Run a security scan on your repo to find hidden instructions in AI assistant config files.
Add rulesentry to your CI pipeline to block malicious commits automatically.
Preview safe fixes and strip invisible characters from your config files.
Generate JSON or SARIF security reports for GitHub's Security tab.
| mohamedzhioua/rulesentry | 4ssh1/mine-sweeper | achawla19/intuitcode-extension | |
|---|---|---|---|
| Stars | 2 | 2 | 2 |
| Language | TypeScript | TypeScript | TypeScript |
| Last pushed | — | 2026-01-31 | — |
| Maintenance | — | Maintained | — |
| Setup difficulty | easy | moderate | moderate |
| Complexity | 2/5 | 2/5 | 4/5 |
| Audience | developer | vibe coder | developer |
Figures from each repo's GitHub metadata at analysis time.
No installation or configuration needed, run directly with npx rulesentry scan.
rulesentry is a tool that catches hidden instructions smuggled into configuration files used by AI coding assistants. These are files like CLAUDE.md, .cursorrules, or copilot-instructions.md that tell an AI agent how to behave. The problem it addresses is that a human reviewer sees one thing when reading these files, but the AI reads the raw underlying data, and attackers can exploit that gap by embedding invisible text. The attacks work through several tricks. Invisible Unicode characters can hide entire commands inside a line that looks innocent to a person. Bidirectional override characters, known as Trojan Source, reorder text so the line you read differs from what the agent reads. Some Claude Code skill files can run shell commands and insert the output before the model sees the file, which means a malicious instruction can execute before anyone notices. Homoglyphs, like a Cyrillic letter that looks like a Latin one, can spoof an allow-list entry. rulesentry detects all of these and shows you exactly what the agent reads versus what you see, with precise byte offsets. You can run it with no installation or configuration using npx rulesentry scan, and it will automatically find and check the relevant agent files in your repository. A clean scan prints one line and exits successfully. If it finds problems, it prints a reveal diff and exits with an error code, which makes it suitable for automated use in continuous integration. You can also run rulesentry fix to preview safe normalization, or add --write to strip invisible characters in place. It only removes characters where the fix is unambiguous, leaving homoglyphs and executable strings flagged for human review. The tool integrates as a GitHub Action or a pre-commit hook, so scans run automatically on pushes and pull requests. It can output results in JSON or SARIF format for GitHub's Security tab. Every scanned file gets a receipt with hashes representing what a human perceives and what the agent actually reads, so code review has a verifiable record when those two values differ. The full README is longer than what was shown.
A scanner that finds hidden malicious instructions embedded in AI coding assistant configuration files like CLAUDE.md or .cursorrules. It detects invisible text, text-reordering tricks, and lookalike characters.
Mainly TypeScript. The stack also includes TypeScript, npx.
No license information was provided in the explanation.
Setup difficulty is rated easy, with roughly 5min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Don't trust strangers blindly. Verify against the repo.