lcp-openai-serve
Develop apps/openai-serve (Go). Keep OpenAI gateway behavior correct and avoid logging prompts/outputs.
When & Why to Use This Skill
This Claude skill provides specialized expertise for developing and maintaining an OpenAI-compatible HTTP gateway (openai-serve) written in Go. It prioritizes architectural robustness through the Single Responsibility Principle (SRP), strict API contract adherence, and critical data privacy by enforcing non-negotiable rules against logging sensitive prompts or model outputs. It is designed to help developers build secure, stable, and high-performance AI middleware.
Use Cases
- Case 1: Developing and extending OpenAI-compatible API endpoints in Go with strict JSON validation and explicit error handling for unsupported fields.
- Case 2: Implementing privacy-first logging configurations that ensure diagnostic data is captured without exposing sensitive user prompts or model completions.
- Case 3: Maintaining high code quality and preventing regressions using Test-Driven Development (TDD) and rigorous linting with golangci-lint.
- Case 4: Refactoring gateway infrastructure to align with SRP and ubiquitous language, ensuring the codebase remains maintainable and scalable.
| name | lcp-openai-serve |
|---|---|
| description | Develop apps/openai-serve (Go). Keep OpenAI gateway behavior correct and avoid logging prompts/outputs. |
| short-description | Work on the OpenAI-compatible gateway |
You are working on the OpenAI-compatible HTTP gateway in apps/openai-serve/.
Scope
- Module:
apps/openai-serve/ - Runtime contract:
apps/openai-serve/README.md
Development principles
These are the consolidated rules that previously lived in apps/openai-serve/AGENTS.md:
- Design for robustness. Keep modules aligned with SRP.
- Use a ubiquitous language. Keep terminology consistent across code and docs.
- Test-first. Prefer TDD when changing behavior.
- Run golangci-lint v2. When changing code, run
make lintand fix findings. - Use
cmp.Diffin tests. Compare expected vs actual withgithub.com/google/go-cmp/cmp.Diff. - Make logging diagnosable. Logging level MUST be configurable (for example via env vars).
- Prefer lnd libraries. Use lnd-provided APIs/libraries before re-implementing.
Non-negotiables
- Logging is sensitive: do not log raw prompts (
messages[].content) or raw model outputs. - Keep behavior stable and explicit: JSON decoding is strict, unsupported fields must be rejected as documented.
Workflow
- Read
apps/openai-serve/README.mdbefore changing behavior. - When adding config/env vars, document them in
apps/openai-serve/README.mdand keep defaults sane. - When changing request/response structures, update tests and ensure the documented constraints still hold.
Validation (run from apps/openai-serve/)
make testmake lintmake fmt