feat(migration): phase 1 — Session + Tracer + RunConfig factory

Three foundation modules per PLAYBOOK §2.8 / §2.9 / §2.10 with all
relevant R2/R3 corrections (C7, C10, C11, C16, C21):

  strix/llm/strix_session.py            SessionABC wrapper around the
                                        legacy MemoryCompressor; on any
                                        compression failure, returns
                                        uncompressed history and
                                        permanently disables compression
                                        for the rest of the run (C10 +
                                        Round 3.4 W5/E2).

  strix/telemetry/strix_processor.py    SDK TracingProcessor that writes
                                        events.jsonl in our schema. All
                                        hooks SYNC per ABC (F3); writes
                                        protected by per-path
                                        threading.Lock (C7); OSError
                                        swallowed and logged (C16); PII
                                        scrubbed via the existing
                                        TelemetrySanitizer.

  strix/run_config_factory.py           make_run_config() with our
                                        defaults: parallel_tool_calls=
                                        False (C1 Phase-1 safe default),
                                        retry policy explicitly excludes
                                        401/403/400 (C11), reasoning
                                        effort + model_settings_override
                                        merge path (C21).
                                        make_agent_context() returns the
                                        canonical per-agent dict
                                        including is_whitebox/diff_scope/
                                        run_id (C21).

32 new smoke tests (197/197 total). mypy strict + ruff clean. Per-file
ignores added for tests/** S105/PT018 and for the two new src modules'
intentional broad-Exception catches (BLE001).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

pyproject.toml

@@ -248,11 +248,13 @@ ignore = [
 "strix/llm/llm.py" = ["PLC0415"]
 "strix/tools/notes/notes_actions.py" = ["PLC0415"]
 "tests/**/*.py" = [
-    "S106",   # Possible hardcoded password
+    "S105",   # Possible hardcoded password (string literal)
+    "S106",   # Possible hardcoded password (function call)
     "S108",   # Possible insecure usage of temporary file/directory
     "ARG001", # Unused function argument
     "ARG002", # Unused method argument (test helpers / fakes mirror SDK signatures)
     "PLR2004", # Magic value used in comparison
+    "PT018",  # Multi-part assertions are a pytest style preference
     "TC002",  # Type-only third-party import (tests import for runtime instantiation)
     "TC003",  # Type-only stdlib import
 ]
@@ -281,6 +283,11 @@ ignore = [
 "strix/tools/_decorator.py" = [
     "TC002",  # FunctionTool imported for annotation
 ]
+# StrixSession + StrixTracingProcessor catch broad Exception intentionally:
+# the whole point is that compressor / sanitizer / disk failures must not
+# tear down the agent run (C10, C16). Calls already log at exception level.
+"strix/llm/strix_session.py" = ["BLE001"]
+"strix/telemetry/strix_processor.py" = ["BLE001"]
 
 [tool.ruff.lint.isort]
 force-single-line = false

strix/llm/strix_session.py

@@ -0,0 +1,106 @@
+"""StrixSession — Session wrapper that runs the legacy MemoryCompressor.
+
+The SDK's `Session` (and ``SessionABC``) protocol owns conversation history
+storage. We delegate the actual storage to any underlying session
+implementation (in-memory, SQLite, Redis, …) and intercept ``get_items`` so
+the legacy ``MemoryCompressor`` runs before the model sees the history.
+
+Why wrap rather than reimplement:
+- ``MemoryCompressor`` already encodes the pentest-tuned summarization
+  prompt and the 90K-token budget that Strix has been tuning for months.
+  Reimplementing inside a Session would lose that institutional knowledge.
+- The SDK gives us a clean seam in ``get_items``: it's the last call before
+  ``call_model_input_filter`` runs, so compressing here means the filter
+  sees a compressed history too.
+
+References:
+    - PLAYBOOK.md §2.8
+    - AUDIT_R2.md §1.5 (C10 — compressor exception → uncompressed fallback)
+    - AUDIT_R3.md §3 row W5/E2 — once compression has failed, set a flag and
+      skip future attempts so we don't infinite-loop on a permanently broken
+      compressor while the agent loop slowly drowns in context.
+"""
+
+from __future__ import annotations
+
+import logging
+from typing import TYPE_CHECKING, Any, cast
+
+from agents.memory.session import SessionABC
+
+
+if TYPE_CHECKING:
+    from agents.items import TResponseInputItem
+
+    from strix.llm.memory_compressor import MemoryCompressor
+
+
+logger = logging.getLogger(__name__)
+
+
+class StrixSession(SessionABC):
+    """Wraps an underlying ``SessionABC`` with Strix's memory compressor.
+
+    The wrapped session owns persistence; ``StrixSession`` only intercepts
+    ``get_items`` to run compression. Writes (``add_items``, ``pop_item``,
+    ``clear_session``) pass through verbatim.
+
+    On compressor failure, the call returns the uncompressed history and
+    a per-instance flag is set so subsequent ``get_items`` calls skip the
+    compressor entirely. This avoids an infinite "compress → fail → grow"
+    loop when the compressor LLM is itself unavailable.
+    """
+
+    def __init__(
+        self,
+        underlying: SessionABC,
+        compressor: MemoryCompressor,
+    ) -> None:
+        self._underlying = underlying
+        self._compressor = compressor
+        self._compression_disabled = False
+        # ``SessionABC.session_id`` is a plain ``str`` field; pass through.
+        self.session_id: str = getattr(underlying, "session_id", "strix-session")
+        self.session_settings = getattr(underlying, "session_settings", None)
+
+    @property
+    def compression_disabled(self) -> bool:
+        """True after the compressor has failed at least once on this session."""
+        return self._compression_disabled
+
+    async def get_items(
+        self,
+        limit: int | None = None,
+    ) -> list[TResponseInputItem]:
+        """Read items from underlying storage and (optionally) compress.
+
+        On any compressor exception, log and return the uncompressed list.
+        Set ``_compression_disabled`` so the next call short-circuits.
+        """
+        items = await self._underlying.get_items(limit=limit)
+        if self._compression_disabled or not items:
+            return items
+        try:
+            # Compressor expects ``list[dict[str, Any]]``; SDK's
+            # ``TResponseInputItem`` is a TypedDict union — structurally
+            # compatible. Compressor mutates content but preserves shape.
+            compressed = self._compressor.compress_history(
+                cast("list[dict[str, Any]]", items),
+            )
+            return cast("list[TResponseInputItem]", compressed)
+        except Exception:
+            logger.exception(
+                "MemoryCompressor failed; returning uncompressed history. "
+                "Compression disabled for this session for the rest of the run.",
+            )
+            self._compression_disabled = True
+            return items
+
+    async def add_items(self, items: list[TResponseInputItem]) -> None:
+        await self._underlying.add_items(items)
+
+    async def pop_item(self) -> TResponseInputItem | None:
+        return await self._underlying.pop_item()
+
+    async def clear_session(self) -> None:
+        await self._underlying.clear_session()

strix/run_config_factory.py

@@ -0,0 +1,206 @@
+"""make_run_config — assemble a Strix-flavored ``RunConfig`` for ``Runner.run``.
+
+Factory pattern: every Strix scan goes through here so the defaults are
+applied uniformly. Per-call overrides are accepted via ``model_settings_override``
+for the rare case a single run wants different reasoning effort or
+``tool_choice`` (C21).
+
+References:
+    - PLAYBOOK.md §2.10
+    - AUDIT.md §2.1 (C1 — parallel_tool_calls=False until Phase 6 relaxes the
+      legacy tool server's per-agent task slot serialization)
+    - AUDIT_R2.md §1.6 (C11 — retry policy explicitly excludes 401/403/400;
+      auth and validation errors must fail fast, not waste retries)
+    - AUDIT_R3.md C21 — RunConfig override + context fields including
+      ``is_whitebox``, ``diff_scope``, ``run_id``
+"""
+
+from __future__ import annotations
+
+from typing import TYPE_CHECKING, Any, Literal
+
+from agents import RunConfig
+from agents.model_settings import ModelSettings
+from agents.retry import (
+    ModelRetryBackoffSettings,
+    ModelRetrySettings,
+    retry_policies,
+)
+from agents.sandbox import SandboxRunConfig
+
+from strix.llm.multi_provider_setup import build_multi_provider
+from strix.orchestration.filter import inject_messages_filter
+
+
+if TYPE_CHECKING:
+    from agents.sandbox.session.base_sandbox_session import BaseSandboxSession
+
+    from strix.orchestration.bus import AgentMessageBus
+
+
+# Phase 1-5 default. Phase 6 relaxes the legacy tool server's per-agent
+# task-slot serialization (``runtime/tool_server.py:94-97``) and flips this
+# to ``True`` after the multi-agent stress tests confirm safety.
+_PHASE1_PARALLEL_DEFAULT = False
+
+# Default retry policy. Explicitly does NOT include 401/403/400 — those are
+# auth and validation errors that retrying cannot fix; they should fail fast
+# so the user sees the real error within seconds. 429/5xx is the right set.
+_RETRYABLE_HTTP_STATUSES = (429, 500, 502, 503, 504)
+
+# Default retry budget. Mirrors the legacy ``llm.py`` retry loop: 5 attempts
+# with ``min(90, 2*2^n)`` backoff.
+_DEFAULT_MAX_RETRIES = 5
+_DEFAULT_BACKOFF = ModelRetryBackoffSettings(
+    initial_delay=2.0,
+    max_delay=90.0,
+    multiplier=2.0,
+    jitter=False,
+)
+
+
+def _default_retry_policy() -> Any:
+    """Build the default retry policy.
+
+    Built from ``retry_policies.any(...)``: any of the listed conditions
+    triggers a retry. ``provider_suggested`` honors server-sent
+    ``Retry-After`` hints; ``network_error`` covers connection / timeout;
+    ``http_status`` whitelists transient HTTP codes.
+    """
+    return retry_policies.any(
+        retry_policies.provider_suggested(),
+        retry_policies.network_error(),
+        retry_policies.http_status(_RETRYABLE_HTTP_STATUSES),
+    )
+
+
+#: Default ``max_turns`` callers should pass to ``Runner.run``.
+#: Mirrors the legacy ``AgentState.max_iterations = 300``
+#: (``HARNESS_WIKI.md §5.2``).
+STRIX_DEFAULT_MAX_TURNS = 300
+
+
+def make_run_config(
+    *,
+    sandbox_session: BaseSandboxSession | None,
+    model: str = "strix/claude-sonnet-4.6",
+    parallel_tool_calls: bool = _PHASE1_PARALLEL_DEFAULT,
+    tool_choice: Literal["auto", "required", "none"] | None = "required",
+    reasoning_effort: Literal["low", "medium", "high"] | None = None,
+    model_settings_override: ModelSettings | None = None,
+    sandbox_client: Any | None = None,
+) -> RunConfig:
+    """Build a ``RunConfig`` with Strix defaults.
+
+    Note: ``max_turns`` and ``isolate_parallel_failures`` are NOT
+    ``RunConfig`` fields — they are passed directly to ``Runner.run``.
+    Use ``STRIX_DEFAULT_MAX_TURNS`` for the budget; pass
+    ``isolate_parallel_failures=False`` to ``Runner.run`` if Phase 6 has
+    not yet flipped ``parallel_tool_calls=True``.
+
+    Args:
+        sandbox_session: Live sandbox session shared by every agent in this
+            scan (one container per scan; see ``strix.sandbox.session_manager``).
+            ``None`` is allowed for unit tests and dry runs.
