agent-learning-kit 0.2.0__tar.gz → 0.2.2__tar.gz

{agent_learning_kit-0.2.0 → agent_learning_kit-0.2.2}/CHANGELOG.md

@@ -24,6 +24,18 @@ Post-rc.1 increments on the release branch. Each one adds gates on top of the
 - Hosted voice workers preserve stdout/stderr in the parent diagnostics stream,
   with per-call correlation, turn-state transitions, transcription dispatch,
   and simulator generation lifecycle events.
+- Voice reports now seal transcripts after bounded STT/native-stream finalization,
+  retain conversation evidence on failures, and explicitly report incomplete
+  capture instead of silently completing with a partial transcript.
+- Retell and LiveKit audio bridges close owned streams and sources, reap media
+  tasks on cancellation, and drain queued playback before closing a provider call.
+- Voice diagnostics include sanitized SIP error codes, transcript-finalization
+  and cleanup results, isolated-worker exit status, and native crash backtraces.
+- Outbound SIP calls no longer block on `wait_until_answered`. LiveKit v1.9.11
+  retried that request with 3s/5s/7s deadlines, cancelling and redialing a
+  ringing call so people and slow lines could never answer; the engine now
+  waits for `sip.callStatus == "active"` in the room within the answer
+  deadline and reports a rejected call's reason (e.g. `user_unavailable`).
 
 ## v1.0.0-rc.1 — 2026-06-10
 

{agent_learning_kit-0.2.0 → agent_learning_kit-0.2.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: agent-learning-kit
-Version: 0.2.0
+Version: 0.2.2
 Summary: Unified Future AGI SDK for agent learning workflows.
 Project-URL: Homepage, https://futureagi.com
 Project-URL: Documentation, https://docs.futureagi.com

{agent_learning_kit-0.2.0 → agent_learning_kit-0.2.2}/pyproject.toml

@@ -5,7 +5,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "agent-learning-kit"
-version = "0.2.0"
+version = "0.2.2"
 description = "Unified Future AGI SDK for agent learning workflows."
 readme = "README.md"
 requires-python = ">=3.10"

{agent_learning_kit-0.2.0 → agent_learning_kit-0.2.2}/src/fi/alk/harness/authoring_runtime_validation.py

@@ -1003,6 +1003,7 @@ async def validate_and_repair(
                     count=job.scenario_count,
                     interactive=False,
                     guidance=[guidance],
+                    job=job,
                 )
             )
 

{agent_learning_kit-0.2.0 → agent_learning_kit-0.2.2}/src/fi/alk/harness/bundle_author_v2.py

@@ -1665,7 +1665,7 @@ def resolve_environment_plan(
                 environment = {**declared_runtime_environment, **environment}
                 environment.setdefault(
                     "LIVEKIT_AGENT_NAME",
-                    "uber-voice-booking-{{JOB_ID}}-w{{WORLD_INDEX}}",
+            "voice-booking-{{JOB_ID}}-w{{WORLD_INDEX}}",
                 )
                 environment.setdefault(
                     "HARNESS_TOOL_TRACE",

{agent_learning_kit-0.2.0 → agent_learning_kit-0.2.2}/src/fi/alk/harness/call_runner.py

