On mention
Summarize a pull request
Paste a PR link in Slack and get a plain summary and risk list.
CI failure digestSee the pattern, not 40 red runs.
Each morning it reads failed runs from the last day. It groups them by cause, marks the flaky ones and says who should look first.
Your first signup includes $50 in free credits.
Weekdays · 8:30
Digest the failing builds
- Reads
- Failed GitHub Actions runs
- Grows
- Failures grouped by cause
- Lands in
- Slack
Ask for it once, in plain words.
No flowchart to draw. Describe the job in chat. TulipFarm sets up the agent and its workflow, and shows you both before anything runs.
You
Every weekday at 8:30, read failed CI runs from the last day. Group them by cause, mark flaky tests, and say who owns each group. Show me before posting to #builds.
TulipFarm
I made an agent called CI digest and a weekday 8:30 schedule. It reads GitHub Actions. Posting to Slack asks you first. Activate it?
ActivateEdit first
One morning, start to finish.
Every run is saved as an agent task you can open, check and rerun.
- Schedule
Weekdays at 8:30
Run startedStarted - Reads
Failed runs and their logs in GitHub
14 failed runs - Groups
Failures by likely cause
3 groups - Checks
Has the same test failed and then passed?
1 flaky - Asks first
Post the digest to the builds channel
Approved - Posts
Message in Slack
Delivered 8:34Completed
The digest at 8:34.
Three groups instead of fourteen alerts. Each group has an owner to look first.
#buildsCI digest APP · 8:34
CI · 14 failed runs, 3 causes
A broken test · 8 runs
- Started after one merge yesterday.
- Look first: the author of that change.
Flaky test · 4 runs
- Passed on retry each time.
- Look first: the test owner.
Network timeout · 2 runs
- Both on the same runner.
What gets planted.
Three parts you own and can change at any time.
- Agent
CI digest
Reads failures and finds the pattern.
- Skill
Cause rules
How you tell a real failure from a flaky one.
- Workflow
Weekdays 8:30
Runs before the team starts.
Uses
- GitHub
- Slack
Add if you use it
Jira
MCPLinear
Google Sheets
It reads quietly. It asks before it acts.
Reads without asking
- Failed workflow runs
- Run logs
- The change that started each failure
Asks you first
- Posting to Slack
- Opening a ticket for a flaky test
It uses your own connections, so it never sees more than you can.
Grow it your way.
Ticket the flaky ones
Draft a Jira ticket for each flaky test, after you approve.
Main branch only
Skip failures on feature branches and watch only main.
Twice a day
Add an afternoon run for fast-moving weeks.
Questions, answered.
Does it re-run builds?
No. It reads runs and logs. It does not start, retry or cancel any run.
Which CI does it read?
GitHub Actions through the built-in GitHub connection. Other CI tools need a Custom MCP connection.
How does it spot a flaky test?
By a pass after a fail on the same code, as set by your skill. It is a guess you can check in the logs.
Fridays · 16:00
Hand off on-call cleanly
Writes the handoff note: open issues, noisy alerts, what to watch.
Weekdays · 10:00
Chase stale pull requests
Finds reviews stuck for days and nudges the right person.
Red builds, made readable.
Public beta, open to everyone. Your first signup includes $50 in free credits.