Reads its own logs, finds failures, opens fix PRs.
Set up a new agent named Self Improver to watch its own production behavior and fix what it gets wrong. Walk me through connecting GitHub, Datadog and Slack, then: read its own production logs daily, identify cases where it failed or produced a bad output, diagnose the root cause, and open a PR fixing the underlying prompt or code with the failing case included as a test. Post a Slack summary of failures found and PRs opened, and never merge without review. Learn what "good" output looks like from its own passing cases, then save it and run daily.
Every Skydive agent ships with these channels. Connect one in your workspace and it gets its own inbox or number. Not agent-specific.
Reusable capabilities the prompt sets up. Browse the open skills ecosystem at skills.sh.
See something off with Self Improver? Suggest an edit. Every agent is one markdown file in the repo.