[bbednarski9]

Overview

After Codex uses a tool, its next model request contains the user’s prompt, the tool call, and the tool’s result. Relay can accidentally export only the result, losing the messages that explain why the tool ran.

This PR fixes the history filter so the exported conversation keeps that context.

For example, in the reproduced case:

Model request:   instructions → user prompt → tool call → tool result
Exported before:                                         tool result
Exported after:  instructions → user prompt → tool call → tool result

The model receives the complete request throughout. The bug affects the history Relay records in telemetry.

• I confirm this contribution is my own work, or I have the right to submit it under this project's license.
• I searched existing issues and open pull requests, and this does not duplicate existing work.

Details

Why it happens. When full-history export is disabled, Relay trims later model requests to the current user turn. OpenAI Responses messages can carry an item ID or extra provider fields. Relay keeps those messages in their original provider format to preserve those fields, but the history filter only recognized its regular User message type. It missed the user prompt in these requests and fell back to keeping the final item—often the tool result.

What changes. The filter reads the message’s role even when Relay preserves its original provider format. It keeps the latest user prompt and everything after it, together with system and developer instructions. Developer instructions also needed to be added to the retained instruction types. Item IDs and extra fields remain intact.

Validation. Reproduced on Codex CLI 0.162.1 with live-model and controlled-provider runs. The focused regression covers messages with and without IDs, prior conversation history, and tool calls/results. Rust, Python, and Node canonical suites passed on a combined checkout of the three related fixes; repository pre-commit hooks passed on this branch.

This restores missing conversation evidence. Custom-tool normalization is covered by #1644, and ownership hints by #1645. Establishing the link between a code-mode wrapper and its inner executions remains separate work.

Where should the reviewer start?

Start with the before/after regression, then the message-role helper and history filter.

Related Issues: (use one of the action keywords Closes / Fixes / Resolves / Relates to)

• Relates to: none

Summary by CodeRabbit

• Bug Fixes
  • OpenAI Responses history projection now correctly identifies user turns from native message roles, including messages with item IDs or additional provider fields. System and developer instructions are retained, while the provider request remains unchanged.

[coderabbitai]

Important

Review skipped

Review was skipped as selected files did not have any reviewable changes.

Run configuration

• Source mode: Auto
• Repository: NVIDIA/NeMo-Relay
• Branch: release/0.11
• File: .coderabbit.yaml
• Commit: d71363b
• Layers: Repository YAML
• Review profile: ASSERTIVE
• Plan: Enterprise
• Run ID: 45162eff-78a1-4ada-817e-24f2e8c26110

Commits

Reviewing files that changed from the base of the PR and between 7448918 and 6dee235.

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• Trigger review

No actionable comments were generated in the recent review.

Walkthrough

Turn projection now recognizes explicit roles in OpenAI Responses native messages of kind message. It uses those roles to identify the current user turn and retain system and developer instructions. Tests and documentation cover the projection behavior.

Changes

OpenAI Responses turn projection

Layer / File(s)	Summary
Role-based turn projection crates/core/src/api/llm.rs, crates/core/tests/unit/llm_api_tests.rs, docs/configure-plugins/observability/opentelemetry.mdx	A shared role lookup recognizes user, system, and developer roles in qualifying native messages. Projection detection and history retention use these roles. Tests cover history variants and preserved fields. Documentation describes the behavior and states that projection does not change the provider request.

Estimated code review effort: 2 (Simple) | ~10 minutes

Change: Bug fix

Merge Risk: Minimal · previous review 73a3d

Current-turn telemetry retains native user messages and instructions. No merge-blocking risk was identified.

Pre-merge checks: 4 · 1

Failed checks: 1 warning

Check name	Status	Explanation	Resolution
Docstring Coverage		Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff with cyclomatic complexity of at least 2. Documentat…	Write docstrings for the functions missing them to satisfy the coverage threshold.