+        model: Model alias to pass to ``MultiProvider``. Defaults to the
+            current production-favored Anthropic alias.
+        parallel_tool_calls: Phase 1 default is ``False`` to keep behavior
+            sequential per the legacy tool server's slot serialization (C1).
+        tool_choice: Forces tool use per turn unless explicitly relaxed.
+            Mirrors the legacy ``4f90a56`` prompt hardening at the model
+            level. Pass ``None`` to omit.
+        reasoning_effort: ``"low" | "medium" | "high"``; routes to
+            ``ModelSettings.reasoning``. ``None`` defers to provider default.
+        model_settings_override: Optional ``ModelSettings`` to merge over
+            the factory defaults (C21 — per-run override path).
+        sandbox_client: Optional pre-built sandbox client (e.g., the Strix
+            Docker subclass). Defaults to ``None``; the SDK will instantiate
+            its built-in if a session is supplied without a client.
+
+    Returns:
+        A ``RunConfig`` ready to pass to ``Runner.run``.
+    """
+    base_settings = ModelSettings(
+        parallel_tool_calls=parallel_tool_calls,
+        tool_choice=tool_choice,
+        retry=ModelRetrySettings(
+            max_retries=_DEFAULT_MAX_RETRIES,
+            backoff=_DEFAULT_BACKOFF,
+            policy=_default_retry_policy(),
+        ),
+    )
+    if reasoning_effort is not None:
+        from openai.types.shared import Reasoning
+
+        base_settings = base_settings.resolve(
+            ModelSettings(reasoning=Reasoning(effort=reasoning_effort)),
+        )
+    if model_settings_override is not None:
+        # ``ModelSettings.resolve`` merges another ModelSettings into self
+        # with override-wins semantics — exactly what we want.
+        base_settings = base_settings.resolve(model_settings_override)
+
+    sandbox_config = (
+        SandboxRunConfig(client=sandbox_client, session=sandbox_session)
+        if sandbox_session is not None
+        else None
+    )
+
+    return RunConfig(
+        model=model,
+        model_provider=build_multi_provider(),
+        model_settings=base_settings,
+        sandbox=sandbox_config,
+        call_model_input_filter=inject_messages_filter,
+        tracing_disabled=False,
+        trace_include_sensitive_data=False,
+    )
+
+
+def make_agent_context(
+    *,
+    bus: AgentMessageBus,
+    sandbox_session: BaseSandboxSession | None,
+    sandbox_token: str | None,
+    tool_server_host_port: int | None,
+    caido_host_port: int | None,
+    agent_id: str,
+    agent_name: str,
+    parent_id: str | None,
+    tracer: Any | None,
+    model: str = "strix/claude-sonnet-4.6",
+    model_settings: ModelSettings | None = None,
+    max_turns: int = 300,
+    is_whitebox: bool = False,
+    diff_scope: dict[str, Any] | None = None,
+    run_id: str | None = None,
+) -> dict[str, Any]:
+    """Build the per-agent ``context`` dict passed to ``Runner.run(context=...)``.
+
+    The dict is the canonical place where bus, sandbox handles, identity,
+    tracer reference, and per-agent toggles live. Tools, hooks, and the
+    ``inject_messages_filter`` all reach in via ``ctx.context.get(...)``.
+
+    C21 (AUDIT_R3): includes ``is_whitebox``, ``diff_scope`` and ``run_id``
+    fields that the legacy code relied on but the original PLAYBOOK §2.10
+    skeleton omitted.
+    """
+    return {
+        "bus": bus,
+        "sandbox_session": sandbox_session,
+        "sandbox_token": sandbox_token,
+        "tool_server_host_port": tool_server_host_port,
+        "caido_host_port": caido_host_port,
+        "agent_id": agent_id,
+        "agent_name": agent_name,
+        "parent_id": parent_id,
+        "tracer": tracer,
+        "model": model,
+        "model_settings": model_settings,
+        "max_turns": max_turns,
+        "turn_count": 0,
+        "agent_finish_called": False,
+        "is_whitebox": is_whitebox,
+        "diff_scope": diff_scope,
+        "run_id": run_id,
+    }

