mohdel 0.102.0 → 0.103.0

package/js/session/adapters/anthropic.js

@@ -203,12 +203,18 @@ export async function * anthropic (envelope, deps = {}) {
   }
 
   const end = String(process.hrtime.bigint())
-  // Estimate thinking tokens from streamed thinking_delta char count
-  // (Anthropic API doesn't report them separately). Cap at total
-  // output tokens reported by usage.
+  // Estimate thinking tokens. Primary path: count streamed thinking_delta
+  // chars (sonnet emits these). Fallback: gap between Anthropic's reported
+  // output_tokens and what actually streamed as visible output (text +
+  // tool input JSON) — catches redacted_thinking blocks (opus 4.7 default)
+  // that consume output tokens but emit no streaming deltas.
+  const streamedOutput = currentOutput()
+  const streamedOutputChars = streamedOutput.length +
+    [...toolBlocks.values()].reduce((s, b) => s + b.inputJson.length, 0)
+  const streamedOutputTokens = Math.ceil(streamedOutputChars / ANTHROPIC_THINKING_CHARS_PER_TOKEN)
   const estimatedThinkingTokens = thinkingChars > 0
     ? Math.min(Math.ceil(thinkingChars / ANTHROPIC_THINKING_CHARS_PER_TOKEN), outputTokens)
-    : 0
+    : Math.max(0, outputTokens - streamedOutputTokens)
   const messageOutputTokens = Math.max(0, outputTokens - estimatedThinkingTokens)
 
   /** @type {import('#core/events.js').DoneEvent} */
@@ -216,7 +222,7 @@ export async function * anthropic (envelope, deps = {}) {
     type: 'done',
     result: {
       status,
-      output: currentOutput() || null,
+      output: streamedOutput || null,
       inputTokens,
       outputTokens: messageOutputTokens,
       thinkingTokens: estimatedThinkingTokens,

package/js/session/run.js

@@ -264,12 +264,16 @@ export async function * run (envelope, {
 function normalizeModelEffort (envelope, resolveSpec) {
   const candidate = effortOf(envelope.model)
   if (!candidate) return { envelope }
-  if (envelope.outputEffort) return { envelope } // explicit wins
 
   const base = catalogKey(envelope.model)
   const baseSpec = resolveSpec(base)
   if (!baseSpec) return { envelope } // base not known — let full string fall through to not-found
 
+  // Explicit outputEffort wins; still strip the suffix so spans/logs see the canonical id.
+  if (envelope.outputEffort) {
+    return { envelope: { ...envelope, model: base } }
+  }
+
   if (!baseSpec.thinkingEffortLevels) {
     return {
       envelope,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "mohdel",
-  "version": "0.102.0",
+  "version": "0.103.0",
   "license": "MIT",
   "author": {
     "name": "Christophe Le Bars",
@@ -87,7 +87,7 @@
     "@opentelemetry/exporter-trace-otlp-grpc": "^0.217.0",
     "@opentelemetry/sdk-node": "^0.217.0",
     "chalk": "^5.4.0",
-    "mohdel-thin-gate-linux-x64-gnu": "0.102.0"
+    "mohdel-thin-gate-linux-x64-gnu": "0.103.0"
   },
   "dependencies": {
     "@anthropic-ai/sdk": "^0.95.1",