Passed checks: 4 passed

Check name	Status	Explanation
Title check		The title follows Conventional Commits format, uses the allowed lowercase type "fix," provides a concise imperative summary, is 63 characters long, and has no trailing period. It accurately describes …
Linked Issues check		Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check		Check skipped because no linked issues were found for this pull request.
Description check		The description follows the required template. It includes an overview, checked confirmations, implementation details, validation results, reviewer starting points, and a Related Issues section using …

Full details: Title check

Explanation

The title follows Conventional Commits format, uses the allowed lowercase type "fix," provides a concise imperative summary, is 63 characters long, and has no trailing period. It accurately describes the main change.

Full details: Description check

Explanation

The description follows the required template. It includes an overview, checked confirmations, implementation details, validation results, reviewer starting points, and a Related Issues section using the required action keyword. The Related Issues entry does not identify a specific issue, but the description is otherwise complete and directly relevant.

Full details: Docstring Coverage

Explanation

Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff with cyclomatic complexity of at least 2. Documentation-only edits and deletions within surviving functions are included; unchanged diff context and fully deleted functions are excluded. Analyzed 3 functions across 2 files. (1 skipped: 1 unsupported.)

Resolution

Write docstrings for the functions missing them to satisfy the coverage threshold.

Finishing Touches

Generate unit tests (beta)

• Create a new PR

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Tip

🎯 Customize your risk assessments

Guide how CodeRabbit assesses merge risk and security architecture risk. Highlight sensitive changes and explain your security boundaries using Merge Risk Instructions and Security Risk Instructions in your review settings. Security architecture risk assessments require the Advanced plan or above.

✅ Automatic approvals, with your rules

Let CodeRabbit approve pull requests once required threads are resolved, pre-merge checks pass, and your risk limits are met. Enable Automatic approvals in review settings, with overrides for teams or file paths.

Recent review info

Run configuration

• Source mode: Auto
• Repository: NVIDIA/NeMo-Relay
• Branch: release/0.11
• File: .coderabbit.yaml
• Commit: 57184ba
• Layers: Repository YAML
• Review profile: ASSERTIVE
• Plan: Enterprise
• Run ID: ff9b9504-a4d2-4594-aa3b-1a115dc8767a

Commits

Reviewing files that changed from the base of the PR and between 57184ba and 73a3d99.

Files selected for processing: 3

• crates/core/src/api/llm.rs
• crates/core/tests/unit/llm_api_tests.rs
• docs/configure-plugins/observability/opentelemetry.mdx

Included review availability: This review used your included allowance. Your plan provides up to 12 included reviews per hour; 11 remain after this review.

Recent review details

Additional context used

Code guidelines: 5

.agents/skills/contribute-docs/SKILL.md — configured
.agents/skills/maintain-observability/SKILL.md — configured
AGENTS.md — configured
.agents/skills/add-binding-feature/SKILL.md — configured
.agents/skills/review-doc-style/SKILL.md — configured

Path-based instructions: 8

Review documentation for technical accuracy against the current API, command correctness, and consistency across language bindings.

CodeRabbit configuration file

Files:

• docs/configure-plugins/observability/opentelemetry.mdx

Tests should cover the behavior promised by the changed API surface, including error paths and cross-request isolation where relevant.

CodeRabbit configuration file

Files:

• crates/core/tests/unit/llm_api_tests.rs

Review the Rust runtime for async correctness, scope isolation, middleware ordering, and event lifecycle regressions.

CodeRabbit configuration file

Files:

• crates/core/tests/unit/llm_api_tests.rs
• crates/core/src/api/llm.rs

Source excerpt: In MDX files, top-of-file comments must use JSX comment delimiters: `{/*` to open and `*/}` to close.

CodeRabbit inference engine (.agents/skills/contribute-docs/SKILL.md)

Files:

• docs/configure-plugins/observability/opentelemetry.mdx

Source excerpt: Core event model and emitted fields `crates/core/src/observability/atif.rs` `crates/core/src/observability/otel.rs` `crates/core/src/observability/openinference.rs` FFI and binding-native wrappers where the config or lifecyc...

CodeRabbit inference engine (.agents/skills/maintain-observability/SKILL.md)

Files:

• docs/configure-plugins/observability/opentelemetry.mdx

Source excerpt: Update public documentation and examples when public behavior, package names, supported bindings, or documented workflows change.

