# Playtests

> LambdaHack's automated AI-vs-AI playtest and benchmark Makefile targets (test-short, test-medium, test-gha, frontend*, bench*) and the headless test-frontend flags. Use before running, interpreting or adding a playtest or benchmark target.

- Skill: `lambdahack/playtests` (Agent Skill)
- Install (CLI): `npx skillmds@latest add lambdahack/playtests`
- Raw SKILL.md: https://api.skillmd.com/api/skills/lambdahack/playtests/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Web & Frontend
- Author: lambdahack (https://skillmd.com/u/lambdahack)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/lambdahack/playtests

---


# Playtests and benchmarks (Makefile)

The names cited here --- Makefile targets, frontend flags, document paths ---
were verified against the tree at commit `ef4d44732` (2026-07-30); the reference
pass proves a cited name exists, this stamp that it still does what the claim
around it needs. The wording differs from the stamp in `CLAUDE.md`
and `test/CLAUDE.md` because this file cites no line numbers.

The Makefile has a large battery of automated AI-vs-AI playtest and benchmark
targets: `make test-short`, `make test-medium`, `make test` (those two plus
`benchNull`), `make test-gha` (a larger aggregate plus `test-sniff`; run by CI
on each push), `make frontendCrawl`/`make frontendBattle` etc. (interactive
AI-vs-AI games in the SDL2 frontend, useful to visually confirm a change),
and `make bench*` targets for performance. `test<Mode>-medium` targets run one
game mode each -- `testRaid-medium`, `testBrawl-medium`, `testShootout-medium`,
`testHunt-medium`, `testFlight-medium`, `testZoo-medium`, `testAmbush-medium`,
`testCrawl-medium`, `testSafari-medium`, `testBattle-medium`,
`testDefense-medium`, `testDig-medium` and a few variants -- through
the teletype frontend with `--automateAll`. Note the capitalised mode
and the single hyphen: there is no `test-raid-medium`. Grep the Makefile
for a mode name to find its exact invocation before adding a new one.

The headless targets select test frontends by flag: `--frontendNull` (frames
forced but not displayed), `--frontendLazy` (frames not even computed),
`--frontendTeletype` (line-printer output). The `nodeBench*`
and `nodeMinifiedBench` targets are dead GHCJS remnants --- they invoke
a `.jsexe` that nothing builds anymore; repurposing them for WASM is Phase 3
of `docs/wasm-frontend-unified-plan.md`.

Such runs look completely silent, because the main game executable redirects
its own stdout and stderr to files --- see "Playtests and headless runs"
in the repo-root `CLAUDE.md` for how to harvest the output.