@@ -113,6 +113,7 @@ SIMULATOR_TTS_MODEL_ALIAS = "SIMULATOR_TTS_MODEL"
 BACKGROUND_NOISE_ALIAS = "ALK_BACKGROUND_NOISE"
 BACKGROUND_NOISE_CATALOG_ALIAS = "ALK_BACKGROUND_NOISE_CATALOG"
 BACKGROUND_NOISE_VOLUME_ALIAS = "HARNESS_BACKGROUND_NOISE_VOLUME"
+CALLER_BARGE_IN_RATE_ALIAS = "HARNESS_CALLER_BARGE_IN_RATE"
 CALL_DIRECTION_ALIAS = "ALK_CALL_DIRECTION"
 VOICEMAIL_CLIP_ALIAS = "HARNESS_VOICEMAIL_CLIP"
 VOICEMAIL_CLIP_TONE_ALIAS = "HARNESS_VOICEMAIL_CLIP_HAS_TONE"
@@ -823,6 +824,7 @@ def _collect_file_tool_calls(runtime: EnvironmentRuntime) -> tuple[Call, ...]:
             arguments = {}
         is_error = bool(record.get("is_error", False))
         output = record.get("output")
+        raw_at = record.get("at")
         calls.append(
             Call(
                 name=name,
@@ -830,6 +832,7 @@ def _collect_file_tool_calls(runtime: EnvironmentRuntime) -> tuple[Call, ...]:
                 result=None if is_error else output,
                 ok=not is_error,
                 error=str(output) if is_error and output is not None else None,
+                at=float(raw_at) if isinstance(raw_at, (int, float)) else 0.0,
             )
         )
     return tuple(calls)
@@ -1063,6 +1066,7 @@ class CallRunnerImpl:
             BACKGROUND_NOISE_ALIAS,
             BACKGROUND_NOISE_CATALOG_ALIAS,
             BACKGROUND_NOISE_VOLUME_ALIAS,
+            CALLER_BARGE_IN_RATE_ALIAS,
         ):
             value = simulator_secret_values.get(alias)
             if value:
@@ -1469,12 +1473,18 @@ class CallRunnerImpl:
         if case is not None and case.result is not None:
             result = case.result
             if result.transcript:
+                transcript_document = {
+                    "schema_version": "futureagi.call-transcript.v1",
+                    "transcript": result.transcript,
+                    "messages": result.messages,
+                }
+                recording_offset_ms = result.metadata.get("recording_offset_ms")
+                if isinstance(recording_offset_ms, (int, float)):
+                    transcript_document["recording_offset_ms"] = int(
+                        recording_offset_ms
+                    )
                 transcript_payload = json.dumps(
-                    {
-                        "schema_version": "futureagi.call-transcript.v1",
-                        "transcript": result.transcript,
-                        "messages": result.messages,
-                    },
+                    transcript_document,
                     sort_keys=True,
                     default=str,
                 ).encode("utf-8")
@@ -1591,6 +1601,20 @@ class CallRunnerImpl:
             reason = (
                 case.failure.message if case.failure is not None else case.status.value
             )
+            if case.failure is not None:
+                safe_details = {
+                    key: value
+                    for key, value in (case.failure.details or {}).items()
+                    if key in {
+                        "operation", "exception_type", "provider_code", "http_status",
+                        "sip_status_code", "stop_reason", "turn_count", "minimum_turn_count",
+                    }
+                }
+                logger.warning(
+                    "voice_call_failed run=%s case=%s code=%s stage=%s details=%s",
+                    report.run_id, case.test_case_id, case.failure.code,
+                    case.failure.stage.value, json.dumps(safe_details, sort_keys=True),
+                )
             # C3 §4.5 step 2 (the one narrow call_runner change): the engine's dispatch-ack ladder
             # surfaces +60s exhaustion via a STRUCTURED marker (`failure.code`), never a substring
             # of `failure.message`. Pass it through on `CallAborted.marker` so the scheduler's

{agent_learning_kit-0.2.0 → agent_learning_kit-0.2.2}/src/fi/alk/harness/call_worker.py

@@ -2,6 +2,7 @@
 
 from __future__ import annotations
 
+import faulthandler
 import logging
 import os
 import sys
@@ -11,6 +12,8 @@ from .isolated_process import run_worker
 
 
 async def main() -> None:
+    faulthandler.enable()
+    os.environ.setdefault("RUST_BACKTRACE", "1")
     from fi.simulate.runtime import SimulationSpec
 
     from .call_runner import _default_place_call
@@ -27,7 +30,7 @@ async def main() -> None:
     fd = os.open(output, os.O_WRONLY | os.O_CREAT | os.O_EXCL, 0o600)
     with os.fdopen(fd, "w") as stream:
         stream.write(report.model_dump_json())
-    logging.getLogger(__name__).info("voice worker completed")
+    logging.getLogger(__name__).info("voice worker result written")
 
 
 if __name__ == "__main__":

