adapter-openai

Files (1)

• SKILL.md

SKILL.md (first lines)

# Adapter: OpenAI

## Focus

- Safe and reliable model invocation patterns.
- Cost/latency-aware configuration.
- Structured output handling and fallbacks.

## Starter Checklist

- [ ] API keys remain server-side.
- [ ] Model outputs are validated before trust.
- [ ] Error and timeout paths are explicit.
- [ ] Prompt context stays scoped to required data.

## Implementation Pattern

Evidence: 3-repo field extraction (2x Next.js, 1x Python agent backend) — see
`.ai/MERGE_REPORT.md`.

- Use one singleton/shared client accessor for the OpenAI SDK, never construct a new client
  per call site.
- Use named model-tier constants (e.g. `FAST_MODEL`, `WRITING_MODEL`) instead of inline model-ID
  string literals scattered across service files.
- Enforce structured/JSON-schema output on calls used for routing, classification, or any
  machine-consumed decision; free-text parsing is only acceptable for pure user-facing prose.
- Fail fast and explicitly at startup/health-check when `OPENAI_API_KEY` is required but
  missing; never let the app continue with a partially initialized AI client.

If the current project already has an established pattern for any of these, follow the existing
pattern instead of this default.

From JahacksV1/day-one-agent-kit, MIT.

More from JahacksV1/day-one-agent-kit