Skills DirectorySkills Directory
SkillsLearnSecurityCategoriesDocsCommunityBlog
Sign InSubmit Skill
Skills Directory

Security-tested agent skills for Claude, coding agents, and AI workflows.

Directory

  • Browse Skills
  • All Skills A–Z
  • Claude Skills
  • Claude Code Skills
  • Agent Skills
  • Categories
  • Submit a Skill

Learn

  • Learn Hub
  • Install Claude Skills
  • Write SKILL.md
  • Skills vs MCP
  • Directories Compared

Security

  • Security
  • Methodology
  • Secure Claude Skills
  • Security Badges

Company

  • About
  • Community
  • Blog
  • API Docs
  • Advertise

2026 Skills Directory. All rights reserved.

Back to skills

Model Migration Strategy

ASecurity

Use when migrating between model families — eval suite first, avoid per-family hyper-optimization, hedge new failure modes, cost attribution. Triggers on "model migration", "gpt-5 gotchas", "em dashes model", "AI telltale signs", "hyper-optimization prompts", "model failure modes".

2 stars
0 votes
0 copies
0 views
Added 9/19/2026
ai-agentsgo

Works with

claude code

Security Analysis

A100/100

Scanned 9/19/2026

Install to Claude Code

$npx -y skills add majinmagros/magros.ai-skills --skill model-migration-strategy --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Model Migration Strategy?

Add the live security badge to your README — it updates automatically with every re-scan.

Security grade badge for Model Migration Strategy
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/majinmagros-model-migration-strategy/badge)](https://www.skillsdirectory.com/skills/majinmagros-model-migration-strategy)

More formats (shields.io, HTML) on the badges page.

Download Zip
Files
SKILL.md
---
name: model-migration-strategy
description: Use when migrating between model families — eval suite first, avoid per-family hyper-optimization, hedge new failure modes, cost attribution. Triggers on "model migration", "gpt-5 gotchas", "em dashes model", "AI telltale signs", "hyper-optimization prompts", "model failure modes".
metadata:
  origin: ECC
  source_docs:
    - https://docs.anthropic.com/en/docs/managed-agents/models
  video_source: "hm8NzEd5io0 - How founders build on Claude Managed Agents (Claude Oficial)"
  related_skills:
    - roteamento-modelos-baratos
    - claude-model-router
    - eval-harness
    - verification-loop
    - cost-aware-llm-pipeline
---

# Skill: model-migration-strategy — Estratégia de Migração de Modelos

Migração segura entre famílias: **eval suite primeiro**, sem hyper-optimization por família, hedge de **novos failure modes**, cost attribution granular. Código em `references/implementation.md`.

## Quando usar

- Nova família de modelo lançada (GPT-5, Claude 4, etc.)
- Precisa migrar produção de um modelo para outro
- Quer evitar **hyper-optimization** para família específica
- Precisa detectar **novos failure modes** por família
- Quer **attribution granularity** de custos por modelo

## Quando NÃO usar

- Roteamento simples → use `roteamento-modelos-baratos`, `claude-model-router`
- Cost tracking genérico → use `cost-aware-llm-pipeline`
- Eval harness genérico → use `eval-harness`
- Subscription tier routing → use `subscription-tier-routing`

---

## Validação Oficial (2026-09-09)

| Claim | Status | Fonte |
|---|---|---|
| Eval suite FIRST before migration | ✅ | Video + Anthropic best practices |
| Avoid hyper-optimization per model family | ✅ | Video |
| GPT-5: mais em dashes, AI telltale signs | ✅ | Video (empirical) |
| New failure modes per family | ✅ | Video |
| Cost attribution granularity | ✅ | Video |

---

## Os 4 Princípios

1. **EVAL SUITE FIRST** — não migre sem evals que peguem regressão (vibes → curated → customer queries; mesmas rubricas offline + outcomes)
2. **AVOID HYPER-OPTIMIZATION** — prompts otimizados para Opus falham no Sonnet; para GPT-4 falham no GPT-5. Prompting robusto, não family-specific
3. **HEDGE NEW FAILURE MODES** — cada família tem gotchas: GPT-5 (em dashes, sentence structures, tom acadêmico), DeepSeek (caracteres chineses), Opus (over-thinking). Teste gotchas ANTES de otimizar
4. **COST ATTRIBUTION** — telemetria fine-grained: que calls custam o quê, por model/feature/operation; alerta se feature passa de 50% do budget

