taskflow

The task lifecycle layer of the hjemmesidekongen/ai methodology. The methodology says structure before code - taskflow is where that structure comes from. It connects your project management tools to the development workflow so every task starts with clear requirements, flagged contradictions, and a defined path to completion.

Version: 0.1.0 · 9 skills · 8 commands

What It Does

taskflow turns Jira tickets into local YAML files, analyzes them for contradictions, and provides a structured lifecycle from ingestion through completion. When you're done coding, it creates PRs with context-rich descriptions and generates QA handover documents grounded in acceptance criteria.

Every MCP integration degrades gracefully. If Jira isn't configured, you get clear messages - never fabricated data.

Commands

Single Task Workflow

/task:ingest PROJ-123          # Pull ticket from Jira → local YAML
/task:start PROJ-123           # Activate, review contradictions
/task:docs                     # Find related Confluence pages
# ... do the work ...
/task:pr                       # Create PR with structured description
/task:done                     # Generate QA handover, close Jira ticket

Sprint Batch Workflow

/task:ingest-bulk <board-url>  # Load entire sprint (deduplicates)
/task:list --status pending    # See what's available
/task:start PROJ-456           # Pick one
/task:done                     # Complete with QA handover

Key Concepts

Contradiction Detection

This is the feature that pays for itself immediately.

When you ingest a Jira ticket, taskflow reads the description and every comment. It identifies where later comments contradict, modify, or extend the original requirements. Each finding is classified by severity:

• Blocker - direct contradiction that changes what you need to build
• Warning - modification that shifts scope or priority
• Info - addition or clarification that doesn't conflict

You see contradictions when you /task:start and again in /task:status. No more discovering mid-implementation that comment #7 reversed the requirement from comment #3.

Structured PR Descriptions

/task:pr creates pull requests with descriptions auto-populated from:

• Task summary and key
• Acceptance criteria
• Test plan derived from criteria
• Pipeline status (checks CI/CD after creation)

The PR description is immediately useful to reviewers - not an empty "ready for review" with a Jira link.

Supports Bitbucket (via MCP) with draft mode. Posts the PR URL as a Jira comment automatically. Falls back to formatted output for manual submission when Bitbucket MCP isn't configured.

QA Handover

/task:done generates a structured QA handover document:

• What changed - file-by-file summary from git diff
• Test scenarios - mapped from acceptance criteria, not generic checklists
• Regression risks - identified from the code diff
• Screenshots - included for UI-heavy changes

Output format matches your project profile - Jira wiki markup, GitHub-flavored markdown, or plain markdown. Goes straight into the right system without reformatting.

Cross-Session Continuity

session-handoff-taskflow extends claude-core's session-handoff with task-specific context:

• Active task state and progress
• Subtask completion (pending/in_progress/done)
• Blockers and why you're stuck
• Open questions for the next session
• Staleness classification on resume

Proactively suggests handoff when 5+ files have been edited in a session.

Component Reference

Skills

Project Profiles

project-profile-loader reads your rules.yml (project root or .ai/rules.yml) to understand project-level conventions:

• Git branch naming patterns
• Commit message style
• QA handover format
• Dev commands (test, build, lint)
• MCP server availability

All downstream skills consume this profile. Your PR descriptions, QA handovers, and git commits automatically follow your project's conventions.

MCP Integrations

No integration is required. taskflow works with whatever MCP servers you have configured and tells you clearly when one is missing.

Local Storage

Everything lives in .ai/tasks/:

.ai/tasks/
├── PROJ-123.yml              # Normalized task with metadata + contradictions
├── PROJ-124.yml
├── active.yml                # Currently active task pointer
├── PROJ-123-handover.md      # QA handover (generated at completion)
├── PROJ-123-pipeline-status.yml  # CI/CD status snapshot
├── handoff.yml               # Cross-session context
├── bulk-ingestion-summary.yml    # Batch ingest report
└── attachments/
    └── PROJ-123/             # Downloaded Jira attachments

All YAML, all local, all version-controllable. No database, no external state.

See also

• claude-core - the foundation plugin that provides MCP skills and session continuity
• dev-engine - the execution pipeline that consumes task context from taskflow
• Workflows - end-to-end recipes using taskflow commands