strix/telemetry/strix_processor.py

@@ -0,0 +1,183 @@
+"""StrixTracingProcessor — SDK trace processor that writes events.jsonl.
+
+Replaces the JSONL output that the legacy tracer wrote in ``telemetry/tracer.py``.
+Hooks into the SDK's tracing pipeline so we keep the existing
+``strix_runs/<run-name>/events.jsonl`` schema and the existing
+``TelemetrySanitizer`` PII redaction without standing up a parallel
+tracing system.
+
+References:
+    - PLAYBOOK.md §2.9
+    - AUDIT_R2.md §1.2 (C7 — JSONL writes must be lock-protected)
+    - AUDIT_R3.md C16 (writes must catch OSError; never tear down the run)
+    - AUDIT_R3.md F3 (every TracingProcessor hook is SYNC, not async)
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+import threading
+from pathlib import Path
+from typing import TYPE_CHECKING, Any
+
+from agents.tracing.processor_interface import TracingProcessor
+
+
+if TYPE_CHECKING:
+    from agents.tracing.spans import Span
+    from agents.tracing.traces import Trace
+
+    from strix.telemetry.utils import TelemetrySanitizer
+
+
+logger = logging.getLogger(__name__)
+
+
+# Module-level lock registry — one per JSONL file so two processors writing
+# different run-dirs don't serialize unnecessarily, but two processors
+# writing the *same* run-dir (e.g., legacy tracer + SDK processor) do.
+_FILE_LOCKS: dict[Path, threading.Lock] = {}
+_GUARD = threading.Lock()
+
+
+def _lock_for(path: Path) -> threading.Lock:
+    with _GUARD:
+        return _FILE_LOCKS.setdefault(path, threading.Lock())
+
+
+class StrixTracingProcessor(TracingProcessor):
+    """Append trace + span events as JSONL into ``run_dir/events.jsonl``.
+
+    Every hook is synchronous — required by ``TracingProcessor`` ABC.
+    Every write is protected by a per-path ``threading.Lock`` so concurrent
+    spans (e.g., from parallel agent tasks) cannot interleave bytes
+    mid-line and corrupt the JSONL (C7).
+
+    Every write is wrapped in ``try/except OSError`` so a full disk or a
+    permission error during the run does NOT propagate up the hook chain
+    and tear down the agent (C16).
+
+    PII scrubbing runs on every event before it hits the file. The
+    ``TelemetrySanitizer`` class is the same one the legacy tracer uses.
+    """
+
+    def __init__(
+        self,
+        run_dir: Path,
+        sanitizer: TelemetrySanitizer | None = None,
+    ) -> None:
+        run_dir = Path(run_dir)
+        run_dir.mkdir(parents=True, exist_ok=True)
+        self.run_dir: Path = run_dir
+        self.events_path: Path = run_dir / "events.jsonl"
+        if sanitizer is None:
+            from strix.telemetry.utils import TelemetrySanitizer
+
+            sanitizer = TelemetrySanitizer()
+        self.sanitizer: TelemetrySanitizer = sanitizer
+
+    # --- Internal helpers -------------------------------------------------
+
+    def _emit(self, event: dict[str, Any]) -> None:
+        """Sanitize ``event`` and append it as one JSONL line.
+
+        Failures are swallowed — we'd rather lose a trace event than fail
+        the run. Errors are logged at WARNING (C16).
+        """
+        try:
+            clean = self.sanitizer.sanitize(event)
+        except Exception:
+            logger.exception("Trace event sanitization failed; dropping event")
+            return
+        try:
+            with (
+                _lock_for(self.events_path),
+                self.events_path.open(
+                    "a",
+                    encoding="utf-8",
+                ) as f,
+            ):
+                f.write(json.dumps(clean, ensure_ascii=True) + "\n")
+        except OSError:
+            logger.exception("Failed to append trace event to %s", self.events_path)
+
+    @staticmethod
+    def _span_kind(span: Span[Any]) -> str:
+        """Map ``SomethingSpanData`` → ``"something"`` for the event_type."""
+        name = type(span.span_data).__name__
+        if name.endswith("SpanData"):
+            name = name[: -len("SpanData")]
+        return name.lower() or "span"
+
+    @staticmethod
+    def _trace_metadata(trace: Trace) -> dict[str, Any]:
+        meta: dict[str, Any] = {"name": getattr(trace, "name", None)}
+        # ``Trace.export()`` includes metadata + group_id when set.
+        try:
+            exported = trace.export()
+            if isinstance(exported, dict):
+                for key in ("metadata", "group_id", "workflow_name"):
+                    if key in exported and exported[key] is not None:
+                        meta[key] = exported[key]
+        except Exception:
+            logger.debug("trace.export failed", exc_info=True)
+        return meta
+
+    # --- TracingProcessor ABC --------------------------------------------
+
+    def on_trace_start(self, trace: Trace) -> None:
+        self._emit(
+            {
+                "event_type": "run.started",
+                "trace_id": trace.trace_id,
+                "metadata": self._trace_metadata(trace),
+            }
+        )
+
+    def on_trace_end(self, trace: Trace) -> None:
+        self._emit(
+            {
+                "event_type": "run.completed",
+                "trace_id": trace.trace_id,
+            }
+        )
+
+    def on_span_start(self, span: Span[Any]) -> None:
+        kind = self._span_kind(span)
+        self._emit(
+            {
+                "event_type": f"{kind}.started",
+                "span_id": span.span_id,
+                "trace_id": span.trace_id,
+                "data": self._safe_export(span),
+            }
+        )
+
+    def on_span_end(self, span: Span[Any]) -> None:
+        kind = self._span_kind(span)
+        self._emit(
+            {
+                "event_type": f"{kind}.completed",
+                "span_id": span.span_id,
+                "trace_id": span.trace_id,
+                "data": self._safe_export(span),
+            }
+        )
+
+    def force_flush(self) -> None:
+        """All writes are synchronous; nothing to flush."""
+
+    def shutdown(self) -> None:
+        """No-op; nothing to release."""
+
+    # --- helpers ----------------------------------------------------------
+
+    @staticmethod
+    def _safe_export(span: Span[Any]) -> dict[str, Any] | None:
+        try:
+            data = span.span_data.export()
+            return data if isinstance(data, dict) else None
+        except Exception:
+            logger.debug("span_data.export failed for %s", span.span_id, exc_info=True)
+            return None

