messages.send: auto_verify body-echo (Phase 2 body-verify defense)

CHANGELOG.md

@@ -6,6 +6,7 @@ All notable changes to cueapi-sdk will be documented here.
 
 ### Added
 
+- **`client.messages.send(auto_verify=True)` body-verify defense (Mike directive 2026-05-11).** New `auto_verify` kwarg, default `True`. When set, the SDK sends `X-CueAPI-Verify-Echo: true` request header. Substrate-side (Phase 1; cueapi-core's lane) echoes the body it received back in the response under `body_received`. SDK diffs sent vs received and raises `BodyVerifyMismatchError` on drift (with `sent_body`, `received_body`, `first_divergence_byte`, `message_id` attributes for programmatic recovery / diagnostic output). Catches the caller-side shell-expansion bug class where `body=f"... {dynamic_var} ..."` or worse `body=os.popen(...)` silently mutated body content upstream. Opt-out via `auto_verify=False` for perf-sensitive flows. Backward-compat: SDK no-ops when substrate omits the echo field (pre-Layer-1 behavior unchanged). New helper: `cueapi.exceptions.first_divergence_byte(a, b)` returns the byte index of the first differing position (pure function; re-usable cross-SDK).
 - `client.cues.bulk_delete(ids)` — delete up to 100 cues in a single call. Returns `{"deleted": [...], "skipped": [...]}`. Per-ID atomic, not batch atomic. Sends `X-Confirm-Destructive: true` header automatically. Wraps `POST /v1/cues/bulk-delete` (cueapi #650). Parity port of cueapi-cli #46. Raises `ValueError` client-side on empty list or > 100 IDs.
 
 ## [0.2.0] - 2026-05-01

cueapi/exceptions.py

@@ -56,3 +56,72 @@ class InvalidScheduleError(CueAPIError):
 
 class CueAPIServerError(CueAPIError):
     """Raised on 5xx — server error."""
+
+
+class BodyVerifyMismatchError(CueAPIError):
+    """Raised when ``messages.send(auto_verify=True)`` detects that the
+    body the server received differs from the body the caller sent.
+
+    Phase 2 of body-verify defense-in-depth (Mike directive 2026-05-11).
+    Caught when ``X-CueAPI-Verify-Echo: true`` request header is sent +
+    the server echoes back the received body in the response. The most
+    likely cause: caller-side shell expansion of ``$(...)`` / backticks /
+    ``${VAR}`` in the body arg BEFORE Python received it (e.g., a bash
+    invocation that assembled the body via double-quoted-string-with-
+    metacharacters before passing as argv).
+
+    Attributes:
+        sent_body: The body the SDK sent in the POST request.
+        received_body: The body the server reports having received.
+        first_divergence_byte: Zero-based byte offset of the first
+            differing position; useful for pinpointing single-char drift.
+            ``-1`` when one body is a proper prefix of the other (length
+            difference rather than content drift).
+        message_id: The server-assigned message ID (server stored the
+            mutated content; caller can inspect via ``messages.get(...)``
+            if needed for diagnostic / recovery purposes).
+    """
+
+    def __init__(
+        self,
+        message: str,
+        *,
+        sent_body: str,
+        received_body: str,
+        first_divergence_byte: int,
+        message_id: str,
+        **kwargs: Any,
+    ):
+        self.sent_body = sent_body
+        self.received_body = received_body
+        self.first_divergence_byte = first_divergence_byte
+        self.message_id = message_id
+        super().__init__(message, **kwargs)
+
+    def __repr__(self) -> str:
+        return (
+            f"BodyVerifyMismatchError(message_id={self.message_id!r}, "
+            f"first_divergence_byte={self.first_divergence_byte}, "
+            f"sent_len={len(self.sent_body)}, "
+            f"received_len={len(self.received_body)})"
+        )
+
+
+def first_divergence_byte(a: str, b: str) -> int:
+    """Return the byte index of the first differing position between
+    ``a`` and ``b``. Returns ``-1`` when ``a == b`` OR when one is a
+    proper prefix of the other (length differs but the shorter is a
+    clean prefix); the caller should distinguish length-mismatch from
+    content-divergence by comparing ``len(a) == len(b)``.
+
+    Pure function; no SDK dependency. Used by
+    ``BodyVerifyMismatchError`` for diagnostic output + can be re-used
+    cross-SDK (cueapi-cli, cueapi-action).
+    """
+    common_len = min(len(a), len(b))
+    for i in range(common_len):
+        if a[i] != b[i]:
+            return i
+    # Equal up to common_len. Either fully equal or one is a prefix of
+    # the other. -1 signals the caller to check length-mismatch.
+    return -1