{agent_learning_kit-0.2.0 → agent_learning_kit-0.2.2}/src/fi/alk/harness/chat.py

@@ -24,6 +24,7 @@ from . import understand as understand_stage
 from .backends import SessionSpec
 from .config import artifact_dir
 from .contract import AgentContract
+from .job import HarnessJob
 from .run import stage as run_stage
 from .session import Stage
 from .sources import AgentSource, resolve
@@ -65,6 +66,7 @@ class Conversation:
     _found: dict[str, Any] = field(default_factory=dict)
     configure_stage: Callable[[str, SessionSpec], SessionSpec] | None = None
     control_only: bool = False
+    job: HarnessJob | None = None
     spent_usd: float = 0.0
 
     def __post_init__(self) -> None:
@@ -190,7 +192,8 @@ class Conversation:
             wanted = written or self.wanted
             # Chat is where "add five" comes from, so only here may aim_for raise the cap.
             self.stage, _ = scenario_stage.open_stage(
-                contract, out=self.out, wanted=wanted, ask=self.ask, can_grow=True
+                contract, out=self.out, wanted=wanted, ask=self.ask, can_grow=True,
+                job=self.job,
             )
             opening = scenario_stage.opening(contract, wanted, written)
         self._prepare_stage(stage_name)
@@ -430,6 +433,7 @@ def open_conversation(
     workspace: Path | None = None,
     configure_stage: Callable[[str, SessionSpec], SessionSpec] | None = None,
     control_only: bool = False,
+    job: HarnessJob | None = None,
 ) -> Conversation:
     source = resolve(kind, name=name, root=path) if name and path else None
     return Conversation(
@@ -440,6 +444,7 @@ def open_conversation(
         workspace=workspace,
         configure_stage=configure_stage,
         control_only=control_only,
+        job=job,
     )
 
 

{agent_learning_kit-0.2.0 → agent_learning_kit-0.2.2}/src/fi/alk/harness/cli.py

@@ -14,6 +14,7 @@ import json
 import os
 import sys
 import time
+import uuid
 from pathlib import Path
 from typing import Any
 
@@ -33,7 +34,7 @@ from .run.targets import supported as target_kinds
 from .scenarios import load as load_written
 from .scenarios import open_stage as scenario_stage
 from .scenarios import opening as scenario_opening
-from .session import ARTIFACT, DONE, RESULT, TEXT, TOOL, Event
+from .session import TEXT, Event
 from .sessions import Session, new_id, save as save_session
 from .sources import resolve, supported
 from .understand import load, open_stage, opening
@@ -495,15 +496,37 @@ async def _scenarios(args: argparse.Namespace) -> int:
     existing = len(load_written(destination))
     wanted = args.count or existing or 10
 
