fix(guardrails): Propagate mutated text to proxied request/response

[JashG]

Summary

• Write input-rail masked/transformed text back onto the last user message in the proxied request (string content only). This matches the library behavior I observe - i.e. only the last user message is modified by rails that might redact/modify content (such as PII redaction).
• Write output-rail masked text back onto choices[0].message.content.
• Mutate request.body in place (sanitize + messages replace) instead of reconstructing InferenceRequest, so fields like typed_body are preserved; IGW re-derives typed_body after the plugin.

Summary by CodeRabbit

• New Features
  • Input rail transformations now sanitize and forward updated upstream messages, rewriting the last role=user message content when text-only.
  • Output rail transformations are propagated into the assistant content returned to clients.
• Bug Fixes
  • Prevents masking/write-back for non-text (multimodal-like) user inputs and avoids updates when rail output matches the original content.
  • Improves safety by avoiding unintended request/response mutations.
• Documentation
  • Clarified the OpenAI-shaped generation response format used for rail-modified content.
• Tests
  • Expanded unit coverage for input masking behavior, multimodal skipping, and output write-back/update detection.

[coderabbitai]

Note

Reviews paused

It looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the reviews.auto_review.auto_pause_after_reviewed_commits setting.

Use the following commands to manage reviews:

• @coderabbitai resume to resume automatic reviews.
• @coderabbitai review to trigger a single review.

Use the checkboxes below for quick actions:

• ▶️ Resume reviews
• 🔍 Trigger review

📝 Walkthrough

Walkthrough

Input rails now propagate transformed user content into sanitized upstream requests without mutating originals. Output rails rewrite assistant content in returned responses while preserving upstream data and handling non-string message content.

Changes

Rail masking propagation

Layer / File(s)	Summary
Response write-back utilities plugins/nemo-guardrails/src/nemo_guardrails_plugin/responses.py, plugins/nemo-guardrails/tests/unit/test_responses.py	Adds input/output rail content resolution, defensive message updates, masked assistant response construction, and unit coverage for unchanged, multimodal, and non-mutating cases.
Middleware input propagation plugins/nemo-guardrails/src/nemo_guardrails_plugin/middleware.py, plugins/nemo-guardrails/tests/unit/test_middleware.py	Sanitizes the existing request body, applies input-rail modifications to a copied message list, forwards updated messages upstream, and tests string versus non-string user content.

Sequence Diagram(s)

sequenceDiagram
  participant Client
  participant process_request
  participant apply_input_rail_modifications
  participant UpstreamModel
  participant build_output_response_body
  Client->>process_request: Request with messages
  process_request->>apply_input_rail_modifications: Input-rail GenerationResponse
  apply_input_rail_modifications-->>process_request: Copied messages with modified user content
  process_request->>UpstreamModel: Sanitized request with updated messages
  UpstreamModel-->>build_output_response_body: Upstream response
  build_output_response_body-->>Client: Response with masked assistant content

Loading

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 16.67% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	Title accurately summarizes the main change: propagating rail-mutated text through proxied requests and responses.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch jgulabrai/guardrails-pii-propagate

---

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

Suite	Lines Covered	Line Rate	Branch Rate
Unit Tests	27113/34832	77.8%	62.1%
Integration Tests	15891/33443	47.5%	19.9%

[gabwow]

thanks, one additional optional nit.