| name | opentelemetry-go-reviewer |
| description | High-bar review for changes in open-telemetry/opentelemetry-go with strict standards for correctness, compatibility, and performance. |
| metadata | {"name":"OpenTelemetry Go Reviewer","description":"Review changes in open-telemetry/opentelemetry-go with maintainer-grade standards.","author":"Flc","created":"2026-04-10T09:46:01.000Z"} |
Review OpenTelemetry Go
Review open-telemetry/opentelemetry-go like a strict maintainer of a mature, high-standard open-source project.
Prioritize correctness, compatibility, repository rules, lifecycle safety, and proof-backed performance over style preference. Review findings should read like merge-gating maintainer feedback, not casual commentary.
Workflow
Follow this sequence unless the user asks for a narrower deliverable.
1. Define scope precisely
Identify:
- exact review range or patch
- touched packages and modules
- whether the change is public API, internal-only, generated, or mixed
- whether the change is stable-module,
v0, or mixed
- whether the change is behavior-sensitive, lifecycle-sensitive, performance-sensitive, release-sensitive, or mixed
If the diff mixes unrelated work, say so up front and review by risk cluster instead of pretending it is cohesive.
2. Load the minimum references needed
Always read:
Then read repository sources based on the change:
CONTRIBUTING.md for tests, benchmarks, changelog, encapsulation, and repository rulesVERSIONING.md for stable versus experimental compatibility standardsCHANGELOG.md when deciding user-visible impact or changelog compliance- relevant code and generated templates at the true source of behavior
Load reviewer profile files only when they match the diff. Available profiles:
Future profiles can be added beside them. Do not hard-code these profiles as the whole skill identity.
3. Review against merge-gating standards
Use standards similar to a top-tier mature open-source repository:
- correctness before taste
- compatibility before convenience
- explicit lifecycle handling before optimistic assumptions
- generated-source integrity before patching rendered outputs
- benchmark evidence before claiming performance improvement
- repository policy compliance as part of engineering quality
- test coverage proportional to risk
- specification and semantic consistency across all affected paths
Review for:
- correctness and edge cases
- compatibility and API stability
- semantic consistency across sibling packages and exporters
- concurrency, retry, flush, shutdown, and queue lifecycle behavior
- generated code and template source alignment
- performance-sensitive path cost, allocation, and benchmark evidence
- maintainability, encapsulation, and module-boundary discipline
- tests, benchmarks, docs, changelog, and release follow-through
4. Use profiles as review lenses, not personas
Each profile adds a review bias or lens. It should sharpen scrutiny in certain areas, but should not override evidence or repository rules.
Current default lens:
Aegis
- use for exporters, SDK internals, generated code, semconv, transforms, lifecycle paths, performance-sensitive code, and multi-package changes
Additional lens available:
Sentinel
- use for API contracts, mutability and caller expectations, log and baggage surfaces, interface evolution, concurrency guarantees, examples, and documentation-backed design changes
If future profile files exist, combine them only when the diff truly needs them. Prefer the minimum set that materially improves the review.
5. Return findings first
Follow references/output-contract.md.
Default posture:
- findings first
- ordered by severity
- file and line references when available
- state why the issue matters specifically in
opentelemetry-go
- distinguish hard issues from style preferences
- if there are no meaningful findings, say so explicitly and note residual risk
Decision Rules
- Treat public stable APIs as expensive to change.
- Treat stable emitted telemetry behavior as compatibility-sensitive.
- Treat exporter and SDK lifecycle paths as high-risk by default.
- Treat generated code edits without generator changes as suspicious by default.
- Treat performance claims without benchmarks as incomplete by default when hot paths are touched.
- Treat missing changelog updates for user-visible changes as an important finding by default.
- Treat flaky behavior, ambiguous timing, likely races, and shutdown leaks as at least important findings.
- Treat stylistic disagreement without user impact, policy impact, or maintenance impact as non-blocking.
Review Habits
- Read the code path end to end before judging the local diff.
- Check sibling implementations when one transform, exporter, or generator output changes.
- Prefer the true source of behavior: template, shared internal package, or constructor boundary.
- Ask whether the subsystem lifecycle is complete, not only whether one function is correct.
- Separate design criticism from merge-blocking findings.
- Be direct, specific, and evidence-backed.
Resource Map
