Sign InOpen Brain
GitHubGitHub RepoNeeds Review

multica-ai/andrej-karpathy-skills

This CLAUDE.md turns common agent failure modes into four operating rules: surface uncertainty, minimize code, avoid unrelated edits, and define verifiable completion criteria.

GitHub
Open Source Open MarkdownOpen JSON
Source Summary

The repository packages **four principles** into one CLAUDE.md: expose assumptions and tradeoffs, write the minimum code, limit edits to the request, and drive work with verifiable goals. A Cursor rule carries the same guidance.

Practical Implication

Merge the rules with project-specific instructions, then add concrete tests and acceptance criteria. The strongest operational pattern is **tests-first verification**, giving an agent a target it can check rather than an underspecified command.

Agent-Ready Context
The repository packages **four principles** into one CLAUDE.md: expose assumptions and tradeoffs, write the minimum code, limit edits to the request, and drive work with verifiable goals. A Cursor rule carries the same guidance.

Merge the rules with project-specific instructions, then add concrete tests and acceptance criteria. The strongest operational pattern is **tests-first verification**, giving an agent a target it can check rather than an underspecified command.

These rules intentionally favor caution and may add friction to trivial fixes. They are behavioral instructions, not evidence that every model or task will improve, so judge them by diff scope, clarification timing, and test results.
Connected Context · Feed7 Judgment

This turns minimal, scoped, assumption-aware coding into a small reusable instruction layer, with tests and acceptance criteria supplying the verifiable target. It reinforces evidence-led harness design, but also narrows the claim: cautious rules are defaults to evaluate by diff size, clarification behavior, and test outcomes, not proof that every model or trivial task benefits.

Do AI Agents Know When a Task Is Simple? Toward Complexity-Aware Reasoning and ExecutionE3 directly reinforces the minimum-code principle by formalizing scope estimation, a minimum viable attempt, and expansion only after verification fails.Evolution of agentic surfaces — Gagan Bhat & Isabella Kai He, AnthropicAnthropic’s account warns that behavioral scaffolding can become harmful after a model change, so these rules should remain replaceable and be reevaluated during upgrades.How Forward Deployed Engineering is done at Factory — Eno ReyesFactory’s use of bounded tasks, validators, and explicit completion criteria provides operational support for the repository’s emphasis on limited edits and verifiable goals.How to build an AI-Native Health Company — Dan Feng, Maven ClinicMaven’s compact specs, reviewable PRs, and repeated tests translate the same principles into team practices while confirming that verification and human review remain necessary.
Context Map
agentcoding#harness-engineering#agent-reliability#context-engineering
Uncertainty
These rules intentionally favor caution and may add friction to trivial fixes. They are behavioral instructions, not evidence that every model or task will improve, so judge them by diff scope, clarification timing, and test results.