Back to Documentation

From Plan to Ship

Your AI agent workflow — from requirements to production code, with full context at every step.

The Workflow Loop

Agiflow provides MCP skills (prompts) for each phase of your development workflow. Your AI agent uses these to manage projects, refine tasks, and execute work systematically.

Plan
Refine
Groom
Execute
Triage

Planning

Skill: project_plan

Break high-level requirements into concrete tasks.

Key Actions

  • - Create 5-15 focused tasks (15 min – 2 hr each)
  • - Define 2-5 acceptance criteria per task
  • - Group 3-8 related tasks into Work Units
  • - Assign priorities and dependencies

Example Prompt

“Plan the shopping cart feature for our e-commerce app”

💡 How It Works: Your agent calls describe_capabilities to discover project tools, then creates tasks with acceptance criteria via create_task.

Task Refinement

Skill: refine_task

Add detail and context to tasks before execution.

Key Actions

  • - Add detailed acceptance criteria (2-5 per task)
  • - Document technical context and constraints
  • - Link related tasks and set dependencies
  • - Assign to team members or agent profiles

Backlog Grooming

Skill: backlog_grooming

Prioritize and organize your backlog.

Key Actions

  • - Review and reorder tasks by priority
  • - Move tasks between work units
  • - Identify blockers and dependencies
  • - Archive or close stale tasks

Execution

Skills: run_work (work unit) and run_task (single task)

Agent executes tasks with full context.

1

Load Task Context

Agent loads the task via MCP — reads acceptance criteria, devInfo, and related context.

2

Understand Conventions

Before coding, calls get-file-design-pattern (Architect MCP) to understand project conventions.

3

Implement

Implements the task following acceptance criteria and discovered patterns.

4

Validate

After coding, calls review-code-change (Architect MCP) to validate against project rules.

5

Update & Commit

Updates task status and devInfo, then commits changes.

💡 Scoped Context: Agents automatically get the right context from scoped MCP. No re-explaining needed between sessions.

Triage & Review

Skills: triage, review_work, daily_standup

Review completed work, provide feedback, and iterate.

Key Actions

  • - Review agent output and artifacts
  • - Add comments for feedback (agent reads these on next session)
  • - Move tasks through status columns: Todo → In Progress → Review → Done
  • - Track progress via devInfo metadata

Agent reads previous comments and devInfo when resuming — no context loss.

Session Continuity (devInfo)

Work Units maintain structured devInfo metadata that agents access via MCP. This means agents resume with full context — files changed, test results, progress.

devInfo Structure
{ "devInfo": { "executionPlan": "Backend → Frontend → Tests", "progress": { "completedTasks": 2, "totalTasks": 6 }, "filesChanged": [ "backend/migrations/003_cart.sql", "backend/src/repos/CartRepo.ts" ], "testResults": { "passed": 12, "coverage": "85%" } } }

Related Documentation

Ready to Organize Your Development Workflow?

Plan, execute, and review — all through your AI agent.