llm-integration

name	llm-integration
description	Add or modify integration-api LLM providers with caller factory wiring, unified provider routing, streaming behavior, and metric/audit compatibility.

LLM Integration Skill

Mission

Add provider callers for chat/stream/verify (plus optional embedding/reranking) with consistent metrics and routing.

Inputs expected from user

1. Required capabilities: chat, stream, verify, embedding, reranking.
2. Auth and endpoint model.
3. Model parameter/tool-calling constraints.

If missing:

• Implement chat + stream + verify first.
• Add embedding/reranking only if provider supports them.

Hard boundaries

In scope:

• api/integration-api/internal/caller/<provider>/...
• api/integration-api/internal/caller/caller.go
• api/integration-api/internal/type/callers.go if interfaces change
• api/integration-api/api/unified_provider.go
• pkg/clients/integration/integration_client.go if routing updates are needed
• provider model metadata in ui/src/providers/<provider>/...

Out of scope:

• assistant-api telephony/STT/TTS/EOS/VAD internals

Implementation workflow

1. Build provider caller package (llm, verify-credential, optional embedding/reranking).
2. Register in caller factory switches.
3. Ensure unified provider endpoint behavior is consistent.
4. Keep MetricBuilder lifecycle consistent (TIME_TAKEN, STATUS).
5. Add tests for success/failure and stream completion.
6. Update provider model metadata where required.

Validation commands

• go test ./api/integration-api/internal/caller/...
• go test ./api/integration-api/api/...
• go test ./pkg/clients/integration/...
• ./skills/llm-integration/scripts/validate.sh --check-diff --provider <provider>

References

• references/checklist.md
• references/llm-checklist.md
• examples/sample.md