[Pujarini Mohapatra (biswapm)]

Summary

Adds the privacy-safe OpenTelemetry attributes required by the Model Observation pipeline.

• Capture gen_ai.response.model when the provider returns the canonical served model.
• Capture gen_ai.usage.cache_read.input_tokens.
• Capture gen_ai.usage.cache_write.input_tokens.
• Normalize the inference gen_ai.operation.name value to lowercase OTel values, including chat.
• Capture model endpoint information as server.address and an optional numeric server.port.
• Support attributes supplied when the inference scope starts or recorded later when the response becomes available.
• Add focused tests covering the emitted attribute names and value types.

Validation

The repository’s targeted inference tests, build, formatting, and linting checks pass.

[Copilot]

🟡 Changes recommended

Multiword operation names become unrecognized and are dropped by the exporter; late-recording tests and the changelog also need completion.

3 open findings

• Add changelog entry for new public inference fields and methods · New
• Map all inference enums to canonical GenAI operation names · New
• Assert response parameter values are emitted in spans · New

What changed in this PR

Adds model-observation attributes to manual inference telemetry.

Changes:

• Records response models and cache token counts.
• Normalizes inference operation names.
• Adds semantic-convention constants and focused tests.

File	Description
src/​a365/​contracts.ts	Extends inference details with response fields.
src/​a365/​scopes/​InferenceScope.ts	Emits new attributes and adds late-recording methods.
src/​genai/​semconv.ts	Defines cache token attribute constants.
test/​internal/​unit/​a365/​scopes.test.ts	Tests new inference attributes and methods.

🧠 Review effort: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

🟡 Changes recommended

The documented operation allowlist omits the newly recognized canonical values.

1 open finding

• Document canonical text_completion and generate_content operation names · New

3 resolved since last review

• Add changelog entry for new public inference fields and methods
• Assert response parameter values are emitted in spans
• Map all inference enums to canonical GenAI operation names

🧠 Review effort: Balanced

[Pujarini Mohapatra (biswapm)]

Fixed in ae0c7c6: added \ ext_completion\ and \generate_content\ to the documented baggage-enrichment allowlist in A365_DOCUMENTATION.md so it matches \GEN_AI_OPERATION_NAMES.

[Copilot]

🟢 Approval recommended

The changes are focused, include relevant regression tests, and have no identified blocking issues.

0 open findings

1 resolved since last review

• Document canonical text_completion and generate_content operation names

🧠 Review effort: Balanced

[Nikhil Navakiran (nikhilNava)]

Two points on operation-name normalization: handling unmapped runtime strings without silently dropping spans, and using the canonical operation in the span name as well as its attribute.

[Copilot]

🟢 Approval recommended

Static review found no blocking issues, and focused regression tests cover the changes, though tests were not executed during review.

No open issues 🎉

_{Review effort: Balanced}