← Files LLM & Agent Builder CopilotARCHIVED FILE

references/implementation-and-sdks.md

1.03 KB · Nov 19, 2026 · 00:40 UTC

↓ Download file

See the change to this file →

# SDK choice and implementation discipline
Choose direct API calls for small, bounded model integrations; choose an agent SDK when orchestration, sessions, tool execution and tracing lower complexity; choose a durable workflow/state machine for critical resume/retry/approval paths. Avoid switching stacks just for popularity.

Inspect project language, lockfile, current SDK version and existing tests. Verify exact APIs in installed packages and official docs. Separate deterministic policy, tool implementation, state persistence and model instructions. Design retry classes: read-only, idempotent write, non-idempotent write; never blanket-retry all operations. Add integration tests through mocks or test providers. The host must really have coding tools to implement; don't fabricate files or executed tests.

Current reference points: https://openai.github.io/openai-agents-python/ ; https://openai.github.io/openai-agents-js/ ; https://platform.openai.com/docs/guides/tools ; https://docs.anthropic.com/ ; https://ai.google.dev/gemini-api/docs

SHA-256: e2fe11a428d046139ffcad0569a0c3405e32decc346fd12f4ee3b8caea2610c6