메뉴
Use when a game "feels off" but you don't know where — bots dominate that shouldn't, prey never gets caught, scores don't match the design, players die invisibly, or one strategy beats every other. Open-ended diagnostic, not a targeted fix.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Use when a game agent (bot, NPC, AI) appears frozen, oscillates between two adjacent tiles, "moves back and forth without making progress", gets stuck near walls, or stops cooperating partway through a session. Symptom is "stopped doing what it should" rather than "doing the wrong thing".
Use when iterating on stag_hunt — tuning prey rewards, flee probabilities, capture coalition sizes, energy economy, or testing a server-side mechanic change. Specific to the stag_hunt game in this repo; for the general process see investigating-game-balance and testing-game-balance-change.
Use when you (or the user) have a specific proposed change to a game tunable — "lower this prob 40 → 30", "give rabbit less energy", "make moose flee more at distance 2" — and you need to confirm it has the intended effect without breaking something else.
| name | investigating-game-balance |
| description | Use when a game "feels off" but you don't know where — bots dominate that shouldn't, prey never gets caught, scores don't match the design, players die invisibly, or one strategy beats every other. Open-ended diagnostic, not a targeted fix. |
Symptoms that should trigger this:
If the user has already named a specific tunable to change ("lower the moose gut prob"), skip this skill and use testing-game-balance-change instead.
Balance bugs hide behind aggregate numbers. A "total captures" count can stay flat while the actual failure mode shifts. You diagnose by spreading the data across three axes simultaneously: bot type, prey kind, and time within the round. The pattern jumps out when you see all three at once.
| Pattern across the data | Likely cause |
|---|---|
| Captures = 0 AND end-energies are 0-1 | Bots are starving, not failing to find prey |
| Captures = 0 AND end-energies are healthy | Bots can't see / can't reach / can't coordinate |
| One bot has score 0 every round | That bot's strategy isn't viable in this roster, OR it's broken |
| Score drops sharply round-over-round | State leaking between rounds (reset bug) |
| One prey kind is never caught | Spawn rate, flee mechanic, or coalition requirement is too tight |
| One prey kind is caught 10× more than expected | That prey is a free lunch — too high reward for the effort |
| Self-play succeeds but mixed-play fails | Cooperation depends on shared assumptions broken by other policies |
| Energies stay high but no captures | Bots resting too defensively — threshold too high |
For each scenario, record a single line. Aim for legibility, not completeness:
roster=4×moose_hunter ticks=1800 rounds=3 seed=1
captures: Moose 5, Rabbit 4, Stag 1
scores : [30, 0, 30, 30] / [0, 0, 0, 0] / [21, 25, 21, 7]
end-energies: [44,30,25,50] / [41,40,32,29] / [37,36,32,40]
→ moose hunting works, round 2 was a wipe (no captures, energy still OK = bots couldn't find a moose)
The "→" line is the whole point. If you can't write one, you didn't look hard enough at the cross-references.
Stop when you can complete this sentence:
"The problem is X, and the smallest change I can make to test a fix is Y."
Then switch to testing-game-balance-change to validate Y.
If you can't finish that sentence, run more scenarios. Don't guess and tweak. A guess masquerading as a fix wastes the next investigation round.
In stag_hunt, a 60s self-play eval of elephant_hunter showed 1 capture/round and looked healthy. An 8-hour interactive session showed the same bots had 162 tramples vs 1 capture — they were starving inside long rounds, getting hit and unable to recover. Three layers told the story:
Without the end-energy snapshot, the score line "29" would have read as "fine" rather than "29 from one good moment, then died." Same lesson everywhere: capture the resource state, not just the result.
testing-game-balance-change — once you know what to changedebugging-stuck-game-agents — when "captures = 0, energies = 0-1" appears./balance_sweep.sh runs a fixed grid of scenarios with the right output shape.