Claude.md

• andrej-karpathy-skills

  multica-ai • Updated 2026 Jul 8 15:12

/plugin marketplace add forrestchang/andrej-karpathy-skillscl

• Map vs Territory
• Prompts/plans are the “map”
• The real codebase/constraints are the “territory”
• The gap between the two is where the unknowns live.

• In Claude Fable 5, the bottleneck is how well the user clarifies unknowns—not model capability
• If results are poor, it’s often not because the model is “dumb,” but because ambiguous parts weren’t properly identified/defined.

• Splits unknowns into four types
• Known knowns: things already explicitly stated
• Known unknowns: things you know you don’t know
• Unknown knowns: implicit preferences/criteria you’d recognize if you saw them, but haven’t articulated
• Unknown unknowns: risks/constraints you haven’t considered at all

• Recommended Claude Code workflow
• Before implementation, do a blind-spot pass to surface “unknown unknowns”
• Don’t jump straight into implementation—start with brainstorming/prototyping
• Have Claude interview you with questions to remove ambiguities that could change the architecture
• If you want specific behavior, provide reference code rather than just describing it
• Before implementation, ask for an implementation plan and review the most change-prone parts first (data model, types, UX)

• During implementation
• Have Claude record decisions/workarounds/edge cases in a scratch doc like implementation-notes.md.
• This makes it easier to postmortem failures or hand off to the next session.

• After implementation
• Write a pitch/explainer doc to convince reviewers
• Ask Claude to quiz you to ensure you truly understand the changes
• E.g., “No merge until I get the quiz right.”

Manage Claude's memory - Claude Code Docs

Learn how to manage Claude Code's memory across sessions with different memory locations and best practices.

https://code.claude.com/docs/en/memory

A field guide to Claude Fable 5: Finding your unknowns | Claude | Claude by Anthropic

Practical patterns for agentic coding with Claude Fable: how to find your unknowns before, during, and after implementation, from the team at Anthropic.

https://claude.com/blog/a-field-guide-to-claude-fable-finding-your-unknowns