applied-cli 0.6.11__tar.gz → 0.6.12__tar.gz

{applied_cli-0.6.11 → applied_cli-0.6.12}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: applied-cli
-Version: 0.6.11
+Version: 0.6.12
 Summary: CLI and shared client library for Applied Labs AI support agents
 Author: Applied Labs
 License-Expression: MIT

{applied_cli-0.6.11 → applied_cli-0.6.12}/applied_cli/__init__.py

@@ -6,7 +6,7 @@ from applied_cli.client import AppliedClient
 from applied_cli.formatters import to_csv, to_json
 from applied_cli.toolkit import ToolResult, ToolSpec
 
-__version__ = "0.6.11"
+__version__ = "0.6.12"
 
 __all__ = [
     "AppliedClient",

{applied_cli-0.6.11 → applied_cli-0.6.12}/applied_cli/client.py

@@ -210,6 +210,21 @@ def _filter_has_explicit_lookup(field: str) -> bool:
     return field.endswith(_EXPLICIT_FILTER_SUFFIXES)
 
 
+# Filters the server declares ONLY in scalar form (method filters, e.g. the
+# evals review-state filters on ConversationFilterset). Rewriting these to
+# `field__in` silently drops them — django-filter ignores unknown params —
+# so they must pass through verbatim.
+_SCALAR_ONLY_FILTER_FIELDS = frozenset(
+    {
+        "verdict",
+        "reviewed",
+        "needs_attention",
+        "exclude_audited",
+        "only_unaudited",
+    }
+)
+
+
 def _serialize_filter_value(value: Any) -> str:
     if isinstance(value, (dict, list)):
         return json.dumps(value, separators=(",", ":"), ensure_ascii=True)
@@ -260,7 +275,11 @@ def _apply_query_filter(params: dict[str, Any], field: str, cond: Any) -> None:
         )
         return
 
-    key = field if explicit_lookup else f"{field}__in"
+    key = (
+        field
+        if explicit_lookup or field in _SCALAR_ONLY_FILTER_FIELDS
+        else f"{field}__in"
+    )
     params[key] = _serialize_filter_value(cond)
 
 

{applied_cli-0.6.11 → applied_cli-0.6.12}/applied_cli/tools.py

@@ -1186,6 +1186,9 @@ async def conversation_query(
             {"sublabel_id": "<intent-uuid>"} — filter by intent
             {"flags__in": "<flag-uuid>"} — has specific flag
             {"resolution": {"nin": ["escalated"]}} — exclude escalated
+            {"verdict": "ai_flagged"} — AI Reviewer flagged (also: pass, fail,
+                disagree, agree, none)
+            {"reviewed": False} — no human review verdict yet
         search: Full-text search across conversation content
         fields: Fields to return, e.g. ["id", "title", "contact.email", "score", "comment"]
             When provided, fields override view-specific projections.

{applied_cli-0.6.11 → applied_cli-0.6.12}/applied_cli/v2/evals.py

@@ -7,9 +7,10 @@ step. These tools expose that loop to an agent — grade conversations, record
 verdicts, teach the Reviewer, and promote a bad conversation into the simulation
 suite as a proposed test (with an auto-generated test customer).
 
-All endpoints are shop-scoped and merchant-authenticated. Discover conversations
-to review with `conversations_search` / `conversations_find`, then feed their ids
-here.
+All endpoints are shop-scoped and merchant-authenticated. Enumerate the review
+queue directly with `review_status(segment="inbox")` (or "disagree"), or discover
+conversations with `conversations_search` / `conversations_find` and feed their
+ids here.
 """
 
 from __future__ import annotations
@@ -43,8 +44,24 @@ class ReviewRunInput(StrictInput):
 
 
 class ReviewStatusInput(StrictInput):
-    conversation_ids: list[str] = Field(
-        description="Conversation ids to fetch verdict pairs + narration for.",
+    conversation_ids: list[str] | None = Field(
+        default=None,
+        description=(
+            "Conversation ids to fetch verdict pairs + narration for. Omit and "
+            "pass segment to enumerate a queue segment instead."
+        ),
+    )
+    segment: Literal["inbox", "disagree"] | None = Field(
+        default=None,
+        description=(
+            "Enumerate a queue segment server-side, newest judgment first: "
+            "'inbox' (AI-flagged, no human verdict yet) or 'disagree' (human "
+            "and AI verdicts differ). Ignored when conversation_ids is set."
+        ),
+    )
+    limit: int | None = Field(
+        default=None,
+        description="Max segment rows (server default 50, max 200).",
     )
 
 
@@ -179,20 +196,42 @@ async def review_run_handler(
 async def review_status_handler(
     client: AppliedClient, params: ReviewStatusInput
 ) -> ToolResult[Any]:
-    ids = params.conversation_ids[:200]
-    try:
-        data = await client.request(
-            "GET", "/v1/evals/status/", params={"ids": ",".join(ids)}
+    ids = (params.conversation_ids or [])[:200]
+    if not ids and not params.segment:
+        return ToolResult(
+            ok=False,
+            error=ToolError(
+                code="input.invalid",
+                message=(
+                    "Provide conversation_ids, or segment='inbox'/'disagree' "
+                    "to enumerate a queue segment."
+                ),
+                retryable=False,
+            ),
         )
+    query: dict[str, Any] = {}
+    if ids:
+        query["ids"] = ",".join(ids)
+    else:
+        query["segment"] = params.segment
+        if params.limit is not None:
+            query["limit"] = params.limit
+    try:
+        data = await client.request("GET", "/v1/evals/status/", params=query)
     except AppliedAPIError as exc:
         return _api_error_result(exc)
     rows = data.get("rows", []) if isinstance(data, dict) else []
+    summary = (
+        f"{len(rows)} row(s) in the '{params.segment}' segment."
+        if not ids
+        else f"Verdict pairs for {len(rows)} of {len(ids)} conversation(s)."
+    )
     return ToolResult(
         data=data,
-        summary=f"Verdict pairs for {len(rows)} of {len(ids)} conversation(s).",
+        summary=summary,
         next_actions=[
             "Use review_get for the full review pair + notes on one conversation.",
-            "Use review_run with the unreviewed ids to grade them.",
+            "Use review_set_verdict to confirm or overrule a verdict.",
         ],
     )
 
@@ -424,7 +463,9 @@ def evals_specs() -> list[ToolSpec]:
             namespace="review",
             description=(
                 "Fetch the verdict pair (AI + human) and the Reviewer's one-line "
-                "reason for a set of conversations, plus shop-wide review counts."
+                "reason for a set of conversations, plus shop-wide review counts. "
+                "Pass segment='inbox' (AI-flagged, unreviewed) or 'disagree' "
+                "instead of ids to enumerate the review queue directly."
             ),
             input_model=ReviewStatusInput,
             output_model=None,

{applied_cli-0.6.11 → applied_cli-0.6.12}/applied_cli.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: applied-cli
-Version: 0.6.11
+Version: 0.6.12
 Summary: CLI and shared client library for Applied Labs AI support agents
 Author: Applied Labs
 License-Expression: MIT

{applied_cli-0.6.11 → applied_cli-0.6.12}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "applied-cli"
-version = "0.6.11"
+version = "0.6.12"
 description = "CLI and shared client library for Applied Labs AI support agents"
 readme = "README.md"
 requires-python = ">=3.11"

{applied_cli-0.6.11 → applied_cli-0.6.12}/tests/test_client.py

@@ -136,6 +136,35 @@ async def test_query_conversations_serializes_metric_event_filter_payload(monkey
     )
 
 
+@pytest.mark.asyncio
+async def test_query_conversations_passes_scalar_only_filters_verbatim(monkeypatch):
+    """Method filters like the evals review-state filters exist server-side only
+    in scalar form — rewriting them to `field__in` silently drops them."""
+    client = AppliedClient(token="test-token")
+    seen = {}
+
+    async def fake_request(method, path, params=None, body=None, shop_id=None):
+        seen["params"] = params
+        return []
+
+    monkeypatch.setattr(client, "_request", fake_request)
+
+    await client.query_conversations(
+        filters={
+            "verdict": "ai_flagged",
+            "reviewed": False,
+            "resolution": "escalated",
+        }
+    )
+
+    assert seen["params"]["verdict"] == "ai_flagged"
+    assert seen["params"]["reviewed"] == "false"
+    assert "verdict__in" not in seen["params"]
+    assert "reviewed__in" not in seen["params"]
+    # Ordinary scalars keep the historical `__in` rewrite.
+    assert seen["params"]["resolution__in"] == "escalated"
+
+
 @pytest.mark.asyncio
 async def test_analytics_report_posts_payload(monkeypatch):
     client = AppliedClient(token="test-token")

{applied_cli-0.6.11 → applied_cli-0.6.12}/tests/test_v2_evals.py

@@ -76,6 +76,30 @@ async def test_review_status_joins_ids_into_query():
     assert client.calls[0]["params"] == {"ids": "a,b,c"}
 
 
+@pytest.mark.asyncio
+async def test_review_status_enumerates_segment_without_ids():
+    client = FakeEvalsClient({"rows": [{"conversation_id": "a"}], "counts": {}})
+    result = await _run("review_status", client, segment="inbox", limit=25)
+    assert result.ok
+    assert client.calls[0]["params"] == {"segment": "inbox", "limit": 25}
+
+
+@pytest.mark.asyncio
+async def test_review_status_ids_win_over_segment():
+    client = FakeEvalsClient({"rows": [], "counts": {}})
+    await _run("review_status", client, conversation_ids=["a"], segment="inbox")
+    assert client.calls[0]["params"] == {"ids": "a"}
+
+
+@pytest.mark.asyncio
+async def test_review_status_requires_ids_or_segment():
+    client = FakeEvalsClient()
+    result = await _run("review_status", client)
+    assert not result.ok
+    assert result.error.code == "input.invalid"
+    assert client.calls == []
+
+
 @pytest.mark.asyncio
 async def test_review_set_verdict_omits_unset_optionals():
     client = FakeEvalsClient({"verdict": "fail"})