tests/llm/test_strix_session.py

@@ -0,0 +1,153 @@
+"""Phase 1 smoke tests for StrixSession (memory compression wrapper)."""
+
+from __future__ import annotations
+
+from typing import Any
+
+import pytest
+from agents.memory.session import SessionABC
+
+from strix.llm.strix_session import StrixSession
+
+
+class _FakeUnderlying(SessionABC):
+    """In-memory SessionABC used to drive StrixSession in tests."""
+
+    def __init__(self, items: list[Any] | None = None) -> None:
+        self.items: list[Any] = list(items or [])
+        self.session_id = "fake-session"
+
+    async def get_items(self, limit: int | None = None) -> list[Any]:
+        if limit is None:
+            return list(self.items)
+        return list(self.items[-limit:])
+
+    async def add_items(self, items: list[Any]) -> None:
+        self.items.extend(items)
+
+    async def pop_item(self) -> Any | None:
+        return self.items.pop() if self.items else None
+
+    async def clear_session(self) -> None:
+        self.items.clear()
+
+
+class _CompressorOK:
+    """Compressor stand-in that compresses by keeping the last item."""
+
+    def __init__(self) -> None:
+        self.calls = 0
+
+    def compress_history(self, messages: list[dict[str, Any]]) -> list[dict[str, Any]]:
+        self.calls += 1
+        return messages[-1:] if len(messages) > 1 else messages
+
+
+class _CompressorBoom:
+    """Compressor stand-in that always raises."""
+
+    def __init__(self) -> None:
+        self.calls = 0
+
+    def compress_history(self, messages: list[dict[str, Any]]) -> list[dict[str, Any]]:
+        self.calls += 1
+        raise RuntimeError("compressor offline")
+
+
+@pytest.fixture
+def items() -> list[dict[str, Any]]:
+    return [
+        {"role": "user", "content": "first"},
+        {"role": "assistant", "content": "second"},
+        {"role": "user", "content": "third"},
+    ]
+
+
+@pytest.mark.asyncio
+async def test_get_items_compresses_when_compressor_ok(items: list[dict[str, Any]]) -> None:
+    underlying = _FakeUnderlying(items)
+    session = StrixSession(underlying, compressor=_CompressorOK())
+
+    out = await session.get_items()
+    assert len(out) == 1
+    assert out[0]["content"] == "third"
+
+
+@pytest.mark.asyncio
+async def test_get_items_returns_empty_without_calling_compressor() -> None:
+    """If underlying has no items, don't even invoke the compressor."""
+    underlying = _FakeUnderlying([])
+    compressor = _CompressorOK()
+    session = StrixSession(underlying, compressor=compressor)
+
+    out = await session.get_items()
+    assert out == []
+    assert compressor.calls == 0
+
+
+@pytest.mark.asyncio
+async def test_get_items_falls_back_to_uncompressed_on_exception(
+    items: list[dict[str, Any]],
+) -> None:
+    """C10 (AUDIT_R2): compressor failure must not tear down the run."""
+    underlying = _FakeUnderlying(items)
+    compressor = _CompressorBoom()
+    session = StrixSession(underlying, compressor=compressor)
+
+    out = await session.get_items()
+    # Uncompressed history returned.
+    assert out == items
+    # Flag set so subsequent calls skip the compressor.
+    assert session.compression_disabled is True
+
+
+@pytest.mark.asyncio
+async def test_compressor_disabled_after_first_failure(items: list[dict[str, Any]]) -> None:
+    """Round 3.4 §E2 / W5 — once the compressor fails, skip it forever."""
+    underlying = _FakeUnderlying(items)
+    compressor = _CompressorBoom()
+    session = StrixSession(underlying, compressor=compressor)
+
+    # First call: compressor invoked, raises, flag set.
+    await session.get_items()
+    assert compressor.calls == 1
+    assert session.compression_disabled is True
+
+    # Second + third call: compressor short-circuited.
+    await session.get_items()
+    await session.get_items()
+    assert compressor.calls == 1
+
+
+@pytest.mark.asyncio
+async def test_writes_pass_through(items: list[dict[str, Any]]) -> None:
+    underlying = _FakeUnderlying()
+    session = StrixSession(underlying, compressor=_CompressorOK())
+
+    await session.add_items(items)
+    assert underlying.items == items
+
+    popped = await session.pop_item()
+    assert popped == items[-1]
+
+    await session.clear_session()
+    assert underlying.items == []
+
+
+@pytest.mark.asyncio
+async def test_session_id_passes_through() -> None:
+    underlying = _FakeUnderlying()
+    session = StrixSession(underlying, compressor=_CompressorOK())
+    assert session.session_id == "fake-session"
+
+
+@pytest.mark.asyncio
+async def test_get_items_respects_limit(items: list[dict[str, Any]]) -> None:
+    """``limit`` is forwarded to the underlying session before compression."""
+    underlying = _FakeUnderlying(items)
+    session = StrixSession(underlying, compressor=_CompressorOK())
+
+    out = await session.get_items(limit=2)
+    # Underlying returned last 2 items; compressor kept the last 1.
+    assert len(out) == 1
+    assert out[0]["content"] == "third"

