
+    const usd = (tokens.input / 1000) * rate.input + (tokens.output / 1000) * rate.output;
+    return usd * 100; // USD → credits (1 credit = $0.01)
+}
 // Sessions are created lazily. Always upsert so project_id gets backfilled
 // when the session was created earlier without a project.
 function ensureSession(sessionId, projectId) {
@@ -89,6 +114,10 @@ const pendingChatIds = new Map(); // `chat:${traceId}` → list of entryIds
 const pendingToolCalls = new Map(); // traceId → tool calls
 const claudePromptEntries = new Map(); // prompt.id → trace entry
 const claudeInteractionEntries = new Map(); // traceId → interaction entry
+// Real Claude Code telemetry sends the child claude_code.llm_request span before its
+// parent claude_code.interaction span within the same batch, so buffer llm_request's
+// token/cost delta here until the interaction entry shows up to receive it.
+const pendingClaudeLlmDeltas = new Map();
 const CLAUDE_STATE_LIMIT = 1000;
 function rememberClaudeEntry(map, key, entry) {
     map.set(key, entry);
@@ -211,13 +240,14 @@ function processSpans(spans, sessionId, projectId, workingDir) {
             const attrs = span.attributes ?? [];
             const inputTokens = Number(getAttr(attrs, 'input_tokens') ?? 0);
             const outputTokens = Number(getAttr(attrs, 'output_tokens') ?? 0);
+            const model = getStringAttr(attrs, 'model', 'gen_ai.request.model');
             const entry = {
                 id: spanId,
                 sessionId: getStringAttr(attrs, 'session.id') ?? sessionId,
                 dateTime: nanoToIso(span.startTimeUnixNano),
                 prompt: getStringAttr(attrs, 'user_prompt') ?? '[Claude Code interaction]',
                 tokens: { input: inputTokens, output: outputTokens, cached: Number(getAttr(attrs, 'cache_read_tokens') ?? 0), reasoning: 0, written: outputTokens, total: inputTokens + outputTokens },
-                aiCredits: Number(getAttr(attrs, 'cost_usd') ?? 0),
+                aiCredits: calcClaudeCredits({ input: inputTokens, output: outputTokens }, model),
                 durationMs: Number(getAttr(attrs, 'interaction.duration_ms')
                     ?? (nanoToMs(span.endTimeUnixNano) - nanoToMs(span.startTimeUnixNano))),
                 toolCalls: [],
@@ -227,6 +257,20 @@ function processSpans(spans, sessionId, projectId, workingDir) {
                 status: (span.status?.code ?? 0) === 2 ? 'error' : 'done',
                 error: span.status?.message,
             };
+            // Apply any llm_request delta that arrived before this interaction span.
+            const pendingDelta = pendingClaudeLlmDeltas.get(traceId);
+            if (pendingDelta) {
+                entry.tokens = {
+                    input: entry.tokens.input + pendingDelta.input,
+                    output: entry.tokens.output + pendingDelta.output,
+                    cached: entry.tokens.cached + pendingDelta.cached,
+                    reasoning: 0,
+                    written: entry.tokens.written + pendingDelta.output,
+                    total: entry.tokens.total + pendingDelta.input + pendingDelta.output,
+                };
+                entry.aiCredits += pendingDelta.credits;
+                pendingClaudeLlmDeltas.delete(traceId);
+            }
             ensureSession(entry.sessionId, resolveProjectId(getAttr(attrs, 'vcs.repository.url'), detectWorkingDir(attrs) ?? workingDir, projectId));
             upsertTrace(entry);
             rememberClaudeEntry(claudeInteractionEntries, traceId, entry);
@@ -238,6 +282,7 @@ function processSpans(spans, sessionId, projectId, workingDir) {
             const inputTokens = Number(getAttr(attrs, 'input_tokens') ?? 0);
             const outputTokens = Number(getAttr(attrs, 'output_tokens') ?? 0);
             const cachedTokens = Number(getAttr(attrs, 'cache_read_tokens') ?? 0);
+            const model = getStringAttr(attrs, 'model', 'gen_ai.request.model');
             const entry = claudeInteractionEntries.get(traceId);
             if (entry) {
                 entry.tokens = {
@@ -248,10 +293,19 @@ function processSpans(spans, sessionId, projectId, workingDir) {
                     written: entry.tokens.written + outputTokens,
                     total: entry.tokens.total + inputTokens + outputTokens,
                 };
-                entry.aiCredits += Number(getAttr(attrs, 'cost_usd') ?? 0);
+                entry.aiCredits += calcClaudeCredits({ input: inputTokens, output: outputTokens }, model);
                 upsertTrace(entry);
                 traceEvents.emit('trace:update', entry);
             }
+            else {
+                // Parent claude_code.interaction span hasn't arrived yet — buffer the delta.
+                const pending = pendingClaudeLlmDeltas.get(traceId) ?? { input: 0, output: 0, cached: 0, credits: 0 };
+                pending.input += inputTokens;
+                pending.output += outputTokens;
+                pending.cached += cachedTokens;
+                pending.credits += calcClaudeCredits({ input: inputTokens, output: outputTokens }, model);
+                rememberClaudeEntry(pendingClaudeLlmDeltas, traceId, pending);
+            }
             continue;
         }
         // ── invoke_agent span = top-level agent turn ──────────────────────────

