octarin-cli 0.3.2 → 0.3.4

package/assets/backfill.py

@@ -290,6 +290,14 @@ def parse_claude_transcript(  # noqa: PLR0915 - top-down jsonl parser; splitting
                                     :_OUTPUT_CAP
                                 ],
                                 "status": "ok",
+                                # subagent turns bill against the session model;
+                                # carry it so per-model analytics can attribute
+                                # their tokens
+                                **(
+                                    {"model": model}
+                                    if stype == "agent" and model
+                                    else {}
+                                ),
                             }
                         )
         prev_ts = ts

package/assets/claude_code/hook.py

@@ -347,6 +347,11 @@ def build_spans(entries: list[dict]) -> tuple[list[dict], dict, list[str], str |
 
     pending_user_text = ""
     pending_user_attachments: list[dict] = []
+    # One API generation streams as SEVERAL transcript entries (one per content
+    # block — text, then each tool_use) that share the same message id and each
+    # repeat the generation's FULL usage. Merge them into ONE span keyed by that
+    # id: usage/cost counted once, outputs concatenated, tool children attached.
+    llm_span_by_id: dict[str, dict] = {}
     # ts of the previous transcript entry; the LLM call started when the user
     # prompt / tool result landed, finished when the assistant message appears.
     prev_ts: str | None = None
@@ -373,6 +378,31 @@ def build_spans(entries: list[dict]) -> tuple[list[dict], dict, list[str], str |
         span_id = _msg(entry).get("id") or uuid.uuid4().hex
         out_text = _truncate(_text(content))
 
+        existing = llm_span_by_id.get(str(span_id))
+        if existing is not None:
+            # Continuation entry of an already-seen generation: extend the span,
+            # never re-count its usage (each entry repeats the full totals).
+            existing["end_time"] = ts
+            if out_text:
+                joined = (
+                    f"{existing['output']}\n{out_text}" if existing["output"] else out_text
+                )
+                existing["output"] = _truncate(joined)
+            for tu in _blocks(content, "tool_use"):
+                _append_tool_span(
+                    spans,
+                    totals,
+                    tu,
+                    parent_span_id=str(span_id),
+                    ts=ts,
+                    results_by_id=results_by_id,
+                    errors_by_id=errors_by_id,
+                    result_ts_by_id=result_ts_by_id,
+                    attachments_by_tool_id=attachments_by_tool_id,
+                )
+            prev_ts = ts
+            continue
+
         in_tok = usage.get("input", 0)
         out_tok = usage.get("output", 0)
         cache_r = usage.get("cache_read", 0)
@@ -399,6 +429,7 @@ def build_spans(entries: list[dict]) -> tuple[list[dict], dict, list[str], str |
         if pending_user_attachments:
             llm_span["attachments"] = pending_user_attachments
         spans.append(llm_span)
+        llm_span_by_id[str(span_id)] = llm_span
         pending_user_text = ""  # consumed by this generation
         pending_user_attachments = []  # consumed by this generation
 
@@ -408,29 +439,17 @@ def build_spans(entries: list[dict]) -> tuple[list[dict], dict, list[str], str |
         totals["total_tokens"] += in_tok + out_tok
 
         for tu in _blocks(content, "tool_use"):
-            tid = str(tu.get("id") or uuid.uuid4().hex)
-            tname = tu.get("name") or "unknown"
-            tu_input = tu.get("input")
-            input_str = (
-                tu_input if isinstance(tu_input, str) else json.dumps(tu_input, ensure_ascii=False)
+            _append_tool_span(
+                spans,
+                totals,
+                tu,
+                parent_span_id=str(span_id),
+                ts=ts,
+                results_by_id=results_by_id,
+                errors_by_id=errors_by_id,
+                result_ts_by_id=result_ts_by_id,
+                attachments_by_tool_id=attachments_by_tool_id,
             )
-            tool_span = {
-                "span_id": tid,
-                "parent_span_id": str(span_id),
-                "name": f"Tool: {tname}",
-                "span_type": "tool",
-                "start_time": ts,
-                "end_time": result_ts_by_id.get(tid, ts),
-                "input": _truncate(input_str),
-                "output": _truncate(results_by_id.get(tid, "")) or None,
-                "status": "error" if errors_by_id.get(tid) else "ok",
-                "attributes": {"tool_name": tname, "tool_id": tid},
-            }
-            tool_atts = attachments_by_tool_id.get(tid)
-            if tool_atts:
-                tool_span["attachments"] = tool_atts
-            spans.append(tool_span)
-            totals["tool_call_count"] += 1
 
         prev_ts = ts
 
@@ -438,6 +457,42 @@ def build_spans(entries: list[dict]) -> tuple[list[dict], dict, list[str], str |
     return spans, totals, models, None
 
