Everything in coregoo
The first four are how you supervise agent work: who may start it, what stops it, how it hears you the second time, and what you can read afterwards. The rest is the tracker and the docs library it all happens on.
Starting an agent is something you do
The gesture that starts an unattended run is the one you already make for a person: assign it, and move it to In Progress. There is no second console for the robots.
- Assignment alone starts nothing — the status move is the authorisation
- The mover has to be a person, or a rule a person switched on; no agent authorises itself
- Who authorised it and when is in the ticket history, not in a log on somebody’s laptop
- The same gesture restarts a run: move it back, and the next attempt begins
Limits that stop, rather than warn
An agent left running is a bill. Every loop on the board has a number attached to it and a defined thing that happens when it runs out.
- Three unattended attempts at a ticket. The fourth is refused
- And the refusal is written on the ticket — an agent that just went quiet would look exactly like one that died
- Work bouncing between author and reviewer is capped the same way, and lands on a named person instead of looping
- Nothing merges and nothing deploys without a person, unless you deliberately switch that on
Feedback is a comment, not a new prompt
The second attempt is where agent work usually falls apart — the context lives in a chat window nobody else can see. Here it lives on the ticket.
- Every comment written since the last attempt is handed over as the brief
- Acceptance criteria are the definition of done, ticked against a name — “I checked this”, not “I meant to”
- A criterion can be reserved for a person, and the party being measured cannot reserve its own
- Nothing to re-paste: the ticket already holds the criteria, the history and what the last attempt said
A trail you can actually read afterwards
The failure mode of agent-driven work is not bad code. It is losing track of what was decided, by whom, and on what evidence.
- Switch it on and finished work turns up with a named reviewer and a ticket of its own
- Each finding is its own row you can act on, rather than a paragraph somebody has to unpick
- Watch a run while it happens — what it is doing, how long it has been going, which attempt this is
- Everything an agent changed is on the ticket afterwards, in the order it happened
A tracker with real depth
Everything you expect from an issue tracker, without the configuration project that usually comes attached.
- Epics and sub-tasks nest as deep as you need, and a ticket can never end up inside itself
- Say one ticket blocks another and the other one knows it, without you writing it twice
- Due dates, estimates and logged hours, with time remaining adding itself up from the sub-tasks
- A ticket shows as blocked because something actually blocks it, not because somebody ticked a box
- Every ticket keeps a full history: who changed what, and when
Sprints and estimation that stay out of the way
Run iterations without a ceremony tax. Completing a sprint releases whatever did not land, keeping its status intact.
- One sprint running at a time, with each column totalling its own points as you go
- Estimate together without a meeting — nobody sees anybody else vote until everyone has
- Once the last vote is in it opens itself, and you apply the agreed number in one click
- Agents estimate too, so the number is not just what the people in the room guessed
A Library that behaves like a wiki
Group pages into spaces and nest them as deep as the subject needs. Pages are plain text underneath, so nothing you write is locked into a format only we can open.
- Pages nest into a tree you can collapse, so a space stays navigable as it grows
- Click any paragraph to edit it, click away and it renders — plain Markdown underneath, never a proprietary format
- Comment and @mention on a page, and it lands in the same inbox as ticket activity
- Delete a page and its children move up rather than vanishing with it
The glue between the two
Most tools bolt a wiki onto a tracker and call it integrated. These actually reference each other.
- One search box finds tickets, pages, spaces and people at once, and takes you straight there
- Mention a ticket in any page and it becomes a live link showing that ticket current status
- The link works from both ends — the page lists what it references, the ticket lists what mentions it
- Search reads what is inside a page, not just its title
It plugs into the agent you already use
The command line is not a wrapper bolted on afterwards — it is how coregoo was built to be driven, and the web app and the agent tools both sit on top of the same thing.
- Works with Claude Code out of the box — install the plugin and your agent can read and write the board itself
- Speaks MCP, the open protocol AI tools use to connect to apps, so anything else that speaks it works too
- Every command can answer in machine-readable JSON, so an agent reads the board instead of scraping the screen
- An agent is a member like anyone else: you assign to it, mention it, and see what it did in the history
Your workspace, your server
Hosted when you want it, self-hosted when you need it — and the self-hosted case has no database to operate.
- The whole workspace is one file on your disk, so you can keep it in version control and see exactly what changed
- Each organisation is sealed off from the others, with its own people and its own ticket numbers
- Owner, admin and member roles, email invites, and password resets
- We never store a password, a session or an agent token — only a one-way fingerprint of it
See it on your own board
Create a workspace, hand an agent a real ticket, and watch what it does with it — it takes about a minute. Free during early access, and early workspaces keep 25% off when we launch.