Bug reports become tested fixes

Mention @Capy in Slack, with screenshots or files attached, or delegate a Linear issue to the Capy agent. On a pull request, comment @capy. Automations can also start a run from a Slack channel, a GitHub event such as a failed workflow run, or a webhook from your deploy pipeline.

Yes, through MCP servers. Ask in any thread, for example “Connect our Sentry MCP server”, and threads can call its tools while they work. A server’s tools act with the access of the account that connected it, and you can hide any tool from the model on the server’s page.

Capy traces the bug to root cause, ships the fix, and adds a regression test so it stays fixed. To make it a rule, write it in AGENTS.md, for example that every bug fix reproduces the bug first and opens a pull request with a test.

A new session uses the Linear team’s mapping, then the connection’s default project. If neither works, Capy posts a configuration error in Linear instead of guessing, and an existing session keeps its project.

Yes. Add an automation with a Slack trigger on the channel, a run_when sentence such as “the message reports a bug”, and single thread mode, so one standing thread keeps its context across reports. A fast model checks each message against run_when before a run starts, and you aren’t billed for that check.

The thread that owns it fixes failing CI, answers review feedback and resolves conflicts. A failing check wakes it at once, and you merge from the pull request page or on GitHub.

Type stop in the Slack thread, or use Linear’s stop signal on the agent session. Either one interrupts the current run and keeps the thread, so a later message starts it again.