-    # The Uber Guest Booking POC policy is supplied only through the platform-owned simulator
+    # The guest-booking POC policy is supplied only through the platform-owned simulator
     # secret channel and is gated against the exact submitted phone target. Keep it in the model's
-    # authoring brief: saved scenarios should be authored with natural PIN behavior, never rewritten
-    # mechanically after generation or intercepted while a call is running.
-    from .poc_guest_booking import guest_booking_pin_guidance
-
-    poc_guidance = guest_booking_pin_guidance(
-        getattr(args, "job", None), scenario_count=wanted
+    # authoring brief: saved scenarios should be authored with natural PIN behavior. The scenario
+    # gate attaches only private fixture facts after selection; it never rewrites scenario intent
+    # or intercepts a live caller turn.
+    from .poc_guest_booking import (
+        TARGET_PHONE_ENV,
+        guest_booking_pin_guidance,
+        guest_booking_policy_job,
     )
+
+    policy_job = None
+    job_input = getattr(args, "job", None)
+    if job_input is not None and os.environ.get(TARGET_PHONE_ENV, "").strip():
+        if isinstance(job_input, (str, os.PathLike)):
+            try:
+                raw_job = json.loads(Path(job_input).read_text(encoding="utf-8"))
+            except (OSError, ValueError) as exc:
+                print(
+                    "Private guest PIN policy disabled: job could not be read "
+                    f"({type(exc).__name__}).",
+                    file=sys.stderr,
+                )
+            else:
+                policy_job = guest_booking_policy_job(raw_job)
+        else:
+            # Internal authoring and repair paths already carry the validated HarnessJob.
+            # Do not reinterpret that object as a filesystem path.
+            policy_job = guest_booking_policy_job(job_input)
+
+    poc_guidance = guest_booking_pin_guidance(policy_job, scenario_count=wanted)
     guidance = [*(getattr(args, "guidance", None) or [])]
     if poc_guidance:
         guidance.append(poc_guidance)
@@ -521,6 +544,7 @@ async def _scenarios(args: argparse.Namespace) -> int:
         wanted=wanted,
         ask=permission_gate(_ask_operator) if args.interactive else None,
         authoring_guidance=poc_guidance,
+        job=policy_job,
     )
     await _converse(
         stage,
@@ -1412,6 +1436,7 @@ def build_parser() -> argparse.ArgumentParser:
             "while preserving existing validated work"
         ),
     )
+    scenarios.add_argument("--job", type=Path, default=None, help=argparse.SUPPRESS)
     scenarios.set_defaults(run=_scenarios, interactive=True)
 
     live = sub.add_parser(

{agent_learning_kit-0.2.0 → agent_learning_kit-0.2.2}/src/fi/alk/harness/hosted_authoring_entrypoint.py

@@ -15,6 +15,7 @@ from pathlib import Path
 
 from . import observability
 from .authoring_entrypoint import main as authoring_main
+from .poc_guest_booking import PIN_ENV, TARGET_PHONE_ENV
 
 _SECRETS_PATH = Path("/run/futureagi/secrets.json")
 _ADC_PATH = Path("/work/.authoring-credentials/google.json")
@@ -28,10 +29,10 @@ _PASSTHROUGH = {
     "ALK_VOICEMAIL_SCENARIOS",
     "ALK_BACKGROUND_NOISE",
     "ALK_BACKGROUND_NOISE_CATALOG",
-    # Temporary, private scenario-authoring policy for the Uber Guest Booking POC. The exact
+    # Temporary, private guest-booking scenario policy. The exact
     # target comes from platform deployment configuration, never from customer environment input.
-    "ALK_UBER_GUEST_POC_TARGET_PHONE_NUMBER",
-    "ALK_UBER_GUEST_POC_PIN",
+    TARGET_PHONE_ENV,
+    PIN_ENV,
     "ALK_CLAUDE_GATEWAY_URL",
     "ALK_CLAUDE_GATEWAY_API_KEY",
     "ANTHROPIC_API_KEY",
@@ -166,8 +167,8 @@ def main(argv: list[str] | None = None) -> int:
                 "AGENTCC_BASE_URL",
                 "ALK_BACKGROUND_NOISE",
                 "ALK_BACKGROUND_NOISE_CATALOG",
-                "ALK_UBER_GUEST_POC_TARGET_PHONE_NUMBER",
-                "ALK_UBER_GUEST_POC_PIN",
+                TARGET_PHONE_ENV,
+                PIN_ENV,
             )
             if gateway_values.get(name)
         }

{agent_learning_kit-0.2.0 → agent_learning_kit-0.2.2}/src/fi/alk/harness/hosted_chat_entrypoint.py

@@ -146,6 +146,8 @@ class HostedChatRuntime:
         metadata = self.job.get("metadata") or {}
         name = str(metadata.get("agent_name") or self.job.get("name") or "agent")
         if not self._control_only():
+            from .poc_guest_booking import guest_booking_policy_job
+
             return open_conversation(
                 name=name,
                 path=str(self.source),
@@ -156,6 +158,7 @@ class HostedChatRuntime:
                 ask=self._ask,
                 configure_stage=self._configure_stage,
                 control_only=False,
+                job=guest_booking_policy_job(self.job),
             )
         identity = self.client.capabilities.identity
         spec = SessionSpec(

