Playtest And Tune

Purpose: Evaluate the current playable build, fix obvious friction, and tune the smallest set of values needed for a better MVP loop.

Use when:

• a playable slice already exists
• timing, readability, feel, or pacing problems are visible in the current build
• the team wants one focused tuning pass with a written outcome

Do not use for:

• first-time integration of systems
• large feature additions disguised as tuning
• speculative balancing before the game loop can actually be played

Inputs / Required Context:

• required: a runnable coherent playable loop that has already been assembled
• optional: reports/mvp-assembly-report.md, design/gdd/systems-index.md, relevant design docs, and previous prototype reports
• optional explicit argument: focus area such as combat, pacing, pickups, readability, or controls

Outputs / Owned Artifacts:

• owns one focused playtest pass on the current build and any small accepted tuning changes
• owns reports/playtest-report.md whenever a real playable build was actually run, even if the final result is that no tuning change was adopted
• may update constants, spawn values, timing, UI feedback, collision or recovery windows, and small QoL logic
• uses gamedev/templates/playtest-report.md

Modes or Arguments:

• no argument: run a general MVP sanity pass
• [focus-area]: bias the session toward one concern such as combat-readability or pickup-pressure

Execution Rules:

1. Before changing anything, verify that a coherent playable loop exists and can actually be run in the current environment.
2. If the loop is missing or the build does not start, stop immediately and route back to assemble-mvp or implement-system; do not write reports/playtest-report.md and do not make tuning edits.
3. Read unresolved High-Risk Systems rows in design/gdd/systems-index.md when present and prefer closing the most important still-open evidence gap during this pass.
4. Run the current build and observe one focused playtest pass before changing values.
5. Identify the smallest set of changes that would most improve readability, feel, pacing, or control clarity.
6. Tune values and lightweight feedback only; do not expand content scope during tuning.
7. Re-run the build after changes and compare the result to the original state.
8. Prefer repo-native verification commands for the chosen runtime over ambient external tooling that is not part of the repo or platform contract.
9. If specialist QA tooling is unavailable but the primary run target already worked, record the exact blocker and continue the tuning pass instead of treating the run as failed.
10. For browser-specific HUD, render, or overlay issues, use browser specialist playtest guidance when available but keep the accepted result and canonical report in this step.
11. Write reports/playtest-report.md using the canonical template and record what changed, or explicitly record that no change was adopted after the real pass.
12. Save accepted evidence in stable project paths or make the verification command reproducible enough that another operator can regenerate it; do not cite ignored scratch dirs as the only durable proof.
13. If this pass closes one or more unresolved High-Risk Systems rows, update their notes and the progress snapshot in design/gdd/systems-index.md in the same pass.
14. If a tuning change becomes the new accepted default, sync the relevant GDD in the same pass or record an explicit follow-up path; do not silently contradict design docs.
15. If the playtest disproves an older prototype or GDD baseline, update the canonical docs to the new accepted read instead of leaving both versions live.
16. Recommend either another small tuning pass, prototype, or the next production step.

Failure / Stop Conditions:

• stop if no coherent playable loop exists yet; route back to assemble-mvp or implement-system
• stop if the build does not run at all; route back to assemble-mvp or implement-system
• do not emit reports/playtest-report.md for a blocked run
• do not treat missing optional specialist QA tooling as a blocked run when the primary playtest target already executed successfully
• stop if the request is really a major redesign instead of a tuning pass
• do not hide major structural problems behind endless constant tweaking

Return Format:

• focus area
• changes made
• observed improvement or lack of improvement
• report path
• docs synced or follow-up path recorded
• high-risk rows closed or still unresolved, if any
• next recommended skill: implement-system, assemble-mvp, or another playtest-and-tune

Example Invocation:

• /playtest-and-tune
• /playtest-and-tune combat-readability

Related Skills / Boundary:

• use assemble-mvp before this skill so there is a coherent playable loop and a real run target
• use platform specialists only for runtime-specific QA depth; keep accepted tuning, reporting, and doc sync in this step
• use implement-system if a missing mechanic, not tuning, is the real blocker

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: avelrl
• Source: avelrl/skills
• License: MIT

Install and usage instructions live in the source repository linked above.