When to use it
- The Sentry inbox is out of control
- Weekly error-budget review
- Before an on-call handover
What it looks like
Help me clean up unresolved issues in api-prod.
find_projectsfind_issuesget_issue_detailsfind_teams+2 more146 unresolved, 38 unassigned. Top 10 by users: 4 real bugs (assign to Payments and Search), 3 already fixed in 2.3.1 (resolve), 3 noise from a crawler user agent (ignore, and add an inbound filter). Apply these 10?
The SKILL.md file
name: sentry-issue-triageTriage a Sentry project's unresolved issues by ranking them by user impact, grouping likely duplicates, and proposing assign, resolve or ignore for each, with approval before changes. Use when the user asks to clean up, triage or prioritise Sentry issues or reduce Sentry noise.
Steps
- Resolve the project with
find_projects. List the backlog withfind_issues, for exampleis:unresolved is:unassigned, sorted by users affected. - Skim each of the top issues with
get_issue_details, enough to tell a real bug from noise such as browser extensions, bots, or network errors the user cannot act on. - Group issues that share a culprit or message pattern; they are often one bug.
- Propose one action per issue:
- Assign, to a team from
find_teamsor a person, when ownership is clear. - Resolve in the next release, when the fix already shipped.
- Ignore until it escalates, for known noise, with the reason.
- Investigate, for real impact with an unknown cause.
- Assign, to a team from
- Suggest inbound filters or a fix to the SDK configuration for repeated noise, with
search_docsfor the setting.
Apply
Show the table: issue, users, events, proposed action and reason. Make only the approved changes with update_issue, and wait for the user's yes before touching any issue.
Sentry tools it uses
find_projectsfind_issuesget_issue_detailsfind_teamssearch_docsupdate_issueAsks first
Tool list from Sentry MCP documentation.
FAQ
Will it ignore real errors?
It only proposes ignoring issues it can explain as noise, gives the reason, and changes nothing until you approve.
How does it rank issues?
By users affected first, then event volume and how recent they are.