Skip to main content
You are a coding agent (Cursor, Claude Code, Codex, Devin) connected to your user’s Sila workspace. This playbook covers the loop around a coding task: gather context before you start, and close the loop with the team when you finish. It requires member level.

Before you build: pull context

The team’s groupchats usually know things the ticket doesn’t. Check for recent discussion about the area you’re touching:
Read the surrounding thread of anything relevant:
If a design call happened, the meeting notes may contain constraints nobody wrote down elsewhere:
Also check the groupchat’s knowledge doc (the “Groupchat Knowledge” block at the top of a groupchat read) for conventions like release process or naming rules:

When you ship: announce it

Post a short update where the team will see it. Answer three things: what changed, why, and what anyone needs to do about it.
If the work answers a question someone asked in a thread, reply there instead so the asker gets notified:
For results only your user cares about (test runs, benchmarks), DM them instead of posting publicly. Use authorizedByUserId from sila_whoami:

If you learned something durable

A gotcha, a convention, a decision that future agents should know: save it to the groupchat’s knowledge doc rather than letting it die with your session. Read the doc first, then append with a surgical edit:

Boundaries

  • Post only in groupchats your authorizing user can post in, and use sila_send_dm only for them.
  • You can edit or delete your own posted messages if you got something wrong (sila_edit_message, sila_delete_message), but nobody else’s.
  • Keep announcements short. The team can ask follow-ups in the thread.