tests/telemetry/test_strix_processor.py

@@ -0,0 +1,201 @@
+"""Phase 1 smoke tests for StrixTracingProcessor."""
+
+from __future__ import annotations
+
+import json
+import threading
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Any
+
+import pytest
+
+from strix.telemetry.strix_processor import StrixTracingProcessor
+
+
+@dataclass
+class _FakeSpanData:
+    """Minimal stand-in for an SDK SpanData class."""
+
+    name: str = "FunctionSpanData"  # used by class .__name__
+    payload: dict[str, Any] = field(default_factory=dict)
+
+    def export(self) -> dict[str, Any]:
+        return dict(self.payload)
+
+
+# Concrete SpanData subclasses so the processor's ``_span_kind`` heuristic
+# (drop ``SpanData`` suffix, lowercase) produces stable event_types.
+class FunctionSpanData(_FakeSpanData):
+    pass
+
+
+class GenerationSpanData(_FakeSpanData):
+    pass
+
+
+class AgentSpanData(_FakeSpanData):
+    pass
+
+
+@dataclass
+class _FakeSpan:
+    span_id: str
+    trace_id: str
+    span_data: _FakeSpanData
+
+
+@dataclass
+class _FakeTrace:
+    trace_id: str
+    name: str = "test-workflow"
+    metadata: dict[str, Any] = field(default_factory=dict)
+
+    def export(self) -> dict[str, Any]:
+        return {"name": self.name, "metadata": self.metadata}
+
+
+def _read_events(path: Path) -> list[dict[str, Any]]:
+    return [json.loads(line) for line in path.read_text().splitlines() if line.strip()]
+
+
+@pytest.fixture
+def run_dir(tmp_path: Path) -> Path:
+    return tmp_path / "strix_runs" / "test-run"
+
+
+def test_constructor_creates_run_dir(run_dir: Path) -> None:
+    StrixTracingProcessor(run_dir=run_dir)
+    assert run_dir.exists()
+
+
+def test_on_trace_start_writes_run_started(run_dir: Path) -> None:
+    p = StrixTracingProcessor(run_dir=run_dir)
+    p.on_trace_start(_FakeTrace(trace_id="t-1", metadata={"scan_id": "abc"}))
+
+    events = _read_events(p.events_path)
+    assert events == [
+        {
+            "event_type": "run.started",
+            "trace_id": "t-1",
+            "metadata": {"name": "test-workflow", "metadata": {"scan_id": "abc"}},
+        }
+    ]
+
+
+def test_on_trace_end_writes_run_completed(run_dir: Path) -> None:
+    p = StrixTracingProcessor(run_dir=run_dir)
+    p.on_trace_end(_FakeTrace(trace_id="t-1"))
+
+    events = _read_events(p.events_path)
+    assert events == [{"event_type": "run.completed", "trace_id": "t-1"}]
+
+
+def test_span_start_and_end_emit_typed_events(run_dir: Path) -> None:
+    """``GenerationSpanData`` → ``generation.started`` / ``generation.completed``."""
+    p = StrixTracingProcessor(run_dir=run_dir)
+    span = _FakeSpan(
+        span_id="s-1",
+        trace_id="t-1",
+        span_data=GenerationSpanData(payload={"model": "gpt-foo"}),
+    )
+
+    p.on_span_start(span)
+    p.on_span_end(span)
+
+    events = _read_events(p.events_path)
+    assert [e["event_type"] for e in events] == [
+        "generation.started",
+        "generation.completed",
+    ]
+    assert events[0]["span_id"] == "s-1"
+    assert events[0]["data"] == {"model": "gpt-foo"}
+
+
+def test_concurrent_writes_yield_valid_jsonl(run_dir: Path) -> None:
+    """C7 (AUDIT_R2): per-path lock prevents JSONL corruption under contention."""
+    p = StrixTracingProcessor(run_dir=run_dir)
+
+    def writer(idx: int) -> None:
+        for i in range(50):
+            p._emit({"event_type": "synthetic", "writer": idx, "i": i})
+
+    threads = [threading.Thread(target=writer, args=(i,)) for i in range(10)]
+    for t in threads:
+        t.start()
+    for t in threads:
+        t.join()
+
+    # 10 threads x 50 events = 500 lines; all valid JSON.
+    lines = p.events_path.read_text().splitlines()
+    assert len(lines) == 500
+    for line in lines:
+        json.loads(line)  # raises on corrupt line
+
+
+def test_emit_swallows_oserror_and_logs(
+    run_dir: Path,
+    caplog: pytest.LogCaptureFixture,
+) -> None:
+    """C16 (AUDIT_R3): a write failure must NOT propagate."""
+    p = StrixTracingProcessor(run_dir=run_dir)
+    # Make events_path point to a directory so open(..., "a") raises.
+    p.events_path = run_dir
+
+    with caplog.at_level("ERROR", logger="strix.telemetry.strix_processor"):
+        p._emit({"event_type": "boom"})
+
+    assert any("Failed to append" in rec.message for rec in caplog.records)
+
+
+def test_span_export_failure_does_not_propagate(run_dir: Path) -> None:
+    """If span_data.export raises, we still emit an event with data=None."""
+    p = StrixTracingProcessor(run_dir=run_dir)
+
+    class _BoomSpanData:
+        def export(self) -> dict[str, Any]:
+            raise RuntimeError("nope")
+
+    # Reuse the lowercase rule: class name has no "SpanData" suffix → "boomspandata"
+    # would not be ideal; use a properly-named subclass.
+    class FunctionSpanDataBroken(_BoomSpanData):
+        pass
+
+    span = _FakeSpan(span_id="s-1", trace_id="t-1", span_data=FunctionSpanDataBroken())
+    p.on_span_end(span)
+
+    events = _read_events(p.events_path)
+    assert len(events) == 1
+    assert events[0]["data"] is None
+
+
+def test_pii_scrubbed_via_sanitizer(run_dir: Path) -> None:
+    """Sanitizer is invoked on every emit before write."""
+    seen: list[Any] = []
+
+    class _StubSanitizer:
+        def sanitize(self, data: Any, key_hint: str | None = None) -> Any:
+            seen.append(data)
+            # Replace any "secret" string with [REDACTED].
+            if isinstance(data, dict):
+                clean = {k: "[REDACTED]" if k == "api_key" else v for k, v in data.items()}
+                if "metadata" in clean and isinstance(clean["metadata"], dict):
+                    md = dict(clean["metadata"])
+                    md.pop("api_key", None)
+                    clean["metadata"] = md
+                return clean
+            return data
+
+    p = StrixTracingProcessor(run_dir=run_dir, sanitizer=_StubSanitizer())
+    p._emit({"event_type": "test", "api_key": "sk-very-secret"})
+
+    events = _read_events(p.events_path)
+    assert events[0]["api_key"] == "[REDACTED]"
+    assert seen and seen[0]["api_key"] == "sk-very-secret"
+
+
+def test_force_flush_and_shutdown_are_noops(run_dir: Path) -> None:
+    p = StrixTracingProcessor(run_dir=run_dir)
+    # Should not raise.
+    p.force_flush()
+    p.shutdown()

