All AI sessions
Context Compass
Session 2 recap · Context Engineering & Model Selection · Sun 11 Oct
Task first, then model, then effort. The one-page recap: six context layers, seven models, six effort levels, and a prompt to keep.
A prompt to keep
Paste only the part the task needs, then point to the rest instead of pasting it.
[Paste only the "Discount rules" section here: 2 pages, not all 40.]
Using the rules above, write test cases for stacked coupons.
If you need payment details, read docs/payments.md.Sources
Everything on the sheet is checked against these.
- Effective context engineering for AI agentsAnthropic (opens in a new tab)
- Prompting best practices (long context)Anthropic (opens in a new tab)
- Get started with Claude CoworkAnthropic (opens in a new tab)
- Models overviewAnthropic (opens in a new tab)
- Choosing a modelAnthropic (opens in a new tab)
- Effort (docs)Anthropic (opens in a new tab)
- How Claude remembers your projectClaude Code (opens in a new tab)
- Memory in ChatGPTOpenAI (opens in a new tab)
- Projects in ChatGPTOpenAI (opens in a new tab)
- ChatGPT Work and CodexOpenAI (opens in a new tab)
- ModelsOpenAI (opens in a new tab)
- Reasoning modelsOpenAI (opens in a new tab)
- Open format for coding agentsAGENTS.md (opens in a new tab)
- Context rot: how input length affects LLMsChroma (opens in a new tab)
