The agent does not obey it.
In a long session, the rule is one line of many. No check fails, thus you do not see the error.
stelint 0.3 Rust and Markdown
Agents write more comments and docs than you can read. They give one object three names. They use words that have no clear meaning. They write two steps that do not agree.
The next agent reads this text and obeys it.
stelint checks each comment and Markdown file against the terms of your project. A bad sentence fails the check, and the agent writes it again.
stelint finds errors in each one. The dictionary of these examples has the terms importer,
batch and archive folder.
An agent wrote these four files. Select a file. Put the pointer on an error to see its help text.
CLAUDE.md is a request. A failed check is not.In a long session, the rule is one line of many. No check fails, thus you do not see the error.
The same text can pass one time and fail the next time. Each check uses tokens and time.
An agent writes a hundred lines of comments in one minute. You read the code, not the comments.
A type checker finds errors in code. stelint finds errors in text. It has no LLM and no network. The same text always gets the same diagnostics, and each diagnostic tells the agent how to correct the text.
The project dictionary applies to each file. A module dictionary applies only to its module, and Claude Code reads it when it opens a file there.
stelint does not include the STE word list. Without it, stelint does not check the words.