[al3xanndru]

Problem

PC-5145: when a DeepRag task fails, the resumed agent job faults with only DeepRag '<name>' did not finish successfully. The agent and the tool span get no reason. In SRE-674993 (prd Japan East) the tenant's AI Trust Layer policy blocked ECS's gemini-2.5-flash. Six DeepRag tasks failed, and the customer could not see that their own governance policy was the cause. SRE-669204, SRE-671555 and SRE-673408 hit the same opaque message.

Root cause

UiPathResumeTriggerReader.read_trigger, DEEP_RAG branch (packages/uipath-platform/src/uipath/platform/resume_triggers/_protocol.py), raises a fixed message for any non-Successful status. It never reads DeepRagResponse.failure_reason (alias failureReason), which the model already parses from the ECS GET response. uipath-runtime builds the job error from exc.message (UiPathRuntimeError.from_resume_trigger_error), so the reason has to be in the message, not in detail.

Change

• When failure_reason is non-empty, the message becomes DeepRag '<name>' did not finish successfully. Reason: <failure_reason>. When it is empty, the message is unchanged.
• The category stays ErrorCategory.USER. See the open questions.
• uipath-platform goes from 0.2.41 to 0.2.42, and both lockfiles are updated, as this repo does for fixes.

Tests

• test_read_deep_rag_trigger_failed now also asserts the full message, including the reason.
• New test_read_deep_rag_trigger_failed_without_reason: with no failure_reason, the message stays as before and has no trailing Reason:.
• Results in packages/uipath-platform:
  • pytest tests/services/test_hitl.py: 144 passed.
  • ruff check .: clean.
  • ruff format --check .: clean.
  • mypy on the changed files: clean.

Risks / open questions

• Category. The ticket asks to derive the category from the failure instead of hard-coding USER. The DeepRag API response has only the free-text failureReason; ECS's internal UserError flag is not exposed. Parsing free text for a category would be brittle. Moving everything to SYSTEM would page Agents for customer-side failures (governance policy, oversized documents). So USER is kept. A real fix needs ECS (Context Grounding) to return a structured user/system flag or error code on the DeepRag GET. Once it does, map that flag here.
• The reason may be empty. SRE-674993 could not confirm whether ECS fills failureReason for the governance-403 case. This change only surfaces what ECS returns. For that failure to become visible, ECS also has to write the LLM Gateway message to failureReason; that is tracked as the external action in the triage.
• Raw ECS text. failureReason is passed through verbatim into a user-visible error. It is ECS's customer-facing field, so this is expected.
• To reach agents, a uipath-platform release has to be picked up by uipath-agents-python / uipath-langchain.

🤖 Generated with Claude Code

---

Drafted by the wabbit-dri draft-fixes loop for PC-5145 (draft #2), for a human to review: not ready to merge as is.

🤖 Generated with Claude Code

[sonarqubecloud]

Quality Gate passed

Issues
1 New issue
0 Accepted issues

Measures
0 Security Hotspots
100.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud