Convert and verify all melee/ranged/magic special attacks

[jelqscape]

Summary

Converts and verifies every melee/ranged/magic special attack (74 weapons) onto the current engine, plus every shared subsystem several of them need to function at all (weapon poison, status effects, charge-weapon tracking, stun/bind, PvP area attacks). A large number of these weapons also had no working normal attack at all - broken or entirely missing outside of their special - and that's fixed here too, listed separately below since it's a genuinely different body of work from the special attacks themselves.

Special attacks

All 74 melee/ranged/magic specials in content/other/special-attacks, each verified against wiki-documented mechanics (damage formulas, accuracy multipliers, timing) rather than ported blind from any single source.

Weapons that needed normal-attack work, not just their special

A special attack being correct doesn't mean the weapon could throw a regular punch - several of these had no normal-attack registration at all, or one that silently fell back to unarmed combat because the cache was missing the params WeaponAttackManager needs to read:

• Toxic/Rosewood blowpipe - normal attack was entirely missing. Both store their own darts (and, for the toxic family, Zulrah's scales) packed