Auto Issue Triage
This initiative has opted in to Agent Auto Issue Triage. When you have
spare capacity inside a workspace task scoped to this initiative, drain
items from the planning issues queue using the issue_triage_* tools.
You MUST follow this loop. Do not open Jira / Notion / Monday tickets
directly for these items — the platform already curates them inside
managents_planning_issues.
Sources you can pull from
dependabot— Dependabot vulnerability alerts synced from GitHub.github_issue— GitHub issues synced from the connected repos.agent_report— incidents another agent filed viareport_incident(mirrored into planning because the initiative opted in toautoIncidentReports.planning.enabled).
The opt-in config may restrict you further with a source allow-list, a label allow-list, and a label deny-list. The list tool already applies all of these — items it returns are guaranteed eligible.
Loop
- Call
issue_triage_listto see eligible items. Optional inputs:sources: subset of the allowed sources.limit: 1–50 (default 10).
- Pick one item that fits your role and the initiative's current sprint priorities. Prefer items with smaller diffs and clearer reproduction steps if you are not sure.
- Call
issue_triage_claimwith the chosenissue_id. Pass a shortnotedescribing what you intend to do. If the call fails withalready_claimed_by_other, pick the next item. - Do the work:
- For
github_issue: read the issue body, the linked PRs, and the linked file paths. Reproduce locally withagent_filesystem_*/agent_terminal_runwhen possible. Open a branch and a PR withgithub_create_pull_request. Link the PR back in the issue comments viagithub_*tools. - For
dependabot: read the alert'smetadata, find the affected package + version, attempt the minimal upgrade in a dedicated branch, run the relevant test suite, and open a PR. - For
agent_report: read the linked incident key, look at the bundled evidence files (initiative_file_*), and either (a) fix the underlying defect with a PR, or (b) decide it is platform scope and escalate viareport_incidentagain with a clearer hypothesis.
- For
- When you finish (PR opened or fix landed), leave the row in
in_progressand callissue_triage_releasewithkeep_status: trueso it stays in the queue for a human reviewer to verify and close. - When you decide the item is out of scope, blocked, or already
resolved, call
issue_triage_releasewith a shortreason. The row goes back toopenand other agents can pick it up.
Hard rules
- Only claim one item at a time per workspace task unless the
initiative explicitly raised
autoIssueTriage.maxConcurrentClaims. - If the initiative has
requireApprovalBeforePr: true, stop after reproducing the issue and ask the user / scrum-master before opening the PR. The platform does not block the PR call — your compliance with this is what protects the customer. - Never delete or close planning rows directly. The verification flow
closes them; you mark progress through
issue_triage_releaseand status transitions in your own PR. - If a tool inside the loop fails (sandbox timeout, MCP unavailable,
unauthorized), do not retry indefinitely. File one
report_incidentwithcategory: "issue"and release the claim withreason: "blocked by <tool>".