Agent Rules Linter
A single offline HTML tool that scores a CLAUDE.md, .cursor/rules, or AGENTS.md file against 12 checks tied to the concrete failure modes that make AI coding agents unreliable — for developers who already have a rules file and want to know if it's actually doing its job.
Description
Paste your rules file in and get a weighted score out of 100, a pass/fail breakdown across 12 deterministic checks (scope boundaries, a verification gate before 'done', read-before-edit, assumption-surfacing, handoff discipline, an independent-review gate, stack specificity, security guidance, and more), and a plain-English fix for every gap, with a Markdown export. It runs entirely in the browser — no login, no backend, no AI calls, nothing leaves your machine. Built as a linter for a config file the same way eslint lints code: it doesn't know if your agent actually follows the rules, but it catches the load-bearing instructions that are missing from the document itself.
What's included
- index.html — the complete tool, one self-contained file, no build step or dependencies
- README.md — usage guide, the full 12-check reference table, and honest limits
Delivery
Instant download after Stripe checkout. File: nightforge-agent-rules-linter.zip