behavior-wrapped 0.8.0 → 0.8.2

package/fixtures/codex-sessions/2026/06/07/rollout-2026-06-07T10-00-00-synthetic.jsonl

@@ -4,5 +4,5 @@
 {"timestamp":"2026-06-07T10:00:05.000Z","type":"response_item","payload":{"type":"message","role":"assistant","content":[{"type":"output_text","text":"Let me check that carefully."}]}}
 {"timestamp":"2026-06-07T10:00:07.000Z","type":"response_item","payload":{"type":"function_call","name":"exec_command","call_id":"synthetic-call-1","arguments":"{}"}}
 {"timestamp":"2026-06-07T10:00:09.000Z","type":"response_item","payload":{"type":"function_call_output","call_id":"synthetic-call-1","output":"synthetic build passed"}}
-{"timestamp":"2026-06-07T10:00:10.000Z","type":"event_msg","payload":{"type":"token_count","info":{"total_token_usage":{"input_tokens":800,"output_tokens":200,"cache_write_input_tokens":0,"cached_input_tokens":1000,"reasoning_output_tokens":50,"total_tokens":2050}}}}
+{"timestamp":"2026-06-07T10:00:10.000Z","type":"event_msg","payload":{"type":"token_count","info":{"total_token_usage":{"input_tokens":1800,"output_tokens":250,"cache_write_input_tokens":0,"cached_input_tokens":1000,"reasoning_output_tokens":50,"total_tokens":2050}}}}
 {"timestamp":"2026-06-07T10:00:12.000Z","type":"event_msg","payload":{"type":"task_complete","started_at":"2026-06-07T10:00:01.000Z","completed_at":"2026-06-07T10:00:12.000Z","duration_ms":11000,"turn_id":"synthetic-turn-1"}}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "behavior-wrapped",
-  "version": "0.8.0",
+  "version": "0.8.2",
   "description": "A private, local-first Wrapped report for Claude Code, Cowork, and Codex behavior.",
   "license": "Apache-2.0",
   "type": "module",

package/server/analysis.mjs

@@ -1,6 +1,7 @@
 import crypto from "node:crypto";
 import { safeEvidenceText, redactText } from "./privacy.mjs";
 import { isFrustratedMessage, isGratefulMessage } from "./frustration-card.mjs";
+import { estimateModelUsageCost } from "./model-pricing.mjs";
 import { displayModelName } from "./model-names.mjs";
 
 export { displayModelName } from "./model-names.mjs";
@@ -251,15 +252,6 @@ function finding(kind, title, summary, method, score, evidence) {
   return { id: crypto.randomUUID(), kind, title, summary, method, confidence: confidence(score), evidence };
 }
 
-function ratesFor(model, agent) {
-  const value = String(model || "").toLowerCase();
-  if (value.includes("opus")) return { input: 5, output: 25, cacheWrite: 6.25, cacheRead: 0.5 };
-  if (value.includes("sonnet")) return { input: 3, output: 15, cacheWrite: 3.75, cacheRead: 0.3 };
-  if (value.includes("haiku")) return { input: 1, output: 5, cacheWrite: 1.25, cacheRead: 0.1 };
-  if (agent === "codex" || value.startsWith("gpt")) return { input: 1.25, output: 10, cacheWrite: 1.25, cacheRead: 0.125 };
-  return { input: 3, output: 15, cacheWrite: 3.75, cacheRead: 0.3 };
-}
-
 function analyzeBehavior(sessionRecords) {
   const findings = [];
   for (const { sessionId, records } of sessionRecords) {
@@ -403,8 +395,7 @@ export function analyzeSessions(sessionRecords) {
         tokenBreakdown.reasoning += reasoning;
         const model = record?.message?.model || `${agent === "codex" ? "Codex" : "Claude"} model`;
         modelTokens.set(model, (modelTokens.get(model) || 0) + recordTokens);
-        const rates = ratesFor(model, agent);
-        estimatedCostUsd += (input * rates.input + (output + reasoning) * rates.output + cacheWrite * rates.cacheWrite + cacheRead * rates.cacheRead) / 1_000_000;
+        estimatedCostUsd += estimateModelUsageCost(usage, model, agent);
       }
       for (const tool of toolUses(record)) {
         toolCalls++;
@@ -460,7 +451,7 @@ export function analyzeSessions(sessionRecords) {
     agents,
     models,
     estimatedCostUsd: Number(estimatedCostUsd.toFixed(2)),
-    costEstimateMethod: "API-equivalent estimate using a local, inspectable model-family rate table.",
+    costEstimateMethod: "API-equivalent estimate using current standard list prices by exact model, including distinct cache-read, 5-minute cache-write, and 1-hour cache-write rates.",
   };
   return { stats, findings: analyzeBehavior(sessionRecords) };
 }

