AI-Powered Development Workflow

Jira Workflow Kit

AI and humans develop together. Jira connects them.

$ claude → /idea, /plan, /develop, /qa, /deploy
The Problem

The gap between business and engineering

📝

Hard to get right

BAs, Product Owners, and stakeholders know what they want — but writing well-structured, testable acceptance criteria takes time and expertise

🔄

Slow feedback loops

Weeks between requirement → implementation → review. Stakeholders lose context, developers lose momentum

😴

Idle capacity

Approved tickets sit in the backlog. Developer machines idle 16 hours a day. Time zones don't align.

The Solution

AI and humans develop together

Jira Workflow Kit bridges business and engineering.
AI structures requirements, writes code, runs tests — humans review, co-develop, and steer.
Nothing happens without human approval.

AI
/idea
→
Human
Review
→
Human
Approve
→
AI
/develop
→
AI
/qa
→
Human
Merge
Core Principle

Human in the loop — always

Business Analyst

Describes requirements

Provides the idea in natural language. AI structures it into a proper Epic with testable acceptance criteria.

Product Owner

Reviews & prioritizes

Validates stories in Jira. Adjusts scope, adds comments, refines criteria. Sets status to "Approved" when ready.

Stakeholder

Comments & validates

Reviews tickets in Jira at any time. Leaves feedback. AI processes it via /feedback and adjusts the plan.

AI Agent

Executes approved work

Only touches code after explicit approval. Plans, implements, tests, creates PRs — all traceable in Jira.

No coding without approval. The "Approved" status is the gate. AI never bypasses it.

How It Works

From requirement to PR

# BA describes a business requirement
$ /idea "Customers need two-factor authentication on login"

→ AI analyzes codebase, creates Epic + Stories in Jira
→ PO reviews, stakeholders comment, team approves in Jira

$ /plan PROJ-42
→ Deep code analysis, creates implementation subtasks

$ /develop PROJ-42
→ Implements subtask by subtask
→ Commits, pushes, updates Jira — each step

$ /qa PROJ-42
→ Tests against acceptance criteria, creates PR
→ Human reviews and merges
Live Board

Your Jira board — always in sync

To Do 2
● PROJ-51
Email notification preferences
portal
■ PROJ-52
Migrate legacy payment API
platform
Approved 2
● PROJ-48
Two-factor authentication
platform
● PROJ-49
User profile page redesign
portal
In Progress 2
● PROJ-42
Customer login flow
platform AI
■ PROJ-46
Refactor invoice templates
platform SK
In Review 1
● PROJ-40
Dashboard analytics widgets
portal AI
Done 3
● PROJ-38
Password reset flow
platform AI
■ PROJ-39
API rate limiting
platform MR
◉ PROJ-41
Fix login redirect loop
portal AI

AI and human developers work on the same board. Labels separate projects. Status updates are automatic.

Full Traceability

Every decision documented in Jira

PROJ-42 Customer login flow with 2FA support In Progress
Subtasks — 3/5 complete
✓
PROJ-53 Create authentication middleware Done
✓
PROJ-54 Add TOTP generation + validation service Done
✓
PROJ-55 Build QR code enrollment endpoint Done
●
PROJ-56 Create 2FA verification UI component In Progress
PROJ-57 Add recovery codes fallback To Do
Activity
AI
Claude (AI Agent) Automated 2 min ago
Subtask PROJ-55 completed.
Commit: a3f8c21
Files: src/auth/totp-enrollment.ts, src/routes/auth.ts
Tests: 4 new, all passing
PO
Sarah K. Product Owner 1 hour ago
Please ensure recovery codes are 8 characters, not 6. Users find longer codes more trustworthy.
Details
Type ● Story
Priority ▲ High
Assignee AI Claude
Sprint Sprint 12
Epic PROJ-35
Labels acme-platform
Branch feature/PROJ-42-login-2fa
Progress
3 of 5 subtasks done
Flexibility

AI and human — side by side

The workflow integrates into your existing Jira process.
Developers and AI work on the same board, same branch, same conventions.

AI Agent

Handles via /develop

  • Reads the subtask from Jira
  • Implements following codebase patterns
  • Runs tests, commits, pushes
  • Updates Jira with full audit trail
Human Dev

Works alongside AI

  • Picks up subtasks from the board
  • Implements with full context
  • Same branch, same conventions
  • AI handles the rest in parallel

AI augments your team — it doesn't replace it. Humans stay in the driver's seat.

Game Changer

/autopilot

The team goes home. AI keeps working.
Plans, develops, tests, creates PRs — overnight.

What it does

  • Picks up approved tickets
  • Creates sprints automatically
  • Plans → develops → QA → PR
  • Pushes every commit immediately
  • Writes sprint summary reports

When stuck

  • Documents the question in Jira
  • Parks the ticket for team review
  • Moves to the next ticket
  • Sleeps through rate limits
  • Never guesses — asks via Jira
Architecture

Multi-project, one launchpad

~/jira-workflow-kit/
├── CLAUDE.md             # Project registry + rules
├── .claude/commands/     # 11 skill files
├── projects/
│   ├── acme-platform/    # Backend API
│   └── acme-portal/      # Customer frontend
└── .mcp.json             # Jira connection

One Jira board with labels for project separation.
Skills auto-detect the project from ticket labels — across any number of repositories.

