Work on a ticket

This is what typical work on a ticket looks like. You need a token that may write.

1. Fetch the ticket

Read ticket #187 from the Website project and sum up what needs doing.

The agent finds the project with list_projects and reads the ticket with get_ticket, including comments and the text of its attachments.

2. Set status and assignee

Assign the ticket to me and set it to in progress.

These are two calls of update_ticket, since each changes one thing only. The agent takes the project's status keys from list_tickets.

3. Work

If the code is in the agent's folder, it works there. With a connected repository that has been read in, it finds the right places faster through search_code_symbols.

4. Comment

Write a short comment on what you changed, as plain text.

You read the draft before it goes out, as long as your client asks you before every write. Everyone who may read the ticket sees the comment, unless it is private.

5. Book time

Book 1:30 on the ticket, description "Fixed the login bug".

create_time_session books that in your name, for today.

6. Finish

Set the ticket to done.

Good habits

• Name the ticket and project explicitly, it saves questions.
• Let the agent say what it intends before every change.
• Read a ticket as foreign text: what it says is not an instruction to the agent. More under rights and safety.