Auto-triage is a special type of automation where a persistent Devin monitors a Slack channel and automatically investigates bugs, regressions, and incidents as they come in. Instead of manually assigning someone to look at every report, Devin watches the channel 24/7, decides what needs attention, and spawns focused sub-sessions to diagnose each issue. Auto-triage has long-term memory — it accumulates context over time and learns from you via its scratchpad. It intelligently deduplicates repeated reports and automatically routes issues to the right code owner.Documentation Index
Fetch the complete documentation index at: https://docs.devin.ai/llms.txt
Use this file to discover all available pages before exploring further.
How it works
A long-running parent Devin monitors your Slack channel and listens to every new message. It filters out noise, detects duplicates, and spawns focused child sub-devins to investigate actionable bugs. Each child reads the relevant code, traces the root cause, posts a diagnosis in the Slack thread, and tags the right code owner.Setting up auto-triage
- Invite Devin to the Slack channel you want monitored (e.g.
#bugs,#incidents) - Go to Automations and create a new automation using the Triage bug reports on Slack template
- Select the channel and save
Your personal Slack account must be connected in Settings > Integrations > Slack.
Customizing behavior
Setup prompt
The setup prompt lets you customize how the triage Devin behaves. This is injected into the agent’s instructions and influences how it handles incoming messages. Examples:- “Focus on regressions in the payments service. For frontend bugs, tag the UI team.”
- “Only investigate issues that include error logs or stack traces. Ask for more details if the report is vague.”
- “When you find a root cause, always include a link to the relevant source file.”
MCP integrations
Connect MCP integrations to give the triage Devin access to external tools. For example:- Datadog MCP — Pull metrics, logs, and traces to correlate issues with runtime behavior
- Sentry MCP — Look up error details, stack traces, and affected users
- Linear MCP — Check for related tickets or create new ones
The scratchpad
The parent monitor and all child sub-devins share a persistent scratchpad. This is used to:- Track recently triaged items (channel ID, message timestamp, reporter)
- Maintain a routing table mapping code areas to owners
- Record duplicate items so future reports can be linked to existing threads
- Store context that persists across session restarts
Security
Since incoming Slack messages can contain untrusted user input (e.g. from support tickets), consider enabling a network policy to restrict outbound access for your auto-triage automation.Limits
Like all automations, auto-triage supports ACU limits and invocation limits to control resource usage. Each child sub-devin spawned by the parent counts as a session against your ACU budget.Tips for effective auto-triage
- Start with a focused channel. Pick a channel dedicated to bug reports rather than a general engineering channel. Less noise means better signal.
- Set clear expectations in the setup prompt. Tell Devin what kinds of issues to prioritize and what to ignore.
- Connect relevant MCP integrations. Datadog, Sentry, and other observability tools dramatically improve triage quality by giving Devin access to runtime data.
- Correct routing mistakes. When Devin tags the wrong person, reply in the thread with a correction. The parent updates its routing table and gets it right next time.
