| Primary scope | Claude Code project files: agent definitions, slash commands, settings, and documentation files such as CLAUDE.md and AGENTS.md. | The repo harness: CLAUDE.md, AGENTS.md, .cursor/rules, GitHub Copilot instructions, CI, hooks, .gitignore, and continuity files. |
| Check model | Schema and structure validation against Claude Code file conventions, with support for custom Zod schema extensions. | 33 evidence-backed harness checks covering findability, instruction quality, workability, safety, and continuity. |
| Reasoning citation | Rules are tied to Claude Code specifications and project-file conventions; custom schemas can add local policy. | Each check is written against a cited primary source, such as prompt-history data, runtime behavior, or published research. |
| Language coverage | A Node CLI for Claude Code configuration and Markdown-style project files, not a general source-code language linter. | Language-agnostic repo audit because it inspects the files agents read and the automation they trigger. |
| Output format | Console output plus JSON and Markdown report options for CI or review workflows. | Score, fix plan, CLI output, and report-oriented workflows designed to show what changed in the harness. |
| Integration | CLI-first, with CI-friendly options such as quiet mode, report files, and configurable failure thresholds. | CLI and CI oriented, with a Claude Code plugin flow for interactive scan-fix-report work. |
| Fix behavior | Best treated as validation and reporting unless your local workflow layers fixes on top. | Audits, scores, and can guide or apply targeted fixes for harness issues it understands. |
| License | MIT, per the upstream repository badge and license metadata. | MIT, matching the AgentLint repository and site footer. |