Implement Spec takes a spec and its tickets, treats the tickets as a task graph with blocking relationships, and works the frontier of ready tickets with fresh implementer subagents, each in its own worktree on its own branch. Merger subagents land each result on a single integration branch.
When every ticket is done it calls code-review on the integration branch and fixes the issues in one fresh implementer subagent.
What You Get
- One integration branch holding the entire spec
- A draft pull request opened after the first merge when the tracker closes work through PRs or you ask for one, marked ready at the end
- Tickets resolved the way the issue tracker closes work
- Optional exploration notes saved outside the repo for every subagent to read
- Cleaned-up implementer worktrees
Common Use Cases
- Building a multi-ticket spec with maximum concurrency
- Letting independent tickets proceed in parallel while blocked ones wait
- Finishing a spec with one review pass over the integration branch
What Makes It Unique
It communicates with subagents through context pointers to the spec, tickets, notes and commits, and keeps messages sparse. Each implementer confirms its worktree sits on the integration branch, drives tdd, and merges the integration tip into its own branch before reporting done. New tickets start as soon as a merge changes the frontier.
Under The Hood
SKILL.md is Markdown with no scripts. The model starts background subagents for exploration, implementation and merging, creates git worktrees and branches, merges them, may open a draft pull request through your tracker’s workflow, and calls tdd and code-review. All of that writes to your repository and its tracker. When no issue tracker was provided, the skill tells you to run setup-matt-pocock-skills.