Summary
A ten-rule CLAUDE.md document attributed to Andrej Karpathy (recently joined Anthropic pre-training) began circulating on X on June 28, 2026, adding six rules to the popular four-rule community template created by Forrest Chang (200K+ GitHub stars). New rules cover verification (test-before-fix), extended goal-driven execution, debugging discipline, dependency caution, communication honesty, and named failure modes (Kitchen Sink, Wrong Abstraction, Optimistic Path, Runaway Refactor). The additions target loop-level agent self-monitoring rather than turn-by-turn coding — aligning with Boris Cherny’s “loop engineering” paradigm where developers design autonomous evaluation loops instead of writing prompts. Authenticity unconfirmed; Karpathy has not commented.
Source Analysis
- Primary source: TechTimes, June 28, 2026
- Community baseline: forrestchang/andrej-karpathy-skills (4 rules, January 2026)
- New focus: Agent self-monitoring at loop scale; stop conditions for autonomous workflows
- Technical context: CLAUDE.md injected as project context in Claude Code; /goal command (v2.1.139) uses separate verifier model
- Security note: Malicious CLAUDE.md in cloned repos documented (Adversa AI, LayerX); use official repos only
- Industry trend: Shift from prompting to loop design (Cherny, Steinberger, Osmani, June 2026)
PreScreening Notes
Score: 5/10 | Priority: medium
Borderline pass. Relevant to software/AI developer audience (loop engineering, Claude Code workflows) and timely (June 28 circulation). However, authenticity is unconfirmed — Karpathy has not verified attribution — making this closer to community trend coverage than verified news. TechTimes is secondary source. Passed per “when in doubt, pass through” given strong practitioner interest in agentic coding practices. Evaluation stage should verify attribution before publication.
Evaluation Report
News Value Assessment
| Dimension | Rating | Notes |
|---|---|---|
| Timeliness | ★★★★☆ | Circulated June 28 on X |
| Impact | ★★☆☆☆ | Community interest only; no verified product/policy change |
| Prominence | ★★☆☆☆ | Karpathy name drives virality but authorship unconfirmed |
| Proximity | ★★★★☆ | Loop engineering relevant to developer audience |
| Novelty | ★★☆☆☆ | Unverified extension of existing community template |
Audience Fit
- Primary audience: Software developers using Claude Code
- Actionability: Content may be useful but provenance is unknown
- Relevance: Loop engineering trend is real; this specific document is not verified news
Risk & Ethics Assessment
- Verification: FAILED — Karpathy has not confirmed authorship
Warning
Independent investigation (Delv, AlphaSignal, MCP.Directory) confirms Karpathy did not write the original four-rule CLAUDE.md (Forrest Chang did). The ten-rule document’s provenance is entirely unverified. Publishing as “Karpathy’s CLAUDE.md” would spread misattribution.
- Misinformation risk: High — attribution is the core news hook
- Fact-checking: Cannot proceed without primary source confirmation
- Ethics: Publishing unverified attribution harms editorial credibility
Publication Strategy
- Decision: REJECT
- Alternative: A future item on “loop engineering” trend could cite confirmed sources (Boris Cherny, Forrest Chang’s repo) without Karpathy attribution claims
Related Wiki Topics
andrej-karpathy — note: Karpathy’s actual project CLAUDE.md files are mundane documentation, not behavioral rule sets
Suggested Angle
N/A — item rejected. If loop engineering is covered separately, angle should be: “Topluluk CLAUDE.md şablonları ve loop engineering paradigması” with attribution to Forrest Chang and Boris Cherny, explicitly noting Karpathy did not author the viral templates.