package/dist/setup.js

@@ -147,11 +147,13 @@ function patchShellProfile(profilePath, port) {
     const block = otelEnvBlock(port);
     // Already has our block?
     if (content.includes('copilot-tracer OTLP config')) {
-        // Check if port matches
-        if (content.includes(`http://localhost:${port}`)) {
+        // Only skip if the embedded block is byte-identical to what we'd generate now —
+        // a marker + matching port isn't enough, since the block's env vars can gain new
+        // keys (e.g. Claude Code support) between tracer versions without the port changing.
+        if (content.includes(block)) {
             return { action: 'already_set' };
         }
-        // Port changed — update
+        // Block is stale (port changed, or vars were added/changed) — replace it in place
         const updated = content.replace(/# >>> copilot-tracer OTLP config[\s\S]*?# <<< copilot-tracer <<</, block);
         fs.writeFileSync(profilePath, updated, 'utf8');
         return { action: 'updated' };
@@ -175,10 +177,9 @@ function patchVSCodeSettings(settingsPath, port) {
     const envKey = 'terminal.integrated.env.osx';
     const existing = (settings[envKey] ?? {});
     const newEnv = vscodeEnvBlock(port);
-    // Check if already set correctly
-    if (existing[OTEL_ENDPOINT_KEY] === `http://localhost:${port}` &&
-        existing[OTEL_CONTENT_KEY] === 'true' &&
-        existing[OTEL_ENABLED_KEY] === 'true') {
+    // Already set correctly only if every current key/value is present — checking a
+    // hardcoded subset let newer keys (e.g. Claude Code support) silently go unset.
+    if (Object.entries(newEnv).every(([k, v]) => existing[k] === v)) {
         return { action: 'already_set' };
     }
     const wasSet = !!existing[OTEL_ENDPOINT_KEY];

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "copilot-tracer",
-  "version": "1.0.6",
+  "version": "1.0.7",
   "description": "Real-time tracing and prompt-refinement companion for GitHub Copilot CLI and VS Code Copilot — tracks tokens, AI credits, tool calls, and refined prompts with console and web UI",
   "keywords": [
     "copilot",

package/web/index.html

@@ -321,7 +321,7 @@ Reply with ONLY the rewritten prompt. No explanation, no preamble.`;
 