CodeRabbit inference engine (AGENTS.md)

Files:

• docs/configure-plugins/observability/opentelemetry.mdx

Source excerpt: [ ] Core function with doc comment in `crates/core/src/api/`

CodeRabbit inference engine (.agents/skills/add-binding-feature/SKILL.md)

Files:

• crates/core/src/api/llm.rs

Source excerpt: Verify MDX files use JSX delimiters for top-of-file SPDX comments.

CodeRabbit inference engine (.agents/skills/review-doc-style/SKILL.md)

Files:

• docs/configure-plugins/observability/opentelemetry.mdx

Additional comments: 3

crates/core/src/api/llm.rs: 1

396-413: LGTM!

Also applies to: 424-425, 455-456

crates/core/tests/unit/llm_api_tests.rs: 1

2850-2898: LGTM!

docs/configure-plugins/observability/opentelemetry.mdx: 1

1183-1188: LGTM!

[github-actions]

Fern docs preview: https://nvidia-preview-pull-request-1643.docs.buildwithfern.com/nemo/relay/dev

[bbednarski9]

fresh before/after smoke test with a live model, using the same prompt on Codex CLI 0.162.1 / gpt-6.1-sol:

Run exactly one shell command printf relay-1643-smoke-ok using tools.exec_command from the exec tool, then reply with the command output. Do not use any other tools or add commentary before the call.

Each run completed one shell invocation, made two model requests, and answered relay-1643-smoke-ok. The comparison below is the second model request, after the tool completed. enable_full_payloads=false in both runs; the after build contains only this PR's fix, without #1644 or #1645.

Follow-up request	Before: release baseline 57184ba79	After: this PR 7448918d3
Items actually sent upstream	9	9
Items retained in telemetry	1	8
Current user prompt	Missing	Retained
Model-requested tool call	Missing	Retained
Tool result	Retained	Retained
Developer instruction/tool-definition items	0	5

Telemetry IDs are anonymized below. Repeated labels refer to the same original identifier.

Before — the complete exported data.content.input contained only the result:

[
  {
    "call_id": "ID001",
    "id": "ID002",
    "output": [
      {
        "text": "Script completed\nWall time 0.2 seconds\nOutput:\n",
        "type": "input_text"
      },
      {
        "text": "relay-1643-smoke-ok",
        "type": "input_text"
      }
    ],
    "type": "custom_tool_call_output"
  }
]

After — the user prompt, tool call, and matching result are all present:

These are the actual non-developer entries from the same data.content.input field. The five retained developer items are omitted from this display; the three entries below preserve the captured content with identifiers replaced by consistent placeholders.

[
  {
    "content": [
      {
        "text": "Run exactly one shell command printf relay-1643-smoke-ok using tools.exec_command from the exec tool, then reply with the command output. Do not use any other tools or add commentary before the call.",
        "type": "input_text"
      }
    ],
    "id": "ID003",
    "role": "user",
    "type": "message"
  },
  {
    "call_id": "ID004",
    "id": "ID005",
    "input": "const r = await tools.exec_command({cmd:\"printf relay-1643-smoke-ok\"}); text(r.output);\n",
    "name": "exec",
    "status": "completed",
    "type": "custom_tool_call"
  },
  {
    "call_id": "ID004",
    "id": "ID006",
    "output": [
      {
        "text": "Script completed\nWall time 0.2 seconds\nOutput:\n",
        "type": "input_text"
      },
      {
        "text": "relay-1643-smoke-ok",
        "type": "input_text"
      }
    ],
    "type": "custom_tool_call_output"
  }
]

The after capture retains the prompt's message ID, the call's original JavaScript input, and the same call_id on the call and result. The earlier environment/user-context item is trimmed according to the existing current-turn policy. Both upstream requests still contained all nine input items.

[mnajafian-nv]

lgtm, thanks

[ericevans-nv]

The history projection now recognizes native Responses user messages and retains system and developer instructions while preserving the original provider items. I found no required corrections in the changed code.