post-plan-workflow_skill

Overview

This skill automates the post-plan materialization and implementation workflow for MCP-managed plans. It creates MCP items from an approved plan, wires dependencies and gating, then dispatches implementation agents and verifies completion. The workflow enforces a strict three-phase sequence: materialize, implement, verify.

How this skill works

After plan approval the skill calls create_work_tree or manage_items to materialize MCP items and tags them to trigger schema gates. It ensures dependency edges and required queue-phase notes are present, then dispatches subagents with item UUIDs so they can own lifecycle transitions and fill work-phase notes. Finally it runs health checks and diagnostics to surface stuck items and missing notes.

When to use it

• After a plan is approved and MCP tracking is enabled
• When you need structured task graphs with dependency gating
• When multiple autonomous agents will execute plan items
• When you require persistent tracking and auditability of work
• When you want automated checks to catch incomplete materialization before dispatch

Best practices

• Always complete full materialization (create items, wire edges, populate required queue-phase notes) before dispatching any agents
• Prefer create_work_tree for structured plans; use manage_items for single or ad hoc items
• Embed expectedNotes guidance into delegation prompts so agents can author required notes correctly
• Dispatch one subagent per MCP item and include the item UUID in the delegation prompt
• Verify dependency readiness with get_blocked_items between waves and avoid dispatching blocked items
• Use query_items and get_context to diagnose and manually advance any stuck work during verification

Example use cases

• Materialize a feature-implementation plan into an MCP work tree, tag items, and dispatch developers as subagents
• Orchestrate parallel research tasks using fan-out/fan-in edges and enforce acceptance-criteria notes before review
• Run automated verification after an implementation wave to detect items that never transitioned
• Recover from a partial create_work_tree failure by querying overview, cleaning partial items, and retrying
• Coordinate multi-agent integration work where each agent reports progress via MCP item transitions