## Fases de Migração (com auto-rollback)

| Fase | Tráfego | Duração | Sucesso | Rollback |
|---|---|---|---|---|
| canary | 5% | 24h | pass ≥ 0.95 | pass < 0.90 |
| partial | 25% | 48h | pass ≥ 0.97 | pass < 0.93 |
| majority | 75% | 72h | pass ≥ 0.98 | pass < 0.95 |
| full | 100% | — | pass ≥ 0.99 | pass < 0.97 |

## Exemplo

```text
Opus → Sonnet em produção: eval suite primeiro (vibes→curated→customer, mesmas rubricas)
Acha gotcha: prompt hiper-otimizado p/ Opus quebra no Sonnet → prompting robusto
Canary 5%/24h (pass 0.96) → partial → full; custo por feature atribuído no dashboard
```

## Unhobbling no Claude 5 (Batch 17f, #77)

Anthropic removeu ~80% do system prompt do Claude Code nos modelos 5
sem perda mensuravel: modelos novos interpretam intent, e instrucoes
defensivas antigas viram atrito (ate orientacoes conflitantes entre
si). Na migracao para Claude 5, audite CLAUDE.md/skills/system prompts
e remova o defensivo obsoleto em vez de empilhar mais regra. Fonte:
Tharik (Anthropic), "new rules of context engineering".

## Auditoria pre-Fable (Batch 17b, #59)

Skills e prompts escritos antes do Fable 5 costumam ser prescritivos
demais e DEGREDAM a qualidade no novo modelo (guia oficial Anthropic).
Na migracao para Claude 5, revise o inventario e remova instrucoes
antigas (steps fechados, double-checks manuais, continuacoes forcadas)
antes de culpar o modelo. Vale tambem para Astra 6 (OpenAI).

## Failure modes da serie nova (leva YouTube rodada 9)

Checklist ao migrar de serie: AI tells novos, formatacao ilegivel, tool-use que mudou de forma, custo/latencia por task (nao por token). Regra: nunca hiper-otimizar a ultima gota na serie antiga na vespera da migracao — congele e migre com eval suite verde.

Attribution

majinmagrosmajinmagros
View sourceMore from majinmagros →
SSkills DirectorySkills Directory

Ship a skill? Prove it's safe.

Free 120-pattern security scan, letter grade, and an embeddable README badge.

Submit a skill

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments (0)

No comments yet. Be the first to comment!

SSkills DirectorySkills Directory

Ship a skill? Prove it's safe.

Free 120-pattern security scan, letter grade, and an embeddable README badge.

Submit a skill

Related Skills

Caveman

Ultra-compressed communication mode. Cuts token usage ~75% by speaking like caveman while keeping full technical accuracy. Supports intensity levels: lite, full (default), ultra, wenyan-lite, wenyan-full, wenyan-ultra. Use when user says "caveman mode", "talk like caveman", "use caveman", "less tokens", "be brief", or invokes /caveman. Also auto-triggers when token efficiency is requested.

1023331 votes

Hyperplan

Adversarial multi-agent planning skill. Self-orchestrates 5 hostile category members (unspecified-low, unspecified-high, deep, ultrabrain, artistry) via team-mode for ruthless cross-critique debate, distills only the defensible insights, then MANDATORILY hands the distilled insight bundle to the `plan` agent for executable plan formalization. Use when planning needs maximum rigor and surfacing of weak assumptions, blind spots, and over-engineering. Triggers: 'hyperplan', 'hpp', '/hyperplan', ...

686011 votes

Mcp Code Execution

Routes multi-tool workflows through MCP servers for large datasets and pipelines. Use when Bash tool overhead is limiting throughput on data-heavy tasks.

3331 votes

catchup

Recovers prior coding-agent session context by running `catchup <agent> --since-compact`, which extracts a clean summary of a previous Codex, Claude Code, Antigravity, OpenCode, or Pi Agent session. Use when the user says "catch up", "what did the last session do", "get me up to speed", "I switched agents", or asks to recover/summarize a previous session before continuing. Do NOT use for the current conversation, git history, or any non-agent log.

611 votes

math-skill

A comprehensive mathematical reasoning skill for AI assistants — handles arithmetic to research-level problems with rigorous step-by-step reasoning, systematic verification, and transparent uncertainty handling

381 votes
View all in ai-agents →