Cloudskill
Govern the AI skills your team depends on
Start new thread

Cloudskill - Govern the AI skills your team depends on

by•
Bring order to the AI agent skills your team depends on. Cloudskill turns scattered skill files into a managed catalogue in seconds, complete with version control, per-person access policies, and a full audit log. When skills are created or updated, every change is reviewed, approved, and tracked.

Add a comment

Replies

Best
Maker
📌
Hey Product Hunt 👋 I'm Tom, maker of Cloudskill, and I'm super excited to share what I've been building! Why I built this: AI agents like Claude, Cursor, and Copilot are only as good as the "skills" they run - the instruction files that tell them how your team actually works. I've watched those skills pile up across organisations, unmanaged and unchecked. So I built Cloudskill. The problem Cloudskill addresses: Teams are creating and pulling in more AI skills every week, and nobody's in control of them. Snyk's 2026 ToxicSkills research found 36% of AI skills on a public marketplace carried prompt-injection risks - and Anthropic's own docs warn that conflicting skills quietly degrade your agents. The skills your team now depends on are scattered, unreviewed, and impossible to govern. What Cloudskill is: A platform that turns the AI skills your team builds into managed software - reviewed, versioned, access-controlled, and audited - so the skills you depend on stay safe and consistent. How it works: 🛡️ Build a catalogue in seconds. Write a skill directly, or upload your existing skill files. Built-in authoring guides keep each one clean and conflict-free. Every edit is versioned, and you can roll back to any version with one click. 🛡️ Distribute without the chaos. Your people see only the skills they're entitled to, and download them in one click. No links to chase, no copy-paste, no stale versions floating around. 🛡️ Control who gets what. Assign access per person with a simple policy matrix. Admins see everything; members see only what's approved for them. 🛡️ Approval built in. Anyone can submit a skill; admins and nominated stakeholders review and approve before it ships. The best skills come from the people doing the work - Cloudskill just keeps a human in the loop. 🛡️ Audit everything. Every change is recorded in a searchable audit log, append-only and ready the moment compliance asks. Over time, your catalogue becomes a living record of how your organisation actually works. 🛡️ Skill files are automatically packed in folders with references, assets, evals and scripts folders along with a design.md and manifest, so your team can add as much context as they need. 🛡️ Works with the agents you already use. Claude, Cursor, Codex, Gemini CLI, GitHub Copilot - Cloudskill is the management layer that sits across all of them. The goal with Cloudskill is simple: treat your team's AI skills like the software they've become. No matter the size or your team, the skills they rely on should be governed, safe, and consistent - not a free-for-all. Try it free at cloudskill.com. I'll be in the comments all day - so ask me anything 🙏

 I'm thinking about the workflow, if team members can add anything to what is approved, how are versions controlled? Or do you only assign certain team members to be contributors?

Maker

 great question, thank you! Once a skill is approved any changes are a new skill that goes through the same process. So every change creates a new skill rather than modifying the existing skill.

This is a sharp problem to take seriously. Skills are starting to look less like docs and more like production dependencies.

One question I would be curious about: do you plan to connect skill approval/versioning with runtime receipts later? For example, when an agent takes a customer-visible or state-changing action, being able to answer which approved skill version shaped the run, which tools it used, and what changed would make the governance story much stronger.

Maker

 That's a great observation and question, thank you!

To answer your question - yes, I am currently developing in that direction and should have an update soon.

  Nice. That is probably the useful jump: not just a skill was approved, but this run used v17, called these tools, changed this state, and left a receipt the team can inspect later.

Maker

 Yep agreed! Any other feature thoughts most welcome!

 I would probably think about drift next: when a skill changes, show which agents or workflows are now affected, and maybe block high-risk runs until the new version is approved for that use case.