PM: Drive Ticket — Autonomous Agents Agentic Workflow

schema_version: 1
id: agent_pm_drive
title: "PM: Drive Ticket"
description: "Drive a ticket to completion — plan, dispatch SE/QA, track PRs, merge, validate, close. Idempotent."
overview: |
  A product manager agent runs this end-to-end chain to keep a parent ticket moving from planning through close without hand-holding. The workflow inspects the ticket state on every invocation, detects the current phase (plan, implement, review, merge, validate, close), and performs only the next useful action. Re-dispatching is safe and expected — operators typically re-run the same workflow until the ticket reaches Done.

  Inputs are a Jira ticket URL and optional agent hints. The PM reads subtasks, loads the relevant knowledge packs for affected modules, performs gap analysis, then delegates subtasks to SE and QA agents, tracks their PRs, drives merges, and handles final acceptance. Use this when a parent ticket needs ongoing coordination instead of a one-shot action.

category:
  level: pipeline
  domain: engineering

metadata:
  agent: true
  role: pm

params:
  ticket_url:
    type: string
    description: "Jira ticket URL (parent with subtasks, or ticket to plan)"
  hint:
    type: string
    default: ""
    description: "Optional extra instructions for the agent"
  entry_path:
    type: string
    default: "~/workspace"
    description: "Working directory for the agent"

steps:
  - type: terminal.open
    title: "Agent: PM Drive Ticket"
    cwd: "{{entry_path}}"
  - type: terminal.run
    cmd: >-
      source ~/.agent-env && claude --dangerously-skip-permissions
      "you are a PM agent. your mission: drive {{ticket_url}} to completion. this workflow is IDEMPOTENT — re-dispatch until the ticket is Done.

      read the ticket, all subtasks, and recent comments. BEFORE planning or dispatching: learn how to PM relevant parts — load skill packs for the domain and each affected module. use this product knowledge for gap analysis and dispatch decisions.

      detect phase:

      PLAN (no subtasks exist):
      gap analysis using product knowledge from skill packs — for each requirement, what exists vs what needs to change.
      decompose into subtasks — each is one shippable behavior change with observable acceptance criteria, sized for one SE session.
      assign execution groups by dependency. create subtasks in Jira. post plan summary comment.

      REFINE (subtasks exist but plan has gaps):
      check each subtask: clear scope? specific acceptance criteria? correct dependencies? right execution group?
      check overall: all parent requirements covered? dependency chain valid?
      create missing subtasks, update incomplete ones. post refinement comment.

      MANAGE (subtasks ready for work):
      for each subtask, take the next lifecycle action:
      To Do → dispatch agent_se_work.
      In Review + CI passing → dispatch agent_qa_validate (PR# in hint). CI failing → dispatch agent_se_followup.
      QA PASS → merge PR → Done. QA FAIL → dispatch agent_se_followup.
      bugs found by QA → create Bug tickets, dispatch agent_se_work.
      silent failure (idle agent + no comment) → re-dispatch to a DIFFERENT agent.
      after merging → dispatch agent_pull_repos to idle agents. respect execution group order.
      try to avoid dispatching jobs to yourself — you are the PM orchestrator, not a worker. check fleet status and prefer dispatching to other idle agents.

      VALIDATE & CLOSE (all subtasks Done, PRs merged):
      verify on live site. bugs found → new subtasks, stay in MANAGE. clean → transition parent to Done.

      post one status comment: progress (N/M done), actions taken, blockers, next steps.

      {{hint}}"