 
+def _append_tool_span(
+    spans: list[dict],
+    totals: dict,
+    tu: dict,
+    *,
+    parent_span_id: str,
+    ts: str,
+    results_by_id: dict[str, str],
+    errors_by_id: dict[str, bool],
+    result_ts_by_id: dict[str, str],
+    attachments_by_tool_id: dict[str, list[dict]],
+) -> None:
+    """Append one ``tool`` child span for a ``tool_use`` block to ``spans``."""
+    tid = str(tu.get("id") or uuid.uuid4().hex)
+    tname = tu.get("name") or "unknown"
+    tu_input = tu.get("input")
+    input_str = tu_input if isinstance(tu_input, str) else json.dumps(tu_input, ensure_ascii=False)
+    tool_span = {
+        "span_id": tid,
+        "parent_span_id": parent_span_id,
+        "name": f"Tool: {tname}",
+        "span_type": "tool",
+        "start_time": ts,
+        "end_time": result_ts_by_id.get(tid, ts),
+        "input": _truncate(input_str),
+        "output": _truncate(results_by_id.get(tid, "")) or None,
+        "status": "error" if errors_by_id.get(tid) else "ok",
+        "attributes": {"tool_name": tname, "tool_id": tid},
+    }
+    tool_atts = attachments_by_tool_id.get(tid)
+    if tool_atts:
+        tool_span["attachments"] = tool_atts
+    spans.append(tool_span)
+    totals["tool_call_count"] += 1
+
+
 def user_ref() -> str:
     """Resolve the engineer's real identity for attribution.
 

package/assets/cursor/lib/canonical.js

@@ -128,6 +128,9 @@ function toolSpanId(input) {
 function fromPostToolUse(input) {
   const tool = input.tool_name || "tool";
   const span = baseSpan(toolSpanId(input), tool, "tool");
+  // Cursor never exposes token usage, so the backend estimates tool spans from
+  // their I/O — stamp the session model so the estimate prices at its real rate.
+  span.model = input.model || null;
   span.input = truncate(asText(input.tool_input));
   span.output = truncate(asText(input.tool_output));
   span.attributes = {
@@ -144,6 +147,7 @@ function fromPostToolUse(input) {
 function fromPostToolUseFailure(input) {
   const tool = input.tool_name || "tool";
   const span = baseSpan(toolSpanId(input), tool, "tool");
+  span.model = input.model || null;
   span.status = "error";
   span.error_message = input.error_message || input.failure_type || "tool failed";
   span.input = truncate(asText(input.tool_input));

package/assets/repo-template/dot-cursor/hooks/lib/canonical.js

@@ -128,6 +128,9 @@ function toolSpanId(input) {
 function fromPostToolUse(input) {
   const tool = input.tool_name || "tool";
   const span = baseSpan(toolSpanId(input), tool, "tool");
+  // Cursor never exposes token usage, so the backend estimates tool spans from
+  // their I/O — stamp the session model so the estimate prices at its real rate.
+  span.model = input.model || null;
   span.input = truncate(asText(input.tool_input));
   span.output = truncate(asText(input.tool_output));
   span.attributes = {
@@ -144,6 +147,7 @@ function fromPostToolUse(input) {
 function fromPostToolUseFailure(input) {
   const tool = input.tool_name || "tool";
   const span = baseSpan(toolSpanId(input), tool, "tool");
+  span.model = input.model || null;
   span.status = "error";
   span.error_message = input.error_message || input.failure_type || "tool failed";
   span.input = truncate(asText(input.tool_input));

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "octarin-cli",
-  "version": "0.3.2",
+  "version": "0.3.4",
   "description": "Octarin's per-user CLI: install AI-coding capture (`octarin init` / `init-repo`) and authorize a machine (`octarin login`). Streams your Claude Code / Cursor / Codex usage to your Octarin workspace.",
   "keywords": [
     "octarin",