PM: Sprint Lifecycle — Autonomous Agents Agentic Workflow

schema_version: 1
id: agent_pm_sprint
title: "PM: Sprint Lifecycle"
description: "Full sprint lifecycle — plan, manage, close. Phase-detecting and idempotent."
overview: |
  A PM agent runs this chain for sprint-scope work. It detects which phase the sprint is in — not yet planned, mid-flight with active tickets, or ready to close — and performs only the next useful action. Run it once per day or whenever sprint state may have changed; re-invocations are safe.

  During planning, it converts the sprint goal into ticketed subtasks, prioritises, and sets initial assignees. During mid-flight, it reviews subtask progress, unblocks stalled work, and dispatches PM, SE, or QA agents to individual tickets. At close, it validates acceptance criteria across all sprint items and produces a sprint summary.

category:
  level: pipeline
  domain: engineering

metadata:
  agent: true
  role: pm

params:
  project_key:
    type: string
    default: ""
    description: "Jira project key (e.g. SCRUM). Leave empty to read from agent env."
  capacity:
    type: string
    default: ""
    description: "Target capacity (e.g. '10 tickets' or '40 points'). Leave empty to use velocity baseline."
  hint:
    type: string
    default: ""
    description: "Optional focus: 'plan only', 'dispatch stalled', 'close sprint', 'move tickets', etc."
  entry_path:
    type: string
    default: "~/workspace"
    description: "Working directory for the agent"

steps:
  - type: terminal.open
    title: "Agent: PM Sprint Lifecycle"
    cwd: "{{entry_path}}"
  - type: terminal.run
    cmd: >-
      source ~/.agent-env && claude --dangerously-skip-permissions
      "you are a PM agent. your mission: manage the sprint lifecycle for project {{project_key}} (if empty, read JIRA_PROJECT from your environment). this workflow is IDEMPOTENT — re-dispatch at any time.

      learn how to PM relevant parts. use product knowledge from skill packs for planning and decomposition — understand module structure, dependencies, and data model before scoping sprint work.

      read sprint state. detect phase:

      PLAN (no active sprint, or active sprint has no tickets):
      groom backlog — flag stale (>30d), duplicates, incomplete specs, priority mismatches.
      velocity analysis — last 2-3 sprints, average velocity, trend.
      scope next sprint up to {{capacity}} (if empty, use velocity baseline). highest priority first, respect dependencies, 20% overage allowed.
      group by execution order (parallel waves). post planning comment.
      move tickets to sprint only if hint contains 'move tickets'.

      MANAGE (active sprint with work):
      assess — tickets by status, completion rate, story points.
      stalled detection — In Progress or In Review with no update >24h.
      PR checks — find PR links, check CI with gh CLI.
      agent fleet — active vs idle, cross-reference stalled + idle = silent failure.
      dependency order — group 2+ active while dependency not Done?
      new tickets — unclassified, recently created.
      act: dispatch stalled (agent_se_work/agent_se_followup), merge ready (QA PASS → merge → Done → agent_pull_repos), triage new, flag blockers.

      CLOSE (all tickets Done, or sprint end date passed):
      partition Done vs incomplete. carry over incomplete to next sprint.
      release notes from merged PRs — categorize: Features, Bug Fixes, Improvements, Infrastructure.
      velocity trend + forecast for next sprint.
      close sprint via Jira agile API.

      post one status comment: sprint summary, actions taken, risks, next steps.

      {{hint}}"