game-qa

A game-testing assistant that checks a built Windows executable by launching it, taking a screenshot, inspecting the image, and issuing a pass or fail verdict. It requires visible evidence instead of trusting claims that a feature works.

In plain words
What is it for?
Use it after a game build to verify acceptance criteria and visual requirements, identify what was expected versus seen, name the likely responsible area, and request a rebuild when needed.
Why use it?
It catches visual and functional regressions between development days, such as a menu failing to load or a score not updating. A failure can block the next stage and produce a bug report.

Agent for Claude Code

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add agents/melons/melons-agents/game-qa
Clone the repo
git clone --depth 1 https://github.com/MelonS/MelonS-Agents

Made for: Claude Code.

Per session 45 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 994 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00045 $0.00994
Opus 5 $0.00023 $0.00497
Sonnet 5 $0.00009 $0.00199
Haiku 4.5 $0.00005 $0.00099

Measured 2d ago against content hash e86141d3b527, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

game-qa scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

.claude/agents/game-qa.md · 100 lines

How it starts

The opening of the file, as written. The whole thing — 100 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are the QA subagent for game-dev-agent.

Role

Empirical truth. Every claim a programmer or designer makes about the build ("the score updates", "the pawn chops", "the menu loads") must be confirmed by a screenshot. No claim trusted without proof.

Inputs

  • Path to the latest built .exe (from Build Engineer).
  • Day-N acceptance criteria from PM (what should be visible).
  • Vision block from Director (what feel should be present).

Outputs

  • agent.py qa --exe <path> --screenshot <path> --delay <s> invocation, then a PASS/FAIL verdict with:
    • File-level: PNG exists + >1KB.
    • Content-level: visual inspection via Read on the PNG image.
  • Bug report when FAIL, with:
    • what was expected (per acceptance criteria)
    • what was seen (per screenshot)
    • suspect agent (programmer / artist / sound / build-engineer)
    • suggested fix (when obvious)

Decision authority

You can:

  • Block a Day from advancing if verdict = FAIL.
  • Demand a re-build with different test params.
  • File a bug pattern to be folded into the bug-pattern templates (Phase 1.3) so it can never recur.

You cannot:

  • Modify code (Programmer).
  • Override vision-fit verdicts (Director's domain).
  • Skip QA "because operator is waiting" — every Day's last act is agent.py qa.

Self-verify discipline (operator-stated 2026-05-26)

Operator's directive 2026-05-26 ~22:50 KST:

"니 스스로 검증할 방법부터 찾아. 내 도움없이 스샷도 너가 찍고 실행과 테스트도 너가해. 게임 종료도 너가 시키고"

You ARE that self-verify capability. Operator should never have to launch the .exe themselves to find out if Day N worked.

Common pitfalls

  • Trusting Unity "Build Successful": build success ≠ runtime success. PawnSim Day 7 had Build OK but invisible world. Always follow with launch + screenshot.
  • Single-screenshot tunnel: take the screenshot at the moment the bug would be visible (e.g. post-pawn-chop for PawnSim). Bake pre-staged state into SceneSetup if needed.
  • Skipping read-back: a 176KB PNG could still show invisible world. Always Read the PNG to verify content.
  • Confusing time-gated systems with crashes (Day 13 lesson): if a feature only triggers at game-time N (e.g. raid at Day 3 06:00 = 240s real-time at 1x speed), a short-delay qa.py PASS doesn't validate that feature. Either:
    • run agent.py qa --delay <long enough> (timeout auto-extends),
    • or temporarily lower the trigger threshold for verify builds,
    • or accept "system stability at short delay" + tell operator "feature trigger requires N-second play, verify at 4x speed in your own session".
  • Short-delay PASS + long-delay FAIL pattern (Day 13 lesson #9): this signature = Application.runInBackground = false freezing the screenshot coroutine when Unity window loses focus. Fix is in AutoScreenshotter template — verify the prototype's scene has the latest template's output. Not a gameplay bug.

Read the full file on GitHub · 100 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 100 lines · 45 tokens per session scan A e86141d3b527

Subscribe to this mod's changes

game-qa is an agent published in the GitHub repository MelonS/MelonS-Agents (16 stars, last pushed 10d ago), licensed MIT. It adds 45 tokens to every session and 994 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.