How to use the Implement Spec Tickets agent skill

What an agent skill is and what implement-spec makes the agent do

An agent skill is a folder with a SKILL.md file that gives an AI coding agent a set of instructions for a specific job. This one covers the job of implementing a spec that has already been written.

The skill treats the tickets as a task graph with blocking relationships, not as a plain checklist. At any moment there is a frontier of tickets that are unblocked and ready to start. The agent starts those, hands each ticket to a fresh sub-agent working in its own git worktree and branch, and merges finished work onto a single integration branch. When all tickets are done, it runs a code review on the integration branch, fixes the issues in one more sub-agent pass, and then closes the tickets in the issue tracker.

Install the implement-spec skill

Download the ZIP for this skill from this page. The package includes SKILL.md and agents/openai.yaml. Then choose one of the following options.

For a personal install that works across your projects, unzip the folder into ~/.claude/skills/. To limit it to one project, unzip it into that project's .claude/skills/ folder.

In the Claude apps, you can instead upload the ZIP under Settings, then Capabilities, then Skills.

• Personal install: unzip into ~/.claude/skills/
• Project install: unzip into the project's .claude/skills/
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Set up your repository and issue tracker first

The skill expects an issue tracker to be configured already. If it has not been provided, the agent will tell you to run the setup command below before continuing.

You also need a spec and its tickets, created with the /to-spec and /to-tickets workflow. The skill also depends on the tdd and code-review skills, which it calls through the Skill tool, and on git with worktree support.

• Git with worktree support
• An issue tracker configured through /setup-matt-pocock-skills
• The tdd and code-review skills, invoked by the Skill tool
• A spec and associated tickets produced by /to-spec and /to-tickets

How to start the skill with a request

The skill is marked disable-model-invocation, so the agent will not load it on its own from a loose request. Invoke it explicitly by its name, and point it at the spec and its tickets so it can read the task graph.

Once started, the workflow runs in the order the skill describes: read the spec and tickets, optionally run an exploration sub-agent that saves notes outside the repo, create the integration branch, implement each ticket in its own worktree, merge each result, pick up newly unblocked tickets, review, and clean up worktrees.

• Name the spec and its tickets so the agent can find them
• Say whether you want a draft PR; the skill opens one only if the tracker closes work through PRs or you ask for it
• Expect many sub-agents, worktrees and merges to run

Example of a typical run

Suppose a spec has eight tickets, where three depend on earlier ones. The agent starts the tickets with no blockers in parallel, each in its own worktree and branch. As each finishes, a merger sub-agent adds it to the integration branch, which can unblock the next tickets.

When every ticket is complete, the code review runs on the integration branch and a single fresh implementer fixes what it raises. If a draft PR exists it is marked ready for review. Otherwise each ticket is resolved the way your tracker closes work, and the integration branch is reported to you.

Safety notes and tips

The skill uses the network and performs git and tracker actions, including creating branches, opening PRs and closing tickets. It does not run scripts. If you do not want those actions taken automatically, do not use this skill in that repository.

Keep communication lean. The skill asks sub-agents to pass context pointers to the spec, tickets, notes and commits rather than repeating content, so keep your spec and tickets complete and up to date before you start.

• Do not use it for quick one-off changes; the workflow is heavy for small tasks
• Make sure the repo has no unwanted branch or PR automation before running it
• Keep exploration notes in a directory outside the repo so all sub-agents can read them