Mistral AI Skill Pack

24 production-ready skills for Mistral AI integration — chat completions, embeddings, function calling, agents, batch API, vision, code generation, and enterprise operations.

SDK: @mistralai/mistralai (TypeScript, ESM-only) | mistralai (Python) API Base: api.mistral.ai | Console: console.mistral.ai By: Tons of Skills / Intent Solutions

Installation

/plugin install mistral-pack@claude-code-plugins-plus

Models Reference

| Model ID | Context | Input $/M | Output $/M | Best For | |----------|---------|-----------|------------|----------| | mistral-small-latest | 256k | $0.10 | $0.30 | General purpose, fast | | mistral-large-latest | 256k | $0.50 | $1.50 | Complex reasoning | | codestral-latest | 256k | $0.30 | $0.90 | Code generation + FIM | | devstral-latest | 256k | $0.40 | $2.00 | Agentic coding | | pixtral-large-latest | 128k | $2.00 | $6.00 | Vision + text | | ministral-latest | 256k | $0.10 | $0.10 | Edge, simple tasks | | mistral-embed | 8k | $0.10 | — | 1024-dim embeddings | | mistral-moderation-latest | — | — | — | Content moderation | | Batch API (any model) | — | 50% off | 50% off | Non-realtime bulk |

Skills (24)

Standard (S01–S12)

| # | Skill | What It Does | |---|-------|-------------| | S01 | mistral-install-auth | SDK installation, API key config, secret manager setup | | S02 | mistral-hello-world | First chat completion, streaming, JSON mode, multi-turn | | S03 | mistral-local-dev-loop | Project scaffold, hot reload, Vitest mocking, integration tests | | S04 | mistral-sdk-patterns | Singleton client, structured output, retry/backoff, usage tracking | | S05 | mistral-core-workflow-a | Chat completions, streaming SSE, guardrails, model selection | | S06 | mistral-core-workflow-b | Embeddings (1024-dim), function calling loop, RAG pipeline | | S07 | mistral-common-errors | HTTP error reference (401/429/400/5xx), circuit breaker, ESM fixes | | S08 | mistral-debug-bundle | Diagnostic bundle script, debug helper, request logger | | S09 | mistral-rate-limits | Token-aware limiter, Retry-After handling, model fallback | | S10 | mistral-security-basics | Key management, prompt injection defense, moderation API | | S11 | mistral-prod-checklist | Pre-deploy verification, health check, circuit breaker, rollout | | S12 | mistral-upgrade-migration | v0.x to v1.x migration, automated transforms, model deprecations |

Pro (P13–P18)

| # | Skill | What It Does | |---|-------|-------------| | P13 | mistral-ci-integration | GitHub Actions workflow, prompt regression tests, cost estimation | | P14 | mistral-deploy-integration | Vercel Edge, Docker, Cloud Run, self-hosted vLLM deployment | | P15 | mistral-webhooks-events | Agents API, Batch API, event bus, BullMQ jobs, Python async | | P16 | mistral-performance-tuning | Model selection, streaming, caching, prompt optimization, FIM | | P17 | mistral-cost-tuning | Pricing calculator, smart model router, budget manager, batch savings | | P18 | mistral-reference-architecture | Layered structure, Zod config, error classes, service layer, prompts |

Flagship (F19–F24)

| # | Skill | What It Does | |---|-------|-------------| | F19 | mistral-multi-env-setup | Per-env config, secret managers (GCP/AWS), feature flags, guards | | F20 | mistral-observability | Prometheus metrics, Grafana panels, alerting rules, structured logs | | F21 | mistral-incident-runbook | Triage script, decision tree, per-error mitigation, postmortem | | F22 | mistral-data-handling | PII redaction, fine-tune sanitization, conversation TTL, GDPR erasure | | F23 | mistral-enterprise-rbac | Workspace isolation, scoped keys, model gateway, spending limits | | F24 | mistral-migration-deep-dive | OpenAI/Anthropic adapter, feature-flag rollout, A/B validation |

Usage

Skills activate automatically based on context:

• "Set up Mistral" activates mistral-install-auth
• "Debug this Mistral error" activates mistral-common-errors
• "Deploy my Mistral app" activates mistral-deploy-integration
• "Migrate from OpenAI" activates mistral-migration-deep-dive
• "Optimize Mistral costs" activates mistral-cost-tuning

Key Capabilities Covered

• Chat Completions — streaming, multi-turn, JSON mode, JSON Schema mode
• Embeddings — text + code, 1024 dimensions, batch processing
• Function Calling — tool definitions, execution loop, parallel calls
• Agents API — stateful conversations, built-in tools, handoffs
• Batch API — 50% cost reduction for bulk workloads
• Vision — image understanding with Pixtral models
• Code Generation — Codestral, Fill-in-the-Middle (FIM)
• Moderation — mistral-moderation-latest, custom guardrails
• Fine-Tuning — dataset preparation, job management
• Self-Hosting — vLLM deployment for data sovereignty

License

MIT