cueapi/resources/messages.py

@@ -5,9 +5,19 @@
 from datetime import datetime
 from typing import TYPE_CHECKING, Any, Dict, Optional, Union
 
+from cueapi.exceptions import BodyVerifyMismatchError, first_divergence_byte
+
 if TYPE_CHECKING:
     from cueapi.client import CueAPI
 
+# Response field where Layer 1 substrate echoes back the body it received
+# (Phase 2 of body-verify defense in depth; Mike directive 2026-05-11).
+# Substrate-side ships this field in the 201 response when the request
+# included ``X-CueAPI-Verify-Echo: true`` header. Field name locked
+# during joint design between cueapi-primary (substrate) + CMA (SDK);
+# update if the design Dock spec finalizes a different name.
+_VERIFY_ECHO_FIELD = "body_received"
+
 
 class MessagesResource:
     """Messages API resource.
@@ -35,6 +45,7 @@ def send(
         metadata: Optional[Dict[str, Any]] = None,
         idempotency_key: Optional[str] = None,
         send_at: Optional[Union[str, datetime]] = None,
+        auto_verify: bool = True,
     ) -> dict:
         """Send a message.
 
@@ -110,8 +121,43 @@ def send(
         headers: Dict[str, str] = {"X-Cueapi-From-Agent": from_agent}
         if idempotency_key is not None:
             headers["Idempotency-Key"] = idempotency_key
-
-        return self._client._post("/v1/messages", json=payload, headers=headers)
+        if auto_verify:
+            # Phase 2 of body-verify defense in depth. When this header is
+            # set, the substrate echoes the body it received in the 201
+            # response under the ``body_received`` field. SDK then diffs
+            # sent vs received + raises BodyVerifyMismatchError on drift.
+            # Substrate ignores the header when Layer 1 isn't deployed
+            # yet; SDK no-ops on missing response field. Backward-compat.
+            headers["X-CueAPI-Verify-Echo"] = "true"
+
+        response = self._client._post("/v1/messages", json=payload, headers=headers)
+
+        # Verify echo if requested. The substrate-side echo lands in
+        # response[_VERIFY_ECHO_FIELD] when Layer 1 is deployed; absent
+        # otherwise (no-op in that case).
+        if auto_verify and isinstance(response, dict):
+            received = response.get(_VERIFY_ECHO_FIELD)
+            if received is not None and received != body:
+                msg_id = response.get("id", "<unknown>")
+                divergence = first_divergence_byte(body, received)
+                if divergence == -1 and len(body) != len(received):
+                    # One body is a proper prefix of the other; length
+                    # mismatch is the signal. Report at boundary of the
+                    # shorter body.
+                    divergence = min(len(body), len(received))
+                raise BodyVerifyMismatchError(
+                    f"Body received by substrate ({len(received)} bytes) differs from "
+                    f"body sent ({len(body)} bytes); first divergence at byte "
+                    f"{divergence}. Likely cause: caller-side shell expansion of "
+                    f"$(...) / backticks / ${{VAR}} in the body arg before Python "
+                    f"received it. Mitigations: pass body via file (Path.read_text) "
+                    f"or use --message-file in cueapi-cli.",
+                    sent_body=body,
+                    received_body=received,
+                    first_divergence_byte=divergence,
+                    message_id=msg_id,
+                )
+        return response
 
     def get(self, msg_id: str) -> dict:
         """Get a single message by ID."""

