ROB-3618: Add chat-analytics fields to HolmesChat schemas (#2058)

* Add chat-analytics fields to HolmesChat schemas

Declare request_type, request_source, source_ref, conversation_id,
conversation_source, and meta as optional fields on HolmesChatParams
and HolmesChatRequest so they're typed and discoverable end-to-end.
The fields already passed through via extra="allow" + Pydantic v1
dict() semantics; this change makes them first-class. HolmesIssueChat*
inherits the fields automatically.

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

* Add holmes_feedback action handler

Proxies user feedback (thumbs up/down) on a Holmes answer from the FE
to the Holmes server's POST /api/feedback, which UPDATEs the
HolmesUsageEvents row keyed by request_id.

Mirrors the holmes_oauth pattern (small synchronous POST with
raise_for_status pass-through), with pydantic-level validation of
request_id presence and sentiment via Literal["thumbs_up", "thumbs_down"]
so bad payloads fail at param construction.

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

* Register holmes_feedback in the manual-action allowlist

The Helm chart's allowlist gates which action names the relay can
forward to the runner. Without this entry the relay rejects
holmes_feedback before it reaches the action handler.

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

* Add is_internal flag to HolmesChat schemas

Optional boolean alongside the other chat-analytics fields (request_source,
conversation_id, etc.) so the FE can mark internal/test traffic and the
Holmes server can filter it out of usage dashboards.

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

* Address CodeRabbit review on holmes_feedback schemas

- HolmesFeedbackParams.request_id: enforce non-empty via Field(min_length=1)
  so empty strings are rejected at param construction.
- HolmesFeedbackRequest: mirror the params contract — sentiment is now
  Literal["thumbs_up", "thumbs_down"] and request_id has min_length=1, so
  the request body class can't be misused outside the action path.
- Add a test covering the empty-request_id rejection.

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

* Forward user_id as query param on holmes_feedback

Holmes' POST /api/feedback reads user_id from request.query_params, not
the JSON body (mirrors /api/chat's resolution path used by the relay).
Posting it in the body returns 400 "missing user_id". Pull user_id out
of the action params and forward it via the URL query string.

- Declare user_id: Optional[str] on HolmesFeedbackParams so it's
  discoverable; pop it before constructing the request body.
- Update tests to assert user_id is on params= and not in the body.

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

* Improve holmes_feedback observability and drop noisy Finding

- handle_holmes_error: log the upstream response body on HTTPError so
  4xx/5xx diagnostics aren't lost in the generic "Holmes internal
  configuration error" branch (e.g. {"detail":"missing user_id"}
  would otherwise be invisible). Benefits all Holmes proxy actions,
  not just feedback.
- holmes_feedback: drop the no-op Finding on success. The FE shows
  the "Thanks" toast optimistically and never reads the response;
  emitting a TYPE_NONE finding for every thumb click was UI noise.
  Replaced with an info log line carrying request_id + sentiment.
- Tests updated to assert add_finding is no longer called.

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

* Remove holmes_feedback action — FE now writes via Supabase RPC

The feedback flow has moved off the runner: the FE now records thumbs
up/down directly via a Supabase RPC, bypassing the relay → runner →
Holmes path. Drop the runner-side proxy:

- HolmesFeedbackParams and HolmesFeedbackRequest schemas
- holmes_feedback action handler
- 7 unit tests
- helm chart lightActions allowlist entry

Kept from earlier PR commits:
- Chat-analytics fields on HolmesChat schemas (request_source,
  conversation_id, source_ref, conversation_source, request_type,
  is_internal, meta) — still useful for /api/chat
- handle_holmes_error response-body logging — benefits all Holmes
  proxy actions (chat, oauth, conversation, issue_chat)

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

---------

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

Author: alonelish

src/robusta/core/model/base_params.py

@@ -207,6 +207,13 @@ class HolmesChatParams(HolmesParams):
     enable_tool_approval: bool = Field(default=False)
     tool_decisions: Optional[List[ToolApprovalDecision]] = None
     additional_system_prompt: Optional[str] = None
+    request_type: Optional[str] = None
+    request_source: Optional[str] = None
+    source_ref: Optional[str] = None
+    conversation_id: Optional[str] = None
+    conversation_source: Optional[str] = None
+    is_internal: Optional[bool] = None
+    meta: Optional[Dict[str, Any]] = None
 
     class Config:
         extra = "allow"

src/robusta/core/playbooks/internal/ai_integration.py

@@ -62,6 +62,14 @@ def handle_holmes_error(e: Exception) -> NoReturn:
             "Holmes endpoint is currently unreachable.",
         )
     elif isinstance(e, requests.HTTPError):
+        # Log the upstream response body so 4xx/5xx diagnostics aren't lost in
+        # the generic "Holmes internal configuration error" branch below
+        # (e.g. {"detail":"missing user_id"} would otherwise be invisible).
+        if e.response is not None:
+            logging.error(
+                f"Holmes responded {e.response.status_code} on {e.request.method if e.request else '?'} "
+                f"{e.request.url if e.request else '?'}: {e.response.text}"
+            )
         if e.response.status_code == 401 and "invalid_api_key" in e.response.text:
             raise ActionException(
                 ErrorCodes.HOLMES_REQUEST_ERROR, "Holmes invalid api key."

src/robusta/core/reporting/holmes.py

@@ -51,6 +51,13 @@ class HolmesChatRequest(BaseModel):
     enable_tool_approval: bool = Field(default=False)
     tool_decisions: Optional[List[ToolApprovalDecision]] = None
     additional_system_prompt: Optional[str] = None
+    request_type: Optional[str] = None
+    request_source: Optional[str] = None
+    source_ref: Optional[str] = None
+    conversation_id: Optional[str] = None
+    conversation_source: Optional[str] = None
+    is_internal: Optional[bool] = None
+    meta: Optional[Dict[str, Any]] = None
 
     class Config:
         extra = "allow"