How to use the Vertical Slice Validator agent skill

What is an agent skill, and what does this one make the agent do?

An agent skill is a folder containing a SKILL.md file that tells an AI agent how to handle a specific kind of task. This skill is for game developers who already have design documents and architecture notes and want a structured go/no-go check before full production.

Once installed, the agent reads your existing design files, proposes a scope, and asks for your confirmation at each step. It saves progress notes to a project file and writes build files with a standard header and quality rules. It also records how long the work takes, which the skill treats as your real production-rate estimate.

• Reads files such as CLAUDE.md, design/gdd/game-concept.md and docs/architecture/ before proposing anything.
• Defines a falsifiable validation question covering both player experience and build feasibility.
• Plans a slice of 3–5 minutes of continuous gameplay and asks you to confirm scope before building.
• Saves a session checkpoint to production/session-state/active.md so work can resume across sessions.
• Adds a header to each build file marking it as a vertical slice that is not for production.

Install the vertical-slice skill from the ZIP

Download the ZIP for this skill from this page and unzip it. The package contains SKILL.md. Then place the folder in one of two locations, depending on how you want to use it.

For a personal install that works across your projects, put it in ~/.claude/skills/. For a single project, put it in that project's .claude/skills/ folder so it travels with the repository.

If you use the Claude apps instead, you can upload the skill under Settings → Capabilities → Skills.

• Personal: ~/.claude/skills/vertical-slice/
• Project: .claude/skills/vertical-slice/ inside your game repository
• Claude apps: Settings → Capabilities → Skills, then upload

Check the requirements before you start

The skill needs Claude Code or a compatible SKILL.md agent runtime. It also expects existing project files that it references: CLAUDE.md, design/gdd/game-concept.md and docs/architecture/. If you have not written these yet, the skill is the wrong stage. The skill itself points to a separate prototype skill for early concept testing.

The skill runs a configuration helper script at hooks/yaml-helper.sh, which is expected to exist in the wider repository, so you need a bash shell to run it. The skill does not use the network.

• Claude Code or a compatible SKILL.md runtime
• CLAUDE.md, design/gdd/game-concept.md and docs/architecture/ (as referenced by the skill)
• A bash shell for hooks/yaml-helper.sh, which is expected in the wider repository
• Works for game projects only; the workflow assumes GDDs and engine-specific structure

Trigger the skill with a natural request

The skill is user-invocable and accepts an optional review argument, --review full|lean|solo, which overrides the review_mode setting for that run. You can also simply describe what you want in plain language once the skill is installed and the request matches its purpose.

Good requests name the stage and the goal, for example asking the agent to validate whether one full loop can be built at production quality before committing to production. Mention your engine and where your design documents live if they are not in the default paths.

• "Run a vertical slice check on our game before we commit to full production."
• "Can we build one complete start-challenge-resolution loop at production quality in two weeks?"
• "Use the vertical slice skill in lean review mode."

Walk through a typical session

The skill works in phases. First it resolves the review mode and loads context, reading each GDD, ADR and the control manifest by section. Then it defines the validation question, using your core fantasy from game-concept.md or game-brief.md, and asks you to confirm the scope before any building starts.

During implementation, the agent asks for permission to create the slice directory at prototypes/[concept-name]-vertical-slice/. After it writes the initial files, it asks you to run the build and report what you observe. You repeat this loop until the full cycle is playable, then conduct at least one playtest.

The skill includes a sunk cost checkpoint on day 3 of the planned timeline. If the full loop is not yet demonstrable, the agent is told to stop and surface the blocker rather than keep iterating.

• Phase 1: resolve review mode and load context
• Phase 2: define the validation question and scope, then confirm
• Phase 3: plan the build and record the checkpoint file
• Phase 4: implement, iterate with your feedback, then playtest

Keep the slice honest: scope, quality and playtesting

The skill is strict about scope. It says to include all core loop systems that a single start-challenge-resolution cycle requires, and to cut content rather than quality. If the slice scope feels too large to build in one to three weeks, the skill treats that as a sign the scope is wrong.

For playtesting, the skill suggests watching someone who has not seen the game play it silently, without explanation. If you have no outside testers, it suggests rotating roles within the team, or stepping away for 2–3 days before playing your own build.

• Cut scope before cutting quality.
• Avoid adding "just one more system" once features feel almost ready.
• Watch first-time players silently and note where they get confused.
• Solo developers: step away for 2–3 days before playing through.