Pathrule
Path-scoped memory, rules & skills for your AI coding agent
20 followers
Path-scoped memory, rules & skills for your AI coding agent
20 followers
AI coding context isn't a retrieval problem. It's a delivery problem. Pathrule delivers the right memories, rules and skills to Claude Code, Cursor, Codex and Windsurf before they start coding, scoped to the files they’re working on.
Hey Product Hunt 👋
Over the last year, one thing kept bothering us while using AI coding tools: every new session started with the same explanations. The same architecture. The same conventions. The same team rules.
We built Pathrule around a simple idea: AI coding context isn't a retrieval problem, it's a delivery problem.
Instead of hoping the AI finds the right information, Pathrule maps your team's memories, rules and skills to the parts of the codebase where they matter and delivers the right context before work begins. It works with Claude Code, Cursor, Codex and Windsurf through a single MCP endpoint, while keeping your source code on your machine.
We're still early and would genuinely love feedback from people building with AI every day. Happy to answer any questions.
Congratulations on the launch. How does this behave in a monorepo — does each subdirectory get its own context or do you still end up with one blob at the root?
mailX by mailwarm
Does Pathrule have a way to surface deprecated rules so the agent doesn’t follow outdated guidance?
@manal_essalek1 Yes, this is a core part of how Pathrule works. We run a server-side sweep of independent detectors across the workspace, and a few of them target exactly this "outdated guidance" problem:
- Zero-hit rules: rules that no longer match anything (usually a sign they've gone stale)
- Shipped milestones: guidance tied to work that's already done
- Expired dates: anything with a date that has passed
- Drifted dependency pins and dead-path refs: rules pointing at versions or paths that no longer exist
Each finding gets scored against specific metrics. Once the score is high enough, we start surfacing the suggestion to you, so you're not buried in low-signal noise and only the things actually worth your attention come up. Every suggestion names the node, which detector fired, and the exact signal that triggered it.
From there you have two paths: handle it manually, or continue with a prompt we prepare for you. That prompt hands your assistant (Claude Code / Cursor / Codex / Windsurf) the places it needs to check and the task to do; the assistant applies the patch, and you review the diff before it lands. Nothing changes silently.
When you do retire a rule, it's a soft delete: archived nodes drop out of routing instantly but stay restorable for 30 days. So the agent stops seeing the stale rule right away, and you keep an undo window.
Mailwarm
Can you set folder level rules, like extra strict guidelines just for migrations or authentication?
@othman_katim Yes. That's actually one of the main use cases.
You can scope memories, rules and skills to specific paths, folders or file patterns. For example, migrations can have their own safety rules, authentication code can require stricter review guidelines, and frontend components can follow an entirely different set of conventions.
When the AI starts working in those areas, Pathrule automatically injects the relevant context instead of relying on a single global instruction file.
@othman_katim One other thing: when the AI encounters a new rule, convention, or issue that should be remembered, it can proactively suggest saving it to Pathrule as a rule or memory. You can also ask it to create or update those entries directly through MCP.
Over time, the system doesn't just deliver context. It helps teams continuously build and maintain their shared context as they work.