sentry-triage-fix | OutcomeCI workflow

Overview

When a new error is reported to Sentry, this workflow decides whether it's actionable, triages its severity and likely root cause, and if both the issue and source repo can be clearly identified, opens pull requests to fix it.

Before you start

You need to provide:

Slack and GitHub authentication credentials
A Slack channel where triage notifications should be posted
A Sentry webhook configured to send errors to this workflow's trigger endpoint

The workflow runs on incoming Sentry events, so no manual triggering is required.

What you get

A Slack notification is always posted summarizing the issue (severity, title, suspected root cause, and Sentry link). If the issue is actionable and the repository is identified, you can discuss and refine the proposed fix plan in a Slack thread; once agreed, the workflow creates a feature branch in the identified repository, makes the minimal code change, adds tests if applicable, and opens a pull request per affected repo with full context and merge order. If the issue lacks sufficient evidence or the repo cannot be identified, only the notification is posted and no fix is attempted. All pull requests remain as drafts if the implementation reveals the plan was unsafe or incomplete, leaving final review and merge decisions to your team.

outcome.ymlRevision 1

apiVersion: outcomeci.workflow/v1
name: sentry-triage-fix
trigger: webhook
secrets:
  slack: vault:publication/setup_value_1
  github: vault:publication/setup_value_2
apis:
  slack:
    uses: slack
    auth: secrets.slack
  github:
    uses: github
    auth: secrets.github
reasoning:
  default:
    runner: claude
  triage-model:
    model: anthropic/claude-sonnet-5-5
steps:
- triage:
    using: triage-model
    reason: triage.md
    from: trigger
    returns:
      decision: enum[actionable, insufficient_evidence, ignore]
      severity: enum[critical, high, medium, low]
      title: string
      summary: string
      sentry_url: string
      repo_hint: string
      repo_status: enum[identified, unknown]
      missing_evidence: string
      suspect_frames:
      - string
      proposed_plan:
        summary: null
        steps:
        - string
      route: enum[discuss, notify_only]
- notify:
    reason: notify.md
    with:
    - triage.decision
    - triage.route
    - triage.title
    - triage.severity
    - triage.summary
    - triage.sentry_url
    - triage.repo_hint
    - triage.missing_evidence
    - triage.suspect_frames
    - triage.proposed_plan
    can:
    - slack.post:
        channel: setup_value_3
    returns:
      posted: bool
- discuss:
    when: triage.route == "discuss"
    converse: slack.thread(notify.calls.slack.post)
    with: triage.proposed_plan
    until: converged
    max_turns: 12
    timeout: 1d
    returns:
    - plan
    - status
- fix:
    when: discuss.status == "converged"
    reason: fix.md
    with:
    - triage.title
    - triage.summary
    - triage.sentry_url
    - triage.repo_hint
    - triage.suspect_frames
    - discuss.plan
    can:
    - github.read
    - github.write
    policy: 'Write only to repositories the PAT already has access to, and only to
      those the agreed plan requires, at most 3 per run. In each, never push to or
      modify the default branch; work on a new branch named sentry-fix/<short-slug>
      (same slug in every repo). The only allowed outcomes are commits on those branches
      and one pull request per repo. No merges, no force-pushes, no repo settings
      changes, no edits to CI config, secrets, or workflow files.

      '
    returns:
      prs:
      - repo: string
        pr_url: string
        branch: string
      merge_order: string
      deviations_from_plan: string
- report:
    when: discuss.status == "converged"
    reason: Reply in the triage thread listing every PR (repo, link, branch) in merge
      order, plus any deviations from the agreed plan, in a few short lines. If no
      PR was opened, say why.
    with:
    - fix.prs
    - fix.merge_order
    - fix.deviations_from_plan
    can:
    - slack.thread:
        channel: setup_value_3
    returns:
      posted: bool