{agent_learning_kit-0.2.0 → agent_learning_kit-0.2.2}/src/fi/alk/harness/hosted_entrypoint.py

@@ -207,6 +207,7 @@ _SIMULATOR_SECRET_ALIASES = frozenset(
         "GOOGLE_CLOUD_PROJECT",
         "GOOGLE_GENAI_USE_VERTEXAI",
         "HARNESS_BACKGROUND_NOISE_VOLUME",
+        "HARNESS_CALLER_BARGE_IN_RATE",
         "HARNESS_OBSERVABILITY",
         "LIVEKIT_URL",
         "LIVEKIT_API_KEY",

{agent_learning_kit-0.2.0 → agent_learning_kit-0.2.2}/src/fi/alk/harness/isolated_process.py

@@ -139,6 +139,10 @@ async def run_json_worker(
                 raise asyncio.CancelledError
             await process.communicate(json.dumps(payload).encode())
             if process.returncode != 0:
+                logging.getLogger(__name__).error(
+                    "isolated_worker_failed module=%s exit_status=%s result_present=%s",
+                    module, process.returncode, result_path.is_file(),
+                )
                 raise IsolatedWorkerError(
                     f"isolated worker exited with status {process.returncode}"
                 )

{agent_learning_kit-0.2.0 → agent_learning_kit-0.2.2}/src/fi/alk/harness/livekit_tool_trace_bootstrap.py

@@ -29,6 +29,7 @@ import os
 from collections.abc import Mapping
 import sys
 import threading
+import time
 from pathlib import Path
 from typing import Any
 
@@ -51,6 +52,8 @@ def _record(event: Any) -> None:
                 "is_error": bool(
                     output is not None and getattr(output, "is_error", False)
                 ),
+                # Epoch seconds when the model issued the call; places it in the transcript.
+                "at": getattr(call, "created_at", None),
             }
         )
     if not records:
@@ -250,12 +253,39 @@ def _install_declared_python_tools() -> None:
         return
     function_names = {function for _, function in declared}
     active = threading.local()
+    import dis
+    import inspect
+
+    suspendable = (
+        inspect.CO_COROUTINE | inspect.CO_GENERATOR | inspect.CO_ASYNC_GENERATOR
+    )
+    # frame id -> (frame, entry time). Holding the frame keeps its id from being
+    # reused while an entry for it is still open.
+    entries: dict[int, tuple[Any, float]] = {}
 
     def profile(frame: Any, event: str, value: Any) -> None:
         if getattr(active, "busy", False):
             return
-        if event != "return" or frame.f_code.co_name not in function_names:
+        code = frame.f_code
+        if code.co_name not in function_names:
+            return
+        if event == "call":
+            # A coroutine re-enters its frame on every resume after an await; the
+            # tool ran from its first entry, so keep that time.
+            entry = entries.get(id(frame))
+            if entry is None or entry[0] is not frame:
+                entries[id(frame)] = (frame, time.time())
             return
+        if event != "return":
+            return
+        # A coroutine also "returns" at every await; only a real return ends it.
+        finished = not code.co_flags & suspendable or dis.opname[
+            code.co_code[frame.f_lasti]
+        ].startswith("RETURN")
+        entry = (
+            entries.pop(id(frame), None) if finished else entries.get(id(frame))
+        )
+        started = entry[1] if entry is not None else time.time()
         filename = frame.f_code.co_filename.replace("\\", "/")
         key = next(
             (
@@ -285,6 +315,7 @@ def _install_declared_python_tools() -> None:
                             "arguments": arguments,
                             "output": value,
                             "is_error": False,
+                            "at": started,
                         },
                         default=str,
                         sort_keys=True,