okstra 0.179.2 → 0.183.2

package/runtime/python/okstra_token_usage/collect.py

@@ -14,16 +14,17 @@ from .claude import (
     find_claude_team_sessions,
 )
 from .codex import codex_session_total, find_codex_sessions
-from .antigravity import (
-    PRINT_MODE_USAGE_NOTE,
-    antigravity_session_total,
-    find_antigravity_sessions,
-)
+from .antigravity import antigravity_session_total, find_antigravity_sessions
+from .grok import find_grok_sessions, grok_session_total
 from .paths import claude_project_dir, utc_now
 from .pricing import antigravity_cost_usd, provider_cost_usd
 from okstra_ctl.dispatch_state import worker_session_ids
 from okstra_ctl.models import provider_wrappers
-from okstra_ctl.wrapper_status import read_wrapper_status, status_path_for_prompt
+from okstra_ctl.wrapper_status import (
+    log_path_for_prompt,
+    read_wrapper_status,
+    status_path_for_prompt,
+)
 
 
 def match_prefixes(worker_id: str) -> list[str]:
@@ -472,6 +473,8 @@ def _cli_sessions_for_windows(
             matches = find_codex_sessions(project_root, started_at, ended_at)
         elif provider == "antigravity":
             matches = find_antigravity_sessions(project_root, started_at, ended_at)
+        elif provider == "grok":
+            matches = find_grok_sessions(project_root, started_at, ended_at)
         else:
             matches = []
         for path in matches:
@@ -488,6 +491,8 @@ def _cli_session_totals(provider: str, session_paths: list[Path]) -> list[dict]:
             total = codex_session_total(session_path)
         elif provider == "antigravity":
             total = antigravity_session_total(session_path)
+        elif provider == "grok":
+            total = grok_session_total(session_path)
         else:
             continue
         if total.get("available"):
@@ -582,24 +587,20 @@ def collect_cli_usage(
     execution = wrapper_execution(status_path)
     execution_note = _execution_note(execution)
 
-    if provider == "antigravity":
-        block = na_block(f"{PRINT_MODE_USAGE_NOTE}; {execution_note}")
+    totals = _cli_session_totals(provider, sessions)
+    if not sessions:
+        block = na_block(
+            f"{execution_note}; CLI usage attribution unavailable because no transcript was found"
+        )
         block["cliNote"] = block["note"]
+    elif not totals:
+        block = na_block(
+            f"{execution_note}; transcript found but no final token snapshot was recorded"
+        )
+        block["cliNote"] = block["note"]
+        block["cliSessionPaths"] = [str(path) for path in sessions]
     else:
-        totals = _cli_session_totals(provider, sessions)
-        if not sessions:
-            block = na_block(
-                f"{execution_note}; CLI usage attribution unavailable because no transcript was found"
-            )
-            block["cliNote"] = block["note"]
-        elif not totals:
-            block = na_block(
-                f"{execution_note}; transcript found but no final token snapshot was recorded"
-            )
-            block["cliNote"] = block["note"]
-            block["cliSessionPaths"] = [str(path) for path in sessions]
-        else:
-            block = _cli_usage_block(provider, _aggregate_totals(totals), sessions)
+        block = _cli_usage_block(provider, _aggregate_totals(totals), sessions)
 
     block["cliExecutionStatus"] = execution["status"]
     if fallback_window_used:
@@ -624,6 +625,51 @@ def _worker_cli_windows(
     return fallback_windows, status_path, True
 
