Cohere Incident Runbook
Overview
Restore a safe user experience first, preserve evidence, and distinguish provider incidents from local regressions before making changes.
Prerequisites
- The target repository, runtime, environment, and accountable owner
- An approved Cohere team and key for any live verification
- Current quality, security, privacy, capacity, and change-control requirements
Tool Discipline
Use Read, Glob, and Grep to inspect code, configuration, and evidence. Use WebFetch only for current Cohere primary documentation. Use Write or Edit only when the user requested implementation and the exact target files are known; never write credentials or customer content.
Current Contract
- Check Cohere's status page, but treat service-specific evidence as authoritative for your workload.
- Separate authentication, billing, throttling, model retirement, provider
5xx, retrieval, and application failures. - Use only pre-approved degradation paths such as queueing, cached answers, read-only mode, or an evaluated model route.
- Never debug by dumping keys, prompts, documents, or unrestricted environment variables.
Authentication
Use an environment-specific key injected from an approved secret manager. Never print, persist, commit, or place CO_API_KEY in an example. Confirm access with the least costly bounded operation appropriate to the task, and treat key creation, rotation, revocation, role changes, and production-capacity requests as owner-approved actions.
Instructions
- Open the incident record and capture start time, impact, environment, endpoint, model, release, and correlation IDs.
- Check provider status and compare control traffic, regions, models, and recent local changes.
- Classify the failure and apply the smallest reversible mitigation within the incident authority matrix.
- Protect data and side effects while monitoring recovery against explicit thresholds.
- Escalate to Cohere with the redacted debug bundle when provider assistance is required.
- After recovery, verify normal quality and capacity, document timeline and cause, and assign preventive work.
Approval Boundaries
Do not expose or rotate keys, change Cohere Team roles, accept commercial terms, enable sensitive production data, increase spend or capacity, switch production models, send a support bundle, or execute model-proposed side effects without the accountable owner's approval. Keep diagnosis read-only unless implementation was requested.
Output
Return the resolved API and model contract, files or settings inspected, evidence collected, validation result, remaining risk, owner, and rollback or next action. Redact keys, authorization headers, prompts, retrieved documents, embeddings, customer identifiers, and unrestricted environment output.
Error Handling
| Condition | Response |
|---|---|
| Unknown cause | Prefer traffic reduction or safe degradation over speculative code changes. |
| Credential suspected | Rotate through the security incident process. |
| Model retired | Use only a pre-evaluated replacement or hold affected traffic. |
| Recovery unstable | Keep the incident open and extend observation. |
Examples
Use this compact handoff shape to keep the selected scope, validation evidence, and operational result reviewable.
Input:
incident=INC-77; symptom=429-spike; release=unchanged; status=investigating
Expected handoff:
class=capacity; mitigation=queue-cap; impact=stable; evidence=preserved