tests/test_messages_resource.py

@@ -20,10 +20,14 @@ def test_minimal_body_and_from_header(self):
 
         r.send(from_agent="sender@x", to="recipient@y", body="hi")
 
+        # Phase 2 of body-verify defense in depth (Mike directive 2026-05-11):
+        # auto_verify=True is the new default → X-CueAPI-Verify-Echo header
+        # always added. Substrate echoes back received body when Layer 1
+        # deployed; SDK diffs + raises BodyVerifyMismatchError on drift.
         mock_client._post.assert_called_once_with(
             "/v1/messages",
             json={"to": "recipient@y", "body": "hi"},
-            headers={"X-Cueapi-From-Agent": "sender@x"},
+            headers={"X-Cueapi-From-Agent": "sender@x", "X-CueAPI-Verify-Echo": "true"},
         )
 
     def test_with_all_optionals(self):
@@ -59,6 +63,7 @@ def test_with_all_optionals(self):
         assert call.kwargs["headers"] == {
             "X-Cueapi-From-Agent": "sender@x",
             "Idempotency-Key": "idemp-key-1",
+            "X-CueAPI-Verify-Echo": "true",
         }
 
     def test_omits_expects_reply_when_default(self):
@@ -88,17 +93,18 @@ def test_idempotency_key_too_long_raises_client_side(self):
         mock_client._post.assert_not_called()
 
     def test_omits_idempotency_key_header_when_unset(self):
-        # Headers should ONLY contain X-Cueapi-From-Agent when no
-        # idempotency_key is passed. Pin so a refactor can't silently
-        # start adding `Idempotency-Key: None` (httpx would coerce).
+        # Headers should ONLY contain X-Cueapi-From-Agent (+ default
+        # auto-verify header) when no idempotency_key is passed. Pin so
+        # a refactor can't silently start adding `Idempotency-Key: None`
+        # (httpx would coerce).
         mock_client = MagicMock()
         mock_client._post.return_value = {"id": "msg_x"}
         r = MessagesResource(mock_client)
 
         r.send(from_agent="x", to="y", body="hi")
 
         headers = mock_client._post.call_args.kwargs["headers"]
-        assert headers == {"X-Cueapi-From-Agent": "x"}
+        assert headers == {"X-Cueapi-From-Agent": "x", "X-CueAPI-Verify-Echo": "true"}
         assert "Idempotency-Key" not in headers
 
 
@@ -167,7 +173,7 @@ def test_send_with_send_at_iso_string(self):
                 "body": "hi",
                 "send_at": "2030-01-01T12:00:00Z",
             },
-            headers={"X-Cueapi-From-Agent": "sender@x"},
+            headers={"X-Cueapi-From-Agent": "sender@x", "X-CueAPI-Verify-Echo": "true"},
         )
 
     def test_send_with_send_at_datetime_auto_isoformats(self):
@@ -198,3 +204,126 @@ def test_send_without_send_at_omits_field(self):
 