package/server/discovery.mjs

@@ -458,8 +458,8 @@ function normalizeCodexRecords(records, { includePrivateToolDetails = false } =
         type: "system",
         timestamp: record.timestamp,
         message: { model: currentModel, usage: {
-          input_tokens: usage.input_tokens,
-          output_tokens: usage.output_tokens,
+          input_tokens: Math.max(0, usage.input_tokens - usage.cached_input_tokens - usage.cache_write_input_tokens),
+          output_tokens: Math.max(0, usage.output_tokens - usage.reasoning_output_tokens),
           cache_creation_input_tokens: usage.cache_write_input_tokens,
           cache_read_input_tokens: usage.cached_input_tokens,
           reasoning_output_tokens: usage.reasoning_output_tokens,

package/server/model-pricing.mjs

@@ -0,0 +1,61 @@
+// Standard API list prices in USD per million tokens, verified 2026-08-26.
+// OpenAI: https://developers.openai.com/api/docs/models/compare
+// Anthropic: https://platform.claude.com/docs/en/about-claude/pricing
+const scaled = (value, multiplier) => Number((value * multiplier).toFixed(12));
+
+const price = (input, output, cacheRead, cacheWrite5m, cacheWrite1h) => ({
+  input,
+  output,
+  cacheRead: cacheRead ?? scaled(input, 0.1),
+  cacheWrite5m: cacheWrite5m ?? scaled(input, 1.25),
+  cacheWrite1h: cacheWrite1h ?? scaled(input, 2),
+});
+
+const modelRates = [
+  [/gpt-5\.6-sol/, price(4, 20)],
+  [/gpt-5\.6-terra/, price(2, 12)],
+  [/gpt-5\.6-luna/, price(0.2, 1.2)],
+  [/gpt-5\.5-pro/, price(30, 180, 0, 30, 30)],
+  [/gpt-5\.5/, price(5, 30)],
+  [/gpt-5\.4-pro/, price(30, 180, 0, 30, 30)],
+  [/gpt-5\.4-mini/, price(0.75, 4.5)],
+  [/gpt-5\.4-nano/, price(0.2, 1.25)],
+  [/gpt-5\.4/, price(2.5, 15)],
+  [/gpt-5\.3-codex/, price(1.75, 14)],
+  [/gpt-5\.2/, price(1.75, 14)],
+  [/(?:claude-)?(?:fable|mythos)(?:-|\s)?5/, price(10, 50)],
+  [/(?:claude-)?opus(?:-|\s)?5/, price(5, 25)],
+  [/(?:claude-)?opus(?:-|\s)?4(?:-|\s)?(?:8|7|6|5)/, price(5, 25)],
+  [/(?:claude-)?opus(?:-|\s)?4(?:-|\s)?1/, price(15, 75)],
+  [/(?:claude-)?opus(?:-|\s)?4(?:\b|-20)/, price(15, 75)],
+  [/(?:claude-)?sonnet(?:-|\s)?5/, price(2, 10)],
+  [/(?:claude-)?sonnet(?:-|\s)?4/, price(3, 15)],
+  [/(?:claude-)?haiku(?:-|\s)?4(?:-|\s)?5/, price(1, 5)],
+  [/(?:claude-)?haiku(?:-|\s)?3(?:-|\s)?5/, price(0.8, 4)],
+];
+
+export function ratesFor(model, agent) {
+  const value = String(model || "").toLowerCase();
+  const match = modelRates.find(([pattern]) => pattern.test(value));
+  if (match) return match[1];
+  return agent === "codex" || value.startsWith("gpt") ? price(2.5, 15) : price(3, 15);
+}
+
+export function estimateModelUsageCost(usage, model, agent) {
+  const input = Number(usage?.input_tokens) || 0;
+  const output = Number(usage?.output_tokens) || 0;
+  const reasoning = Number(usage?.reasoning_output_tokens) || 0;
+  const cacheRead = Number(usage?.cache_read_input_tokens) || 0;
+  const cacheWrite = Number(usage?.cache_creation_input_tokens) || 0;
+  const declared5m = Number(usage?.cache_creation?.ephemeral_5m_input_tokens) || 0;
+  const declared1h = Number(usage?.cache_creation?.ephemeral_1h_input_tokens) || 0;
+  const cacheWrite5m = declared5m + Math.max(0, cacheWrite - declared5m - declared1h);
+  const rates = ratesFor(model, agent);
+  return (
+    input * rates.input
+    + (output + reasoning) * rates.output
+    + cacheRead * rates.cacheRead
+    + cacheWrite5m * rates.cacheWrite5m
+    + declared1h * rates.cacheWrite1h
+  ) / 1_000_000;
+}