Skip to content

gamedev-skills/awesome-gamedev-agent-skills / skills/genres/tower-defense

tower-defense

Verified

Designs tower defense games in any engine: lane and maze pathing, wave pacing, targeting priorities, armoured and flying counters, and the gold economy, with a wave-curve checker. Use for a Bloons- or Kingdom Rush-like, or late waves that are impossible or trivial or a dominant tower. Not for pathfinding (`game-ai`) or survival base building (`survival-crafting`).

Tower defense

Owns the TD loop: paths, waves, towers, targeting and the economy that links them. The core balance question is whether the DPS the player can afford keeps pace with the HP each wave sends, so design both curves together and check them with the script before playtesting.

When to use

  • Towers attack waves on their own, the player spends earned currency to build, upgrade and sell, and leaks cost lives. Includes lane defense, mazing TD and TD-action hybrids.
  • Balancing waves and economy, choosing targeting modes, adding enemy types, fixing pathing.

When not to use: the player shoots directly (fps-shooter); free-form base building with needs (survival-crafting); the A* or flow-field code itself (game-ai).

Workflow

  1. Pick the map model: fixed lanes (most TDs; full control of pacing), maze (towers shape the path; needs placement validation) or open field with flow fields (large crowds).
  2. Write towers, enemies and waves as data. Each tower has a role (single-target DPS, splash, slow, anti-air, armour pierce, support) and each enemy type exists to counter one lazy build: fast (beats slow single-target), armoured (beats rapid low-damage), flying (skips ground-only), swarm (beats single-target), boss (HP spike), regenerating or shielded (punish slow DPS).
  3. Set the curves: enemy HP per wave grows geometrically (start near 1.12-1.18x per wave) and income grows from bounty growth plus enemy count. The two compound against each other, so small changes matter: in the script's example, HP growth 1.15 holds for 30 waves while 1.25 walls by wave 2.
  4. Run the curve check (path relative to this skill folder): python3 scripts/td_curve_check.py waves.json. It buys a reference tower with all gold before each wave and prints each wave's pressure (needed DPS / owned DPS x efficiency). Errors are walls (pressure above 1.0); warnings are runs of trivial waves. Aim for 0.6-0.95 with breathers between spikes. --help shows the spec format; --json for tooling.
  5. Pace the session: a short build phase between waves, a next-wave preview, an early-call bonus for confident players and a 2-3x fast-forward toggle.
  6. Check it. Run the curve check and these tests, then report the numbers:
    • a fast enemy (speed x 10) on a path with sharp corners keeps its total travel time within one tick of path_length / speed (refs §1);
    • maze maps: every placement that would cut the spawn or any walking enemy off from the goal is refused (refs §1);
    • targeting "first" picks the enemy with the most path progress even when another is closer to the tower.

Economy rules

LeverStart hereWhy
Wave 1beatable with starting gold at pressure about 0.5-0.7the first wave teaches, it does not test
Kill bountymost income early; wave-clear bonus grows laterlate income should reward survival, not farming
Leakscost lives and the bountyotherwise leaking can be optimal
Sell refund50-80% of money spentrepositioning possible, never free
Interestoptional, cappedrewards saving but snowballs
Upgradescompetitive gold-per-DPS with new towers, with diminishing tiers"tall" and "wide" both viable
Early callgold bonus scaled by time skippedspeeds the game for experts
Targetingper tower: first (default), last, strongest, weakest, closestchanges play more than raw stats

Pattern 1: wave data

# A wave is data: groups spawn in order, each with its own spacing.
WAVES = [
    {"groups": [{"type": "grunt", "count": 10, "spacing": 0.6},
                {"type": "runner", "count": 6, "spacing": 0.3}], "clear_bonus": 30},
    {"groups": [{"type": "armored", "count": 4, "spacing": 1.2}], "clear_bonus": 35,
     "preview": "armored"},                     # shown one wave ahead in the UI
]

Telegraph every new enemy type a wave ahead, and introduce it alone before mixing it with others, so the counter feels learnable rather than random.

Gotchas

  • Late game impossible or trivial → HP growth and income growth set separately. Tune them together with the curve check; geometric HP needs geometric income or upgrade efficiency.
  • One tower wins everything → no enemy counters it. Add the counter type (step 2) and diminishing upgrade returns.
  • Enemies stuck, or the goal unreachable → a maze placement severed the path. Validate with one search from the goal before committing (refs §1).
  • Fast enemies slow down at corners or overshoot → movement stops at each waypoint. Carry the leftover distance (refs §1).
  • Towers ignore the enemy about to leak → "closest" used as the default. Default to "first" by path progress.
  • Wave timers drift on slow machines → spawn spacing counted in frames. Use the game clock; fast-forward multiplies the clock, not the frame rate.
  • Specials feel unfair → no preview. Show the next wave's composition.
  • Monotonic ramp exhausts players → every wave harder than the last. Spike, breather, spike.

References

  • references/balancing.md: read when choosing path representations and writing waypoint or maze validation code, targeting mode code, DPS and "can this lane hold" math, HP scaling, economy details, lives and wave pacing.
  • scripts/td_curve_check.py --help: spec format and the pressure model's limits.

Related: game-ai (A*, flow fields), level-design (map layouts), godot-tilemap or unity-tilemap-2d (grid maps), godot-resources or unity-scriptableobjects (tower and wave data), game-ui-ux (build menu, wave preview, HUD), game-feel and audio-design.

Version notes

Last verified: 2026-11-11.

  • Godot 4.7 Path2D and PathFollow2D (progress, progress_ratio, loop) and AStarGrid2D.set_point_solid() (no update() needed afterwards) used in the reference were checked against the 4.7.2 class reference (doc/classes/*.xml); PathFollow2D.loop defaults to true, so set it to false for leak detection.
  • Curve figures (1.15 holds, 1.25 walls) come from running the bundled checker on its test spec, not from a shipped game.
brainstormingobra/superpowersYou MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation).Stars: 322kUpdated Claude CodeCodexVerifiedInstall
brand-guidelinesanthropics/skillsApplies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
frontend-designanthropics/skillsGuidance for distinctive, intentional visual design when building new UI or reshaping an existing one. Helps with aesthetic direction, typography, and making choices that don't read as templated defaults.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
mcp-builderanthropics/skillsGuide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK).Stars: 198kUpdated Claude CodeCodexVerifiedInstall