Automate your project the way you need it.
Sandboxed scripts, skills and webhooks extend a project from the inside. Agents may write the code. An admin decides what it may do.
Sandboxed scripts, skills and webhooks extend a project from the inside. Agents may write the code. An admin decides what it may do.
Sandboxed TypeScript that talks to Manager without a token. It acts as its author or a member an admin picks, confined to the project.
A script starts at times of day or intervals, on chosen weekdays, or when a task is created or changes. A guard stops it looping on its own changes.
An agent can write a script and request the hosts it needs. Granting them, adding secrets and approving the code stay with an admin; a code edit clears the grant.
A project skill is a Markdown page of instructions that every agent loads when it needs it: definition of done, runbooks, vocabulary. Write it once, and every agent follows it.
Manager posts project events to your endpoint as they happen, filtered to the events you choose and signed with HMAC-SHA256. Every delivery's outcome is on record.
Your instance can require that someone reads a script's code before it may run, with a second reviewer if you like. The approval holds for exactly the code that was read.
A connected agent with the project admin role writes scripts and skills through MCP. Granting hosts, adding secrets and approving code stay with you.
Say to your agentManager: Write a script for PRJ that posts the tasks finished in the last 7 days to our Slack webhook, and request the host it needs.
It writes the TypeScript and requests the one host. Granting that host, adding the secret and setting when the script runs stay with an admin.
Say to your agentManager: Change the Slack script in PRJ to group the tasks by assignee. Touch nothing else.
It edits just that part of the code. The code changed, so the host grant lapses until an admin grants it again.
Say to your agentManager: Which scripts in PRJ are still waiting for a host grant?
The agent lists the project scripts with their state and names the ones whose requested hosts nobody has granted yet.
Say to your agentManager: Read the last three retrospectives in PRJ and turn the recurring review complaints into a code-review skill.
It writes a project skill from what keeps coming up. From then on, every agent working in PRJ can load it.
Everything you can automate with, at a glance.
Guides that put this feature to work in a real project.
Set WIP limits, pull work when there is room and let an agent watch the flow, so your team finishes more by starting less.
Run service requests, incidents, problems and changes for your servers and workplace IT, with priorities by impact and a CAB for risky changes.
Make an agent your project manager: it plans with you, watches the deadlines and takes the paperwork off your hands.
Everything you add runs under a real account and within its permissions. Code can come from you or from an agent; what it may do is decided by an admin in the interface, never by the code itself.
Get early access