Changelog
What changed in Contexory
New features, improvements and fixes, newest first. Each entry says what you can do now that you couldn't before.
What changed in Contexory
Showing all 16 entries
- Improved
A home page that walks through the product
The home page now tells the story chapter by chapter, from the objection a team has to the release that answers it, with the trial length, prices and FAQ read from the product itself. On a phone, or with reduced motion, it reads top to bottom with nothing sticky.
See the home page → - New
Integrations: does it work with my agent?
A new integrations page and one page per host (Claude Code, OpenClaw, GitHub and the signed API) show what each one reads, the install command, where files land and what is not delivered there.
See integrations → - New
How releases work, on its own page
A new page walks through an agent's release: what a release pins, how a label moves it to production, how eval cases hold a failing release out of production, and how a rollback works.
How releases work → - New
Compare Contexory with a shared doc or a repo
A new page lays out where Contexory, a shared doc and a repo of prompt files each fit, including the cases where one of the other two is the better choice.
Compare the three → - Improved
Install in one command
The docs now open on a quickstart for Claude Code, OpenClaw and your own code: install, connect a token, then check that it worked.
Open the quickstart → - Fixed
Navigation on phones
The app now has a navigation menu on a phone, so you can reach every section and sign out from a small screen.
Start a trial and try it → - New
Runtime SDK
Your own code can fetch the release behind a label through the signed API, using a read-only runtime key that works only for the agents you pick. Responses are cached, with a stale fallback.
Read the SDK reference → - New
Claude Code plugin follows an agent's release
Set an agent in the plugin and Claude Code gets the skills that agent's release pins, not the whole library. With a runtime key it can also serve the release's skill text and tool descriptions, after checking the release's signature.
Install for Claude Code → - New
Releases and labels
An agent's instructions, skills and tool descriptions are cut into one numbered release that never changes. A label such as production points at a release: moving the label is the deploy, and moving it back is the rollback.
How releases work → - New
Eval cases hold the production label
When an agent has eval cases, they run against a release before the production label can move to it. If the release fails, production stays on the release it already points at. An owner or admin can override with a written reason.
- New
Name your own categories
Each organization now names its own skill categories in Settings instead of using a fixed set. Retiring a category moves its skills to another one you choose.
- Improved
Any text file in a skill's assets folder
Templates, CSV checklists and JSON stubs in a skill's assets folder now open in the editor and reach the agent as text. Binary files such as images and PDFs are unchanged.
- Improved
Several questions on one card
When the assistant needs more than one answer, it asks them together on one card and you reply once, instead of the first answer closing the rest.
- Improved
Claude Code plugin 2.0: skills no longer load twice
After an upgrade, skills no longer load twice, and sync now runs on its own and applies edits in the running session. This is a major version: the /contexory:mode, /contexory:reviewers and /contexory:disconnect commands are removed. A skill folder of your own with the same name is skipped with a warning, never overwritten.
Install for Claude Code → - Fixed
MCP tool names in allowed-tools
The skill validator no longer rejects MCP tool names such as mcp__github__create_issue in allowed-tools, and the Properties panel no longer turns the list form into a comma-separated value.
Read about allowed-tools → - New
Skill Doctor
A free SKILL.md checker that runs in your browser, with no sign-up. It points at the field most likely to stop an agent loading your skill, which is usually the description.
Check a skill now →