The Tradeics AI skill teaches agents how to use Tradeics assistive AI and the OpenAI-compatible LLM proxy safely — embeddings, chat completions, and product helpers — without treating model output as a payment or award authority.
Goals
- Point agents at Tradeics AI guides and the LLM API.
- Use AI MCP for tool-shaped assistants; use HTTPS for high-volume inference.
- Keep a hard boundary: assistive, not authoritative.
Surfaces
| Surface | Use for | Docs |
|---|---|---|
| Product AI helpers | Summaries, classification, drafting | Capabilities |
| LLM proxy | Your app’s Chat Completions / embeddings | Call the LLM |
| AI MCP | Agent-discovered assist tools | AI MCP |
Base (placeholder):
https://llm.tradeics.example/v1Behavior contract
- Do not auto-award POs, release wallets, or settle invoices from model output alone.
- Prefer structured JSON when chaining into REST writes.
- Rate-limit autonomous loops; cache embeddings when possible.
- Label AI-generated text in operator UIs when policy requires it.
Contents checklist
- Auth notes (
Authorizationandx-tradeics-llm-api-key). - Model selection / placeholders until GA catalog ships.
- Prompt patterns for RFX summary and catalog classification.
- Escalation to REST for durable storage of AI results.
- Pairing with Platform skill when AI touches live RFXs.