Moveset shapeshifting replaces the boss’s whole active package with another authored package: locomotion, targeting, attacks, reactions, and recovery change together. Clear transition rules let one encounter revisit learned opponents without becoming an unreadable bag of moves.
The colossus package signals a heavy stance, releases one expanding slam from 0.55 to 1.35 seconds, and completes its own recovery before transformation.
Autoplay is disabled by reduced motion. Scrub across 0.20, 0.55, 1.35, 1.72, 2.05, 2.35, 3.20, 3.48, 3.82, 4.12, 5.00, and 5.42 seconds.
Define each form’s package id, entry and exit conditions, transition cue, minimum tenure, attack eligibility, geometry, movement, vulnerabilities, cooldown ownership, shared health and status, deterministic sequence, rollback, and telemetry.
Treat locomotion, target selection, attacks, hit reactions, hurtboxes, vulnerabilities, effects, and recovery as one versioned contract. A form may share health and status with the encounter, but it should not inherit incompatible local timers or animation state.
Choose deterministic conditions such as elapsed tenure, health bands, a fixed sequence, or an authored weighted table. Finish or cancel the outgoing action explicitly, signal the handoff, then enable the next package only after its rig, collision, and cues are ready.
Classify every value as encounter-wide or form-local. Preserve health, progression, ownership, and approved status effects; reset or translate cooldowns, targets, spawned entities, weak points, and pending events according to a documented handoff table.
A random attack salad removes the learned relationship between pose, movement, geometry, and response. Keep a package active long enough to read and practice, and reserve cross-package combinations for explicitly authored exceptions.
Old hitboxes, projectiles, targets, or cooldown callbacks can fire after the new rig appears. Give every transition an id, cancel or transfer owned events once, and test backward seeking, latency, death, interruption, reload, and rollback at both boundaries.
A package is more than an attack list. Specify its entry pose, movement model, legal actions, signature pressure, recovery cadence, exit lock, and fallback if an animation or target is unavailable.
One health bar and encounter progress can make the transformations feel like the same boss. Local cooldowns and entities usually belong to their package; translating everything across forms creates invisible dependencies and balance surprises.
Record formId, packageVersion, transitionId, reason, tenure, entry and exit time, action ids, canceled events, carried statuses, damage taken per form, repeat frequency, and failures to reach a valid recovery before handoff.
The elemental widens into a colossus for a stone-shaking slam, uncoils into a serpent’s lane, then raises an oracle crown and casts three rays. Each silhouette completes its own rhythm before the next rune takes hold.
Offer a fixed sequence, longer minimum tenure, stronger transition cues, persistent form icons, and an option that removes one package. Announce geometry through shape, sound, text alternatives, and haptics rather than color alone.
A phase can change encounter goals while keeping the same action system. Shapeshifting replaces the active moveset contract itself. A loadout mirror derives a package from player data; this mechanic selects complete authored packages, often borrowed from other opponents.
From 0:15, Delirium repeatedly becomes bosses from the run and uses their attack patterns. Its documented multipart and timing quirks show why borrowed packages need explicit ownership, cleanup, and bounds rather than raw reuse.
From 0:52, Zant begins a sequence of recreated boss arenas and matching response packages. The long, ordered handoffs preserve the identity of each prior lesson instead of interleaving all of their attacks at once.
At 1:30, Lavos begins cycling through attack modes based on earlier bosses. The turn-based example makes package boundaries explicit and demonstrates how one shared health target can revisit many learned action sets in a deterministic order.
Found an inaccuracy or have a suggestion for this page? Propose an edit on GitHub