     <div class="stats-bar">
       <div class="stat"><span class="stat-label">Total Prompts</span><span class="stat-value" id="s-prompts">0</span></div>
-      <div class="stat"><span class="stat-label">AI Credits</span><span class="stat-value credits" id="s-credits">0.00 cr | $0.0000</span></div>
+      <div class="stat"><span class="stat-label">Est Cost</span><span class="stat-value credits" id="s-credits">$0.0000</span></div>
       <div class="stat"><span class="stat-label">Total Tokens</span><span class="stat-value tokens" id="s-tokens">0</span></div>
       <div class="stat"><span class="stat-label">Cached</span><span class="stat-value tokens" id="s-cached">0</span></div>
       <div class="stat"><span class="stat-label">Reasoning</span><span class="stat-value" style="color:#d2a8ff" id="s-reasoning">0</span></div>
@@ -338,7 +338,7 @@ Reply with ONLY the rewritten prompt. No explanation, no preamble.`;
             <tr>
               <th>Date / Time</th>
               <th>Prompt</th>
-              <th>AI Credits</th>
+              <th>Est Cost</th>
               <th>Duration</th>
               <th>Cached Tokens</th>
               <th>Written Tokens</th>
@@ -488,7 +488,7 @@ Reply with ONLY the rewritten prompt. No explanation, no preamble.`;
         <div class="dash-total-value tokens">${data.totals.tokens.toLocaleString()}</div>
       </div>
       <div class="dash-total-card">
-        <div class="dash-total-label">Total AI Credits</div>
+        <div class="dash-total-label">Total Est Cost</div>
         <div class="dash-total-value credits">${fmtCredits(data.totals.credits)}</div>
       </div>
     `;
@@ -525,7 +525,7 @@ Reply with ONLY the rewritten prompt. No explanation, no preamble.`;
               <span class="dash-project-stat-value tokens">${p.totalTokens.toLocaleString()}</span>
             </div>
             <div class="dash-project-stat">
-              <span class="dash-project-stat-label">AI Credits</span>
+              <span class="dash-project-stat-label">Est Cost</span>
               <span class="dash-project-stat-value credits">${fmtCredits(p.totalCredits)}</span>
             </div>
           </div>
@@ -559,7 +559,7 @@ Reply with ONLY the rewritten prompt. No explanation, no preamble.`;
     return ms < 1000 ? ms + 'ms' : (ms/1000).toFixed(1) + 's';
   }
 
-  function fmtCredits(c) { return (c||0).toFixed(2) + ' cr | $' + ((c||0)*0.01).toFixed(4); }
+  function fmtCredits(c) { return '$' + ((c||0)*0.01).toFixed(4); }
 
   function fmtDateTime(dt) {
     if (!dt) return '—';
@@ -650,7 +650,7 @@ Written:   ${(t.tokens?.written||0).toLocaleString()}
 Reasoning: ${(t.tokens?.reasoning||0).toLocaleString()}
 Output:    ${(t.tokens?.output||0).toLocaleString()}
 Total:     ${(t.tokens?.total||0).toLocaleString()}
-AI Credits: ${fmtCredits(t.aiCredits)}
+Est Cost:  ${fmtCredits(t.aiCredits)}
         </div>
       </div>
       ${t.toolCalls?.length ? `
@@ -725,14 +725,14 @@ AI Credits: ${fmtCredits(t.aiCredits)}
   function showCreditDetail(id) {
     const t = traces[id];
     if (!t) return;
-    document.getElementById('modal-title').textContent = `AI Credits: ${fmtCredits(t.aiCredits)}`;
+    document.getElementById('modal-title').textContent = `Est Cost: ${fmtCredits(t.aiCredits)}`;
     document.getElementById('modal-body').innerHTML = `
       <div class="detail-box" style="line-height:2">
 Input tokens:     ${(t.tokens?.input||0).toLocaleString()} × $0.003/1k = $${((t.tokens?.input||0)/1000*0.003).toFixed(5)}
 Written tokens:   ${(t.tokens?.written||0).toLocaleString()} × $0.006/1k = $${((t.tokens?.written||0)/1000*0.006).toFixed(5)}
 Reasoning tokens: ${(t.tokens?.reasoning||0).toLocaleString()} × $0.009/1k = $${((t.tokens?.reasoning||0)/1000*0.009).toFixed(5)}
 ─────────────────────────────────────────
-Total AI Credits: ${fmtCredits(t.aiCredits)}
+Total Est Cost:   ${fmtCredits(t.aiCredits)}
 Duration:         ${fmtDuration(t.durationMs)}
       </div>
     `;