Skills might improve Claude's adherence to procedures compared with claude.md
The author experimented with Claude Code, using a file called CLAUDE.md to list rules the AI should follow. While the AI read the file, the author observed that compliance was inconsistent—sometimes the AI skipped steps or produced slightly different results despite identical instructions. Adding more lines to the file often made it longer and harder for the AI to process, leading the author to…
Key points
- Claude reads CLAUDE.md but does not follow rules consistently.
- Separating repeated tasks into skills made the AI follow procedures more reliably.
- The author notes that longer CLAUDE.md files become harder for the AI to process.
After seeing a post on X suggesting that long CLAUDE.md or AGENTS.md files are problematic, the author tried breaking repetitive procedures into separate "skills". Skills bundle a specific procedure and are referenced from CLAUDE.md, allowing the AI to receive only the necessary instructions for each task. The author reports that this approach seems to make the AI follow the same procedure more reliably, though the claim remains anecdotal. The piece also notes contrasting opinions that future AI improvements might eliminate the need for such detailed instructions, but the author argues that exact control still requires some level of explicit guidance.
If you want AI to follow procedures, skills might be better than CLAUDE.md
note.com · 24 October 2026
Loading the full article…
This text was published by note.com and written by ふくゆー. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Agents & Tools
All →- Claude Code agents hallucinate non-existent blockers in multi-agent workflows · 1 src
- AI novice builds game, custom skills, and daily news automation with Claude in one day · 1 src
- OpenAI's Dots reaches 1 million users in a week as Amazon blocks access · 1 src
- Claude Code documentation defines harness as layer around model · 1 src
- OpenAI adds scheduled automations to Agents API for recurring tasks · 1 src
Comments
via GitHub Discussions