tests/test_run_config_factory.py

@@ -0,0 +1,181 @@
+"""Phase 1 smoke tests for make_run_config / make_agent_context."""
+
+from __future__ import annotations
+
+from agents import RunConfig
+from agents.model_settings import ModelSettings
+from agents.retry import ModelRetryBackoffSettings
+
+from strix.orchestration.bus import AgentMessageBus
+from strix.run_config_factory import (
+    _RETRYABLE_HTTP_STATUSES,
+    make_agent_context,
+    make_run_config,
+)
+
+
+def test_make_run_config_returns_run_config() -> None:
+    cfg = make_run_config(sandbox_session=None)
+    assert isinstance(cfg, RunConfig)
+
+
+def test_default_parallel_tool_calls_is_false() -> None:
+    """C1 (AUDIT.md): Phase 1 default is sequential to match legacy tool server."""
+    cfg = make_run_config(sandbox_session=None)
+    assert cfg.model_settings is not None
+    assert cfg.model_settings.parallel_tool_calls is False
+
+
+def test_default_tool_choice_is_required() -> None:
+    cfg = make_run_config(sandbox_session=None)
+    assert cfg.model_settings is not None
+    assert cfg.model_settings.tool_choice == "required"
+
+
+def test_call_model_input_filter_is_wired() -> None:
+    cfg = make_run_config(sandbox_session=None)
+    assert cfg.call_model_input_filter is not None
+    # Wired to inject_messages_filter (validated by name to keep import light).
+    assert cfg.call_model_input_filter.__name__ == "inject_messages_filter"
+
+
+def test_retry_settings_have_max_retries_5() -> None:
+    cfg = make_run_config(sandbox_session=None)
+    assert cfg.model_settings is not None
+    retry = cfg.model_settings.retry
+    assert retry is not None
+    assert retry.max_retries == 5
+
+
+def test_retry_backoff_uses_strix_defaults() -> None:
+    """Mirrors legacy llm.py: min(90, 2*2^n) with initial 2s, max 90s, x2."""
+    cfg = make_run_config(sandbox_session=None)
+    assert cfg.model_settings is not None
+    retry = cfg.model_settings.retry
+    assert retry is not None
+    backoff = retry.backoff
+    assert isinstance(backoff, ModelRetryBackoffSettings)
+    assert backoff.initial_delay == 2.0
+    assert backoff.max_delay == 90.0
+    assert backoff.multiplier == 2.0
+
+
+def test_retry_http_codes_exclude_401_403_400() -> None:
+    """C11 (AUDIT_R2): auth/validation errors must NOT be in the retry list."""
+    assert 401 not in _RETRYABLE_HTTP_STATUSES
+    assert 403 not in _RETRYABLE_HTTP_STATUSES
+    assert 400 not in _RETRYABLE_HTTP_STATUSES
+    # And 429 / 5xx must be present.
+    for code in (429, 500, 502, 503, 504):
+        assert code in _RETRYABLE_HTTP_STATUSES
+
+
+def test_trace_include_sensitive_data_is_false() -> None:
+    cfg = make_run_config(sandbox_session=None)
+    assert cfg.trace_include_sensitive_data is False
+
+
+def test_model_settings_override_merges() -> None:
+    """C21 (AUDIT_R3): per-call override path."""
+    override = ModelSettings(tool_choice="auto", parallel_tool_calls=True)
+    cfg = make_run_config(sandbox_session=None, model_settings_override=override)
+    assert cfg.model_settings is not None
+    assert cfg.model_settings.tool_choice == "auto"
+    assert cfg.model_settings.parallel_tool_calls is True
+    # Retry settings (not in override) preserved from base.
+    assert cfg.model_settings.retry is not None
+    assert cfg.model_settings.retry.max_retries == 5
+
+
+def test_reasoning_effort_propagates() -> None:
+    cfg = make_run_config(sandbox_session=None, reasoning_effort="high")
+    assert cfg.model_settings is not None
+    assert cfg.model_settings.reasoning is not None
+    assert cfg.model_settings.reasoning.effort == "high"
+
+
+def test_max_turns_default_is_300() -> None:
+    """Mirrors legacy AgentState.max_iterations=300 (HARNESS_WIKI §5.2)."""
+    # max_turns is RunConfig-level; we default 300 in make_agent_context for
+    # the per-agent context dict. RunConfig itself sets max_turns at run call
+    # time via Runner.run(max_turns=...). Verify our context.
+    bus = AgentMessageBus()
+    ctx = make_agent_context(
+        bus=bus,
+        sandbox_session=None,
+        sandbox_token=None,
+        tool_server_host_port=None,
+        caido_host_port=None,
+        agent_id="root",
+        agent_name="root",
+        parent_id=None,
+        tracer=None,
+    )
+    assert ctx["max_turns"] == 300
+
+
+def test_make_agent_context_full_shape() -> None:
+    """C21 — context dict carries every field tools/hooks reach for."""
+    bus = AgentMessageBus()
+    ctx = make_agent_context(
+        bus=bus,
+        sandbox_session=None,
+        sandbox_token="bearer-xyz",
+        tool_server_host_port=48081,
+        caido_host_port=48080,
+        agent_id="agent-1",
+        agent_name="root",
+        parent_id=None,
+        tracer="not-a-real-tracer",
+        is_whitebox=True,
+        diff_scope={"changed_files": ["src/app.py"]},
+        run_id="strix_runs/abc_def",
+    )
+
+    assert ctx["bus"] is bus
+    assert ctx["agent_id"] == "agent-1"
+    assert ctx["parent_id"] is None
+    assert ctx["agent_finish_called"] is False
+    assert ctx["turn_count"] == 0
+    assert ctx["is_whitebox"] is True
+    assert ctx["diff_scope"] == {"changed_files": ["src/app.py"]}
+    assert ctx["run_id"] == "strix_runs/abc_def"
+    assert ctx["sandbox_token"] == "bearer-xyz"
+    assert ctx["tool_server_host_port"] == 48081
+    assert ctx["caido_host_port"] == 48080
+
+
+def test_make_agent_context_is_whitebox_defaults_false() -> None:
+    bus = AgentMessageBus()
+    ctx = make_agent_context(
+        bus=bus,
+        sandbox_session=None,
+        sandbox_token=None,
+        tool_server_host_port=None,
+        caido_host_port=None,
+        agent_id="r",
+        agent_name="root",
+        parent_id=None,
+        tracer=None,
+    )
+    assert ctx["is_whitebox"] is False
+    assert ctx["diff_scope"] is None
+
+
+def test_sandbox_config_omitted_when_no_session() -> None:
+    cfg = make_run_config(sandbox_session=None)
+    assert cfg.sandbox is None
+
+
+def test_model_default_is_strix_claude() -> None:
+    """Production default per AUDIT/PLAYBOOK convention."""
+    cfg = make_run_config(sandbox_session=None)
+    assert cfg.model == "strix/claude-sonnet-4.6"
+
+
+def test_multi_provider_is_built() -> None:
+    """Verify the factory wires our custom MultiProvider, not the SDK default."""
+    cfg = make_run_config(sandbox_session=None)
+    # MultiProvider is opaque, but our build_multi_provider returns
+    # an instance with our prefix routes installed.
+    assert cfg.model_provider is not None