plan-execute_skill

Overview

This skill orchestrates end-to-end execution for non-trivial engineering requests: features, bugs, stories, epics, spikes, and multi-step tasks. It accepts a ticket URL, a file path containing a spec, or plain text, then assembles an agent team, breaks the work into structured tasks, and manages the lifecycle from research through implementation, review, deploy, and empirical verification.

How this skill works

The skill reads the request source (ticket via CLI, file, or text) and determines if the work is simple or requires a coordinated plan. For complex work it enumerates available agent types, justifies exclusions, builds a team (always including the Explore agent), spawns each agent in plan mode, and uses a Team Lead session to name the plan, pick branches, classify the request, and produce verifiable tasks. It enforces research, metadata population, reviews, and a strict commit/push/PR/deploy/verify workflow.

When to use it

• Implementing a new feature that spans multiple components or teams
• Fixing a bug that needs replication, testing, and coordinated changes
• Delivering an epic or story that requires several coordinated subtasks
• Running a spike or research task before implementation
• Any multi-step task requiring automated verification and deployment

Best practices

• Always supply a ticket URL, spec file path, or detailed plain-text prompt to provide full context
• Let the Explore agent research the codebase first and populate relevant_documentation for each task
• Require every spawned agent to run in plan mode and submit plans for Team Lead approval
• Include explicit acceptance criteria, testing requirements, and a verification command in each task metadata
• Commit all outstanding local changes (except secrets), push, open a PR with auto-merge, and monitor CI/deploy

Example use cases

• Turn a Jira story URL into a scoped plan with tasks, implementation branches, and PR ready for review
• Read a spec file for a cross-service feature, create tasks for backend, frontend, and integration tests, and run verification
• Reproduce and fix a flaky production bug by creating empirical replication steps, tests, and a monitored deploy
• Run a spike to evaluate third-party integrations, record learnings, and produce a follow-up implementation plan