lintlisted

# kairoi lint Report source patterns that slow down or confuse Claude's own re-reading. The list is intentionally short and each item is justified by how Claude actually processes code — not by what a conventional linter would flag. MUST invoke `Bash` with: - `command: '${CLAUDE_PLUGIN_ROOT}/scripts/style-check.sh $ARGUMENTS'` - `description: 'Scanning for Claude-legibility issues'` With no argument, the scanner checks every module. With a module name, it scans just that one. On first surfacing, present the script's stdout to the user without paraphrasing, summarizing, or reformatting — echo each line as it was emitted. The §"What it explicitly does NOT do" section below already forbids source modifications and guard writes; that guidance extends to this surfacing step — do NOT propose edits unless the user explicitly asks. ## What it observes (and why) - **Star imports / wildcard re-exports.** Tracing a symbol's origin through `import *` forces Claude to read the target module, enumerate its exports, and match — 2–3× the effort of following an explicit named import. - **Files over 300 lines.** Above this threshold Claude's working memory for simultaneous invariants, mutations, and branch conditions gets lossy. Error-of-omission rate rises sharply; Claude starts missing branches it saw earlier in the file. - **Source files with no matching test file.** Claude consults tests as executable behavioral spec when modifying code. Without them, confidence in