combat-design

Combat Design & Game Feel

How to make combat that feels powerful and reads fairly — especially in a high-lethality game where an unreadable hit is a broken promise. Juice sells the hit; telegraphing keeps it fair; encounter structure keeps the fight legible; commitment + a generous retry loop make hard feel fair instead of cheap.

When to use this

• Tuning game feel: juice budgets, hit-stop, screen shake, the impact bundle
• Authoring or auditing telegraphs and a game-wide danger-cue vocabulary
• Designing enemies (silhouettes, combat roles) and multi-enemy encounters
• Building a Souls-like lethality loop: commitment, stamina, comeback mechanics, checkpoints
• Diagnosing combat that feels floaty, noisy, unfair, swarmy, or memorization-gated

Scope

This skill owns moment-to-moment combat and its feel. Adjacent concerns live in sibling skills, cross-linked from the guide:

• General reward/flow/difficulty theory and the "interesting decision" test → game-design-fundamentals
• The single-save consequence model & meta-progression → permadeath-and-lethality
• Loot/itemization that feeds builds, damage tiers as economy → rpg-systems
• Enemy placement in the world, sightlines to a fight → open-world-design
• Generating enemies/encounters within these constraints, and gating them → procedural-generation, procgen-review
• Multiplicative systems that make fights emergent → systemic-emergent-design
• Boss/quest framing → narrative-and-quest-design

How the pieces fit

• GUIDE.md — the cited why, in four sub-domains: game feel/juice (the inverted-U dose curve), readability/telegraphing (fairness as a contract), enemy & encounter design (silhouettes, roles, readable chaos), and Souls-like lethality (commitment, death-as-teacher, comeback, checkpoints). Each rule carries an exemplar, a source, a test-for criterion, the named failure mode, and the procedural/headless implication.
• CHECKLIST.md — Do/Don't + machine-checkable Test-for criteria, grouped by sub-domain. Written to be enforced as automated validators in a generation loop, not just read.

The one idea to anchor on

In a lethal game, telegraphing is fairness, not decoration — and juice has a ceiling. Every attack that can kill must be preceded by a readable, consistent wind-up whose salience scales with its damage ("relative damage must match perceived danger"). And feedback follows an inverted-U: the largest study on the subject (Kao 2020, N=3,018) found Extreme juice hurts experience, motivation, and raw performance as much as None does. Tune to the Medium–High band; never maximize.

Why this matters doubly for a generator: a human team feels an un-telegraphed kill or a juice-overloaded screen as "off." An autonomous generator does not. Hand-author the cue vocabulary, the damage→telegraph curve, the feedback table, and the stamina/checkpoint numbers as inviolable contracts, then let generation recombine vetted parts inside them — and run headless validators (telegraph checker, juice linter, fairness ghost-runner) before any content is committed.

Start with GUIDE.md, then apply CHECKLIST.md.