workflow-implement-phases_skill

Overview

This skill orchestrates multi-phase implementation from a plan document by extracting phases, analyzing dependencies, and coordinating sub-agents to execute each phase. It chooses an optimal execution strategy (parallel, sequential, or mixed), presents a clear plan for confirmation, and aggregates results into a coordination directory. The orchestrator never implements phases directly; every phase is run via a Task() sub-agent.

How this skill works

The skill reads a plan file (or lists plans in docs/plans/), parses phases, acceptance criteria, and explicit/implicit dependencies. It builds an execution graph and selects an execution strategy using heuristics for file, API, config, and build dependencies. For each phase the orchestrator spawns a dedicated Task() sub-agent with a full implementation prompt, waits for results, handles failures, and aggregates outputs into .claude/phase-coordination.

When to use it

• You have a multi-phase plan document and need safe, auditable orchestration of work.
• You want to maximize parallelism where safe while preserving dependency ordering.
• You need strict context isolation to avoid context saturation in the main agent.
• You require structured artifacts and results per phase for downstream tasks or reviews.
• You must ensure failures in one phase do not corrupt other phase executions.

Best practices

• Always spawn one Task() sub-agent per phase — never implement a phase directly in the orchestrator.
• Use the auto strategy to let the skill analyze dependencies; override with --strategy only when sure.
• Topologically sort dependencies and group independent phases into parallel levels for mixed execution.
• Populate clear acceptance criteria and file/import paths in the plan to improve dependency detection.
• Store artifacts and results under .claude/phase-coordination/artifacts and .claude/phase-coordination/results for traceability.
• When confidence is low, prefer sequential execution or ask the user to confirm risk trade-offs.

Example use cases

• Implement feature rollouts described in docs/plans/my-plan.md with mixed parallel/sequential phases.
• Run independent backend and frontend tasks in parallel while enforcing order for shared schema migrations.
• Automate infrastructure and deployment tasks where certain phases depend on created secrets or artifacts.
• Orchestrate a stepwise migration where each migration phase writes artifacts consumed by later phases.
• Coordinate multiple engineers' work by isolating each phase implementation into its own sub-agent for independent review.