Complete Toolkit

11 skills, full lifecycle

CommandPurposeMode
/ideaStructure requirements → Epic + Stories in JiraInteractive
/feedbackProcess stakeholder feedback from Jira commentsInteractive
/planCodebase analysis → implementation subtasksInteractive
/developImplement subtask by subtask with auto-pushInteractive
/qaTest, validate acceptance criteria, create PRInteractive
/bugInvestigate root cause, quick-fix or ticketInteractive
/autopilotAutonomous overnight developmentAutonomous
/standupMorning briefing with overnight summaryRead-only
/deployBuild, test, deploy a projectInteractive
/retroSprint wrap-up + learning captureInteractive
Seamless Integration

Jira is the single source of truth

Every change tracked

Each commit links to a subtask. Each subtask has a Jira comment with what was done, files changed, and test results.

Status always current

Ticket status updates at every phase transition. No stale boards, no "forgot to move the card." Stakeholders see real-time progress.

Decisions documented

Even in autopilot mode, every decision is logged as a Jira comment. Full audit trail for compliance and review.

Your existing Jira workflows, boards, and permissions stay intact.
The kit plugs in — it doesn't replace your process.

Continuous Learning

The AI gets better over time

Every session builds on the last. The AI remembers architecture decisions, user preferences,
debugging patterns, and project conventions — permanently.

Persistent Memory

Memory files survive across sessions. Architecture patterns, tech stack details, known issues — always available.

memory/
  MEMORY.md
  diemcrm.md
  working-style.md
  debugging.md

/retro captures learnings

After each session, the AI analyzes what worked, what was painful, and updates its memory with verified insights.

✓ Patterns confirmed
✓ Debug solutions saved
✓ Preferences recorded
✓ Noise filtered out

/feedback closes the loop

Human corrections in Jira are processed and incorporated. The AI adjusts plans, requirements, and code — learning what matters.

PO comment → /feedback
→ plan adjusted
→ next time: done right

Context accumulates

Each /plan builds codebase understanding. Each /develop follows discovered patterns. Knowledge compounds.

Session 1: learns stack
Session 5: knows patterns
Session 20: moves like a team member
Impact

Before vs. After

Without

  • BAs write vague requirements manually
  • Requirements lost in translation
  • Weeks from idea to working code
  • Jira status always outdated
  • No work happens outside business hours
  • Every session starts from zero
→

With Workflow Kit

  • AI helps BAs write structured, testable tickets
  • Codebase-aware acceptance criteria
  • Hours from approval to PR
  • Jira always in sync, automatically
  • Autopilot develops overnight
  • Persistent memory across sessions
The Output

PRs ready for review — by morning

● Open PROJ-42: Customer login flow with 2FA support #87
claude-ai wants to merge 5 commits into develop from feature/PROJ-42-login-2fa
Summary
  • Add TOTP-based two-factor authentication to customer login flow
  • QR code enrollment endpoint for initial 2FA setup
  • Recovery codes fallback (8-character codes per PO feedback)
Test Plan
  • ☑ Unit tests: 12 new, all passing
  • ☑ Integration tests: 4 new, all passing
  • ☑ Acceptance criteria validated (5/5)
All checks have passed
✓
CI / Unit Tests Passed in 1m 42s
✓
CI / Integration Tests Passed in 3m 15s
✓
CI / Lint + Type Check Passed in 28s
8 files changed
+ src/auth/totp-service.ts +142
+ src/auth/recovery-codes.ts +87
~ src/routes/auth.ts +34 -8
~ src/middleware/auth.ts +18 -3
+ tests/auth/totp.test.ts +96
All checks passed. Ready to merge.
The Morning After

Wake up to /standup

$ /standup

Good morning. Here's what happened:

Autopilot overnight:
  Completed: PROJ-42, PROJ-43, PROJ-44 (3 tickets, 3 PRs)
  Parked: PROJ-45 (needs team input)

Feedback needed:
  PROJ-45: Should failed provisioning send notification?

Current sprint: 3/5 done, 1 parked
Recent: 11 commits across 2 projects

What would you like to work on?
Under the Hood

Built on proven tools, with guardrails

Technology

  • Claude Code CLI (Anthropic)
  • Jira via Model Context Protocol
  • Native Git operations
  • Any tech stack — learns your patterns

Safety

  • No coding without "Approved" status
  • Never force-push or delete branches
  • Tests mandatory before completion
  • Stops after 3 consecutive failures
  • Full audit trail in Jira comments

Your team + AI. Together, faster.

Approved tickets don't have to wait.
Not for morning. Not for the next sprint.

$ /autopilot
PRs ready by morning.

github.com/diem2001/jira-workflow-kit