cohere-platform-operations-2026

Files (3)

• SKILL.md
• openai.yaml · agents/
• sources.md · references/

SKILL.md (first lines)

# Cohere Platform Operations (2026)

Use this skill when Cohere is the production platform or ranking/search stack, not just an interchangeable model endpoint.

## Workflow

1. Check current Cohere models and release notes before implementation or migration.
2. Choose the model family by task: generation, embeddings, rerank, or transcription.
3. Prefer official Cohere SDK patterns for auth, request schema, batching, and streaming.
4. Define retry, timeout, and rate-limit behavior before load testing.
5. Validate one representative endpoint per workload before broader rollout.

## Guardrails

- Keep model names pinned in production rather than depending on implicit defaults.
- Treat rerank, embed, and transcription as separate operational surfaces with different latency and cost profiles.
- Record SDK version, endpoint, and region or deployment assumptions in test evidence.
- Keep one fallback path for critical retrieval or generation workflows.

## Output Requirements

- State the chosen Cohere model family and endpoint.
- State one compatibility or migration note.
- State one operational safeguard such as timeout, retry, quota, or fallback.

From mdbabumiamssm/LLMs-Universal-Life-Science-and-Clinical-Skills-, MIT.

More from mdbabumiamssm/LLMs-Universal-Life-Science-and-Clinical-Skills-