Prompt para Avaliar e Projetar Gestão de Contexto em IA (Claude, Notion, Kuse, MyMind)

Base on the observed approaches to context management (Claude: system prompts + conversation summaries to compress history; Notion: document-level context; Kuse: agentic folder system to narrow context; MyMind: human-curated inputs before prompting). These trade-offs affect context length, relevance, and control. Your task is to design a comprehensive, reusable prompt to guide an AI in analyzing, comparing, and prototyping context-management strategies, and to produce an implementation plan.

Instructions:
1) Define a unified rubric for context management in LLMs, with metrics including: context length (tokens), relevance/quality score, latency (ms), cost (compute), controllability, robustness to noise, privacy risk, and user effort.
2) Describe how to implement each of the four approaches within prompts and tooling:
   - System-prompt + summaries (Claude-like)
   - Document-context (Notion-like)
   - Agentic folders (Kuse-like)
   - Human-curated inputs (MyMind-like)
   For each, provide: input data flows, example prompts, storage/recall strategies, and a minimal orchestration blueprint.
3) Propose three hybrid approaches that combine strengths of the four methods. For each hybrid, specify prerequisites, a step-by-step implementation plan, and a concrete evaluation plan.
4) Provide a minimal viable product (MVP) plan to pilot the framework and a structured evaluation sheet template to collect results across experiments. The evaluation sheet should include fields for approach name, measured context length, average relevance score, latency, cost, user control score, and notes.
5) Deliver a concise one-paragraph conclusion recommending, by use-case, which approach or hybrid is most suitable (e.g., real-time chat, document search, long-running agent tasks), plus a quick-start MVP checklist.
6) Add a brief section on ethical considerations: privacy, consent, data handling, and potential bias in context selection.
7) Output a final recommendation on which AI tool among: chatgpt, claude, deepseek, gemini, copilot, perplexity would be best suited to implement and iterate on this framework, with justification.

Output note: Produce the required outputs as a structured JSON snippet that could be fed into another tool for execution, including a clear mapping from each approach to its evaluation plan and to the proposed MVP tasks.