Recently, I've had agents reach for the wrong skill or tool. This tends to happen in the middle of longer reasoning chains, which frustratingly means lots of chatter polluting the context before a correction can be issued.

To chase these issues down, I built `drskill`: a CLI tool that checks for conflicts and security issues among your agent's Skill and MCP loadout.

In addition to spotting conflicts that might trip up your agents, security alerts were added as well. Did an MCP provider silently add a tool due to a supply chain attack? Running `drskill` on bootup or during CI will flag those changes.

`drskill` works without access to an LLM, though connecting it to one will yield better conflict reports and allows for suggesting alternate, non-clashing descriptors.

A ledger is maintained globally and locally, tracking acknowledgements and preferences.

This is the type of tool that heavily benefits from contributions (your loadout is very different than mine!), so please file issues or submit PRs if you're interested.