New
Swan skills now follow the open skill convention — A skill is a folder with a SKILL.md at its root and whatever supporting material it needs beside it: reference documents, scripts, examples. That's the same structure the wider agent ecosystem has settled on, which is the whole point — a skill is no longer a Swan-specific object.
So you can bring in any skill — A skill you already use somewhere else, or one you found published online, comes into Swan as it is. Nothing to rewrite, convert, or paste into a box. And a skill you build in Swan isn't trapped there either.
Swan can search and install from gtmskills.com — GTM Skills is an open-source public library of go-to-market skills that we helped build. Swan searches it straight from chat: ask for help with something your current skills don't cover, and it browses the catalog, picks a skill that fits, and installs it into your organization. Once installed it behaves exactly like a skill you wrote yourself, and you can edit it.
Improvements
Salesforce campaign members — A new tool for reading and managing Salesforce campaign members.
Connected accounts tidied up — Email and LinkedIn account settings moved into overflow menus so the list reads cleanly, while blocked-account recovery stays visible. Healthy LinkedIn accounts can now be reconnected without disconnecting first.
Personas belong to one target market — Each persona is now tied to exactly one target market. Personas that were previously linked to several were split into one per market, so targeting stays unambiguous.
Triggers show who created them — Creator avatars make it clear who set up an automation.
Fixes
A pasted LinkedIn profile URL was ignored — Dropping a single LinkedIn profile link into a request sometimes did nothing, because the URL was being swallowed by the text around it. It's now picked up.
Long email threads were cut off mid-message — When Swan read a long thread, the end of it could silently disappear, so replies missed context that was actually there.
Cancelling a chat showed an error — Stopping Swan mid-response, or stopping something already stopping, could surface an error even though the cancel had worked.