ROB-3820: Forward chat-analytics fields through holmes_issue_chat (#2075)

* Forward chat-analytics fields through holmes_issue_chat

holmes_chat passes the full HolmesChatParams payload to /api/chat via
params.dict() (b14f68d), so source_ref and the other analytics fields
land on Holmes. holmes_issue_chat was missed: it built
HolmesIssueChatRequest with five hand-picked fields, silently dropping
source_ref, request_type, request_source, conversation_id,
conversation_source, is_internal, meta, plus enable_tool_approval,
tool_decisions, and additional_system_prompt.

Slack/Teams threaded chats on an existing alert hit /api/issue_chat,
which is why source_ref arrived from the relay populated and was
stored as NULL in HolmesUsageEvents.

Mirror the holmes_chat pass-through pattern: take params.dict() minus
runner-only/structurally-handled fields and splat the rest, then add
investigation_result and issue_type from params.context. Same
client/server-upgrade-without-middleware-changes property the
HolmesChat schemas docstring describes.

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

* Log chat-analytics fields before forwarding to Holmes

Diagnostic log on both /api/chat and /api/issue_chat paths so we can
verify what the runner actually forwards (source_ref, request_source,
request_type, conversation_id, conversation_source, is_internal). If
these show up in the runner log but land NULL in Holmes' usage events,
the gap is downstream of the runner.

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

* Log raw action_params keys for holmes_* actions at receiver

Targeted info-level log at the websocket boundary to distinguish
"relay didn't send source_ref" from "something downstream stripped
it." Prints the full key list so we can see whether source_ref is
present at all when the relay hands off to the runner, plus the
analytics field values for direct comparison with the
holmes_chat forwarding log line.

Scoped to action names starting with holmes_ so the rest of the
action traffic stays at debug verbosity.

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

* Remove diagnostic logging — issue traced to relay

The receiver-side and forwarding-side logs confirmed the relay was
omitting source_ref and the other chat-analytics fields from the
holmes_chat action_params on the Slack path. Fix lives in the relay
repo, not here. Reverting the diagnostic log lines now that the
investigation is done; keeping the holmes_issue_chat pass-through
fix.

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/playbooks/internal/ai_integration.py

@@ -264,12 +264,11 @@ def holmes_issue_chat(event: ExecutionBaseEvent, params: HolmesIssueChatParams):
     conversation_title = build_conversation_title(params)
     params_resource_kind = params.resource.kind or ""
     try:
+        params_dict = params.dict(exclude={"holmes_url", "render_graph_images", "resource", "context"})
         holmes_req = HolmesIssueChatRequest(
-            ask=params.ask,
-            conversation_history=params.conversation_history,
+            **params_dict,
             investigation_result=params.context.investigation_result,
             issue_type=params.context.issue_type,
-            model=params.model,
         )
         result = requests.post(f"{holmes_url}/api/issue_chat", data=holmes_req.json())
         result.raise_for_status()