 
+def _antigravity_log_paths(
+    project_root: Path,
+    worker: dict,
+    status_path: Path | None,
+) -> list[Path]:
+    """agy stream-json 은 워커 로그에 남는다. 홈 세션 디렉터리가 아니다."""
+    paths: list[Path] = []
+    prompt_path = _resolve_project_path(project_root, str(worker.get("promptPath") or ""))
+    if prompt_path is not None:
+        log_path = log_path_for_prompt(prompt_path)
+        if log_path.is_file():
+            paths.append(log_path)
+    status = read_wrapper_status(status_path) if status_path else None
+    if status is not None and status.log_path is not None and status.log_path.is_file():
+        if status.log_path not in paths:
+            paths.append(status.log_path)
+    return paths
+
+
+def _worker_cli_usage_block(
+    *,
+    provider: str,
+    project_root: Path,
+    worker: dict,
+    fallback_windows: list[tuple[str, str]],
+) -> dict:
+    """공급자 프로세스 트랜스크립트와 실행 증거로 워커 사용량 블록을 만든다."""
+    windows, status_path, used_fallback = _worker_cli_windows(
+        project_root,
+        worker,
+        fallback_windows,
+    )
+    session_paths = _cli_sessions_for_windows(provider, project_root, windows)
+    if provider == "antigravity":
+        for log_path in _antigravity_log_paths(project_root, worker, status_path):
+            if log_path not in session_paths:
+                session_paths.append(log_path)
+    return collect_cli_usage(
+        provider=provider,
+        status_path=status_path,
+        sessions=session_paths,
+        fallback_window_used=used_fallback,
+    )
+
+
 def _attach_cli_usage(
     block: dict,
     provider: str,
@@ -638,8 +684,8 @@ def _attach_cli_usage(
     window produces several rollout jsonls; pricing only the latest (the old
     ``find_codex_session`` -> ``sessions[-1]`` behavior) dropped the earlier
     attempts' tokens. We sum every in-window session here so the redispatched
-    CLI spend is fully reported. (agy print mode leaves no transcript, so the
-    antigravity path still resolves to "session not found" → na cost.)
+    CLI spend is fully reported. antigravity 는 워커 로그의 stream-json
+    ``result.usage`` 를 같은 합산 경로에 붙인다.
     """
     session_paths = _cli_sessions_for_windows(provider, project_root, windows)
     cli = collect_cli_usage(
@@ -676,17 +722,11 @@ def _collect_cli_runtime_usage(state: dict, project_root: Path) -> dict:
             )
             continue
         worker_id = str(worker.get("workerId") or "").strip()
-        windows, status_path, used_fallback = _worker_cli_windows(
-            project_root,
-            worker,
-            windows_by_worker.get(worker_id, []),
-        )
-        session_paths = _cli_sessions_for_windows(provider, project_root, windows)
-        worker["usage"] = collect_cli_usage(
+        worker["usage"] = _worker_cli_usage_block(
             provider=provider,
-            status_path=status_path,
-            sessions=session_paths,
-            fallback_window_used=used_fallback,
+            project_root=project_root,
+            worker=worker,
+            fallback_windows=windows_by_worker.get(worker_id, []),
         )
     _populate_usage_summary(state, team_name=resolve_team_name(state),
                             sessions_found=0, needle_source="none")
@@ -871,6 +911,7 @@ def collect_claude_runtime_usage(
     sessions_dir = claude_project_dir(cwd)
     # sid 로 귀속한 세션 id 전체 — 아래 unattributed 폴드에서 빼는 데 쓴다.
     attributed_sids: set[str] = set()
+    cli_windows_by_worker = _codex_worker_windows(cwd, state)
     for worker in state.get("workers", []):
         worker_id = worker.get("workerId")
         agent = worker.get("agent")
@@ -922,10 +963,21 @@ def collect_claude_runtime_usage(
                 matched_sids.add(entry[0])
 
         if not matched:
-            worker["usage"] = na_block(
-                "no Claude session jsonl in the run window for dispatched "
-                f"sessionIds {dispatched_sids} or agentName prefixes {prefixes}"
-            )
+            provider = _cli_assignment_provider(worker)
+            if provider and str(worker.get("runner") or "") == "cli-wrapper":
+                worker["usage"] = _worker_cli_usage_block(
+                    provider=provider,
+                    project_root=cwd,
+                    worker=worker,
+                    fallback_windows=cli_windows_by_worker.get(
+                        str(worker_id or "").strip(), []
+                    ),
+                )
+            else:
+                worker["usage"] = na_block(
+                    "no Claude session jsonl in the run window for dispatched "
+                    f"sessionIds {dispatched_sids} or agentName prefixes {prefixes}"
+                )
             continue
 
         # Stable order by startedAt so the "primary" session is the first one.

package/runtime/python/okstra_token_usage/grok.py

@@ -0,0 +1,127 @@
+"""Grok Build session collectors.
+
+공식 세션 문서는 대화를 ``~/.grok/sessions/`` 에 둔다. cwd 는 퍼센트 인코딩된
+디렉터리 이름이다. 누적 사용량은 ``updates.jsonl`` 의
+``params.update.usage.modelUsage`` 마지막 스냅샷이다.
+"""
+from __future__ import annotations
+
+import os
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Mapping
+from urllib.parse import quote, unquote
+
+from .jsonl_io import iter_jsonl
+from .paths import ts_in_window
+
+
+def grok_sessions_root(
+    home: Path | None = None,
+    env: Mapping[str, str] | None = None,
+) -> Path:
+    environ = env if env is not None else os.environ
+    configured = str(environ.get("GROK_HOME") or "").strip()
+    base = Path(configured).expanduser() if configured else (home or Path.home()) / ".grok"
+    return base / "sessions"
+
+
+def _iso_from_unix(value: object) -> str | None:
+    if not isinstance(value, (int, float)) or isinstance(value, bool):
+        return None
+    return datetime.fromtimestamp(value, timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ")
+
+
+def _usage_payload(record: dict) -> dict | None:
+    params = record.get("params")
+    if not isinstance(params, dict):
+        return None
+    update = params.get("update")
+    if not isinstance(update, dict):
+        return None
+    usage = update.get("usage")
+    return usage if isinstance(usage, dict) else None
+
+
+def _model_snapshot(usage: dict) -> tuple[dict, str | None]:
+    models = usage.get("modelUsage")
+    if isinstance(models, dict) and len(models) == 1:
+        name, payload = next(iter(models.items()))
+        if isinstance(payload, dict):
+            return payload, name if isinstance(name, str) else None
+    return usage, None
+
+
+def grok_session_total(updates_path: Path) -> dict:
+    """마지막 modelUsage 스냅샷. 세션 누적이라 더하지 않는다."""
+    last: dict | None = None
+    model: str | None = None
+    started: str | None = None
+    ended: str | None = None
+    for record in iter_jsonl(updates_path):
+        usage = _usage_payload(record)
+        if usage is None:
+            continue
+        iso = _iso_from_unix(record.get("timestamp"))
+        if iso and started is None:
+            started = iso
+        if iso:
+            ended = iso
+        snapshot, snapshot_model = _model_snapshot(usage)
+        last = snapshot
+        if snapshot_model:
+            model = snapshot_model
+    if last is None:
+        return {"totalTokens": 0, "available": False}
+    return {
+        "totalTokens": last.get("totalTokens", 0) or 0,
+        "inputTokens": last.get("inputTokens", 0) or 0,
+        "outputTokens": last.get("outputTokens", 0) or 0,
+        "cachedInputTokens": last.get("cachedReadTokens", 0) or 0,
+        "reasoningOutputTokens": last.get("reasoningTokens", 0) or 0,
+        "durationMs": last.get("apiDurationMs", 0) or 0,
+        "model": model,
+        "startedAt": started,
+        "endedAt": ended,
+        "available": True,
+    }
+
+
+def _session_in_window(updates_path: Path, started_at: str, ended_at: str) -> bool:
+    for record in iter_jsonl(updates_path):
+        iso = _iso_from_unix(record.get("timestamp"))
+        if iso and ts_in_window(iso, started_at, ended_at):
+            return True
+    return False
+
+
+def find_grok_sessions(
+    cwd: Path,
+    started_at: str,
+    ended_at: str,
+    *,
+    session_root: Path | None = None,
+) -> list[Path]:
+    """cwd 로 인코딩된 세션 중 창 안에 갱신이 있는 updates.jsonl."""
+    if not started_at or not ended_at:
+        return []
+    root = session_root or grok_sessions_root()
+    if not root.is_dir():
+        return []
+    target = str(cwd)
+    matches: list[Path] = []
+    for child in root.iterdir():
+        if not child.is_dir():
+            continue
+        decoded = unquote(child.name)
+        if decoded != target:
+            continue
+        for session in child.iterdir():
+            updates = session / "updates.jsonl"
+            if updates.is_file() and _session_in_window(updates, started_at, ended_at):
+                matches.append(updates)
+    return sorted(matches)
+
+
+def grok_session_dir_name(cwd: Path) -> str:
+    return quote(str(cwd), safe="")

package/runtime/schemas/final-report-v2.0.schema.json

@@ -8943,6 +8943,7 @@
       ],
       "additionalProperties": false,
       "properties": {
+        "stageRefs": {"type": "array", "minItems": 1, "uniqueItems": true, "items": {"type": "integer", "minimum": 1}},
         "id": {
           "type": "string",
           "pattern": "^VC-\\d{3,}$"
@@ -9142,6 +9143,8 @@
       ],
       "additionalProperties": false,
       "properties": {
+        "setAside": {"type": "array", "items": {"type": "object", "required": ["id", "reason"], "additionalProperties": false, "properties": {"id": {"type": "string", "minLength": 1}, "reason": {"type": "string", "enum": ["record", "observed", "deferred"]}}}},
+        "stageLedger": {"type": "object", "additionalProperties": {"type": "string", "enum": ["done", "active", "ready", "blocked"]}},
         "uniformVerifiers": {
           "type": "array",
           "description": "Verifiers whose every vote this round was one verdict. Advisory: a unanimous round is legitimate, but the gate reads as a three-way cross-check unless this sits beside it.",
@@ -9242,6 +9245,8 @@
             ],
             "additionalProperties": false,
             "properties": {
+              "block": {"type": "string", "enum": ["execution", "record"]},
+              "stageScope": {"type": "array", "minItems": 1, "uniqueItems": true, "items": {"type": "integer", "minimum": 1}},
               "id": {
                 "type": "string",
                 "minLength": 1
@@ -9280,6 +9285,9 @@
                   ],
                   "additionalProperties": false,
                   "properties": {
+                    "claimKind": {"description": "Set only on DISAGREE. `fact` claims are settled by reproduction, `judgement` by quorum.", "enum": ["fact", "judgement", null]},
+                    "reproduction": {"description": "Set only on a `fact` claim. The probe okstra runs; see okstra_ctl/claim_reproduction.py.", "type": ["object", "null"]},
+                    "reproductionResult": {"description": "Written by okstra, not by the worker.", "enum": ["reproduced", "not-reproduced", "not-runnable", null]},
                     "worker": {
                       "type": "string",
                       "minLength": 1
@@ -9752,6 +9760,7 @@
       ],
       "additionalProperties": false,
       "properties": {
+        "stageRefs": {"type": "array", "minItems": 1, "uniqueItems": true, "items": {"type": "integer", "minimum": 1}},
         "id": {
           "type": "string",
           "pattern": "^R-\\d{3,}$"