bug-triage-agent

Triage a bug ticket end to end. Disposition first, then release correlation, workaround discovery and priority scoring, with an audit log entry. Use when the user asks to triage a ticket, prioritize bugs, analyse a tracker bug, or asks what priority a bug should be. Triggers include "triage ABC-1234", "priorizá estos bugs", "qué prioridad tiene", "is this a bug or expected behavior", "which release broke this". A bare ticket key is a request to triage it. A bare word routes to its skill without asking - queue to triage-queue, log to triage-history, doctor to triage-readiness, config to triage-config, review to triage-override, auto to triage-auto.

Resolves every enrichment source (tracker, releases, the metrics source, the warehouse, code search) to live MCP calls, recorded fixtures, or a typed unavailable result. Load this before any enrichment. It defines the source contract that the rubric consumes, so live and fixture modes are indistinguishable downstream.

Identify which release most likely touched the area a bug affects, using fix versions and merged pull requests over a full release-cycle lookback. Use when triaging a bug, asking which release caused an issue, or asking what shipped near a report date.

Automatic, unattended triage for a team whose config turns it on (automation.enabled). Picks new, changed and overdue bugs, triages them without asking questions, and writes a review digest. Never writes to the tracker. Use when a scheduled task or a tracker event runs triage-auto, or when the user asks to run automatic triage now. Off by default: if the config does not enable it, say so and stop.

Create or update a team config for the bug triage agent by asking the user a short series of questions, discovering which connectors are available in the session, and writing a validated teams file into the team's own folder. Use when the user asks to set up triage for a team, configure the triage agent, onboard a new team, or when a triage run finds no config for the requested team.

Decide whether a ticket is a defect, expected behavior, a config or data issue, a duplicate, or a voice-of-customer request, before any priority is assigned. Use when triaging a bug, or when asked whether something is really a bug.

Show the bug triage agent's audit log: recent triage decisions, one ticket's triage history, and how often people agreed with the agent. Use when the user asks what was triaged, the triage log or history, what the agent decided for a ticket, or how accurate the triage agent has been. Also use it when the whole request is just log, history or accuracy. Show it straight away.

Record a human override of a bug triage recommendation, with the reason, so the agent's accuracy can be measured and the rubric calibrated. Use when the user disagrees with a triage result, says a ticket's disposition or priority should be something else, or wants to log a correction. Also use it when the request starts with review or override followed by a ticket.

Assign P1 to P4 with escalation caps and a stated confidence basis, for tickets already dispositioned as defects. Use after the disposition gate, never before it.

List the open bugs in the team's connected tracker, show which the triage agent has already handled and what it decided, flag priority gaps and tickets that changed since their last triage, and suggest what to triage next. Use when the user asks for the bug queue, the backlog of untriaged bugs, what is pending, what needs triage, or what to look at next. Also use it when the whole request is just the word queue, as in invoking the bug-triage-agent plugin with queue. Show the queue straight away for the resolved team; do not ask what the user wants to do with it.

Check that a bug triage team config is valid and every configured source is reachable, with each tool binding present in the session, and list sources that are off but could be connected. Use when the user asks to check, test or diagnose the triage setup, why a source is not working, or whether a team is ready to go live. Also use it when the whole request is just doctor or check, optionally with a team. Run it straight away.

Find and state a workaround that unblocks the customer today, drawn from resolved tickets, docs, or the code path. Use when triaging a bug or when asked how to unblock a customer while a fix is pending.