         call = mock_client._post.call_args
         assert "send_at" not in call.kwargs["json"]
+
+
+class TestAutoVerify:
+    """Phase 2 of body-verify defense in depth (Mike directive 2026-05-11).
+
+    auto_verify=True (default) adds X-CueAPI-Verify-Echo: true header.
+    Substrate echoes back received body in 201 response under
+    body_received field; SDK diffs sent vs received + raises
+    BodyVerifyMismatchError on drift.
+    """
+
+    def test_default_adds_verify_echo_header(self):
+        mock_client = MagicMock()
+        mock_client._post.return_value = {"id": "msg_x", "delivery_state": "queued"}
+        r = MessagesResource(mock_client)
+
+        r.send(from_agent="x", to="y", body="hi")
+
+        headers = mock_client._post.call_args.kwargs["headers"]
+        assert headers.get("X-CueAPI-Verify-Echo") == "true"
+
+    def test_opt_out_omits_header(self):
+        mock_client = MagicMock()
+        mock_client._post.return_value = {"id": "msg_x", "delivery_state": "queued"}
+        r = MessagesResource(mock_client)
+
+        r.send(from_agent="x", to="y", body="hi", auto_verify=False)
+
+        headers = mock_client._post.call_args.kwargs["headers"]
+        assert "X-CueAPI-Verify-Echo" not in headers
+
+    def test_byte_identical_response_returns_normally(self):
+        """When server echoes back the same body, send() returns response."""
+        mock_client = MagicMock()
+        mock_client._post.return_value = {
+            "id": "msg_x",
+            "delivery_state": "queued",
+            "body_received": "hi",
+        }
+        r = MessagesResource(mock_client)
+
+        result = r.send(from_agent="x", to="y", body="hi")
+
+        assert result["id"] == "msg_x"
+
+    def test_raises_on_body_mismatch(self):
+        """When server echo differs from sent body, raises BodyVerifyMismatchError."""
+        from cueapi.exceptions import BodyVerifyMismatchError
+
+        mock_client = MagicMock()
+        mock_client._post.return_value = {
+            "id": "msg_mutated",
+            "delivery_state": "queued",
+            "body_received": "body with INJECT (caller's shell command-substituted)",
+        }
+        r = MessagesResource(mock_client)
+
+        with pytest.raises(BodyVerifyMismatchError) as exc:
+            r.send(
+                from_agent="x", to="y",
+                body="body with $(echo INJECT) (intended literal)",
+            )
+
+        assert exc.value.message_id == "msg_mutated"
+        assert "$(echo INJECT)" in exc.value.sent_body
+        assert "INJECT (caller" in exc.value.received_body
+        assert exc.value.first_divergence_byte >= 0
+
+    def test_no_op_when_substrate_omits_echo_field(self):
+        """Backward-compat: pre-Layer-1 substrate doesn't include
+        body_received field → SDK doesn't raise; returns normally."""
+        mock_client = MagicMock()
+        mock_client._post.return_value = {"id": "msg_x", "delivery_state": "queued"}
+        r = MessagesResource(mock_client)
+
+        result = r.send(from_agent="x", to="y", body="hi")
+
+        assert result["id"] == "msg_x"
+
+    def test_opt_out_skips_verify_even_if_substrate_echoes(self):
+        """auto_verify=False: even if substrate sends body_received, don't check."""
+        mock_client = MagicMock()
+        # Mismatched echo but opt-out → no exception
+        mock_client._post.return_value = {
+            "id": "msg_x", "body_received": "DIFFERENT BODY",
+        }
+        r = MessagesResource(mock_client)
+
+        result = r.send(from_agent="x", to="y", body="hi", auto_verify=False)
+
+        assert result["id"] == "msg_x"
+
+
+class TestFirstDivergenceByte:
+    """Pure helper for diagnostic byte-position-of-first-difference."""
+
+    def test_equal_strings_return_minus_one(self):
+        from cueapi.exceptions import first_divergence_byte
+        assert first_divergence_byte("abc", "abc") == -1
+
+    def test_prefix_returns_minus_one(self):
+        from cueapi.exceptions import first_divergence_byte
+        assert first_divergence_byte("abc", "abcd") == -1  # one is prefix of other
+
+    def test_first_char_diff(self):
+        from cueapi.exceptions import first_divergence_byte
+        assert first_divergence_byte("Xbc", "abc") == 0
+
+    def test_middle_diff(self):
+        from cueapi.exceptions import first_divergence_byte
+        assert first_divergence_byte("abXd", "abcd") == 2
+
+    def test_metachar_substitution_scenario(self):
+        """Realistic case: caller's shell substituted $(echo X) → 'X'.
+
+        Sent:     'pre $(echo INJ) post'  (caller intended literal)
+        Received: 'pre INJ post'           (shell already substituted)
+        First divergence at byte 4 (start of '$' in sent vs 'I' in received).
+        """
+        from cueapi.exceptions import first_divergence_byte
+        sent = "pre $(echo INJ) post"
+        received = "pre INJ post"
+        assert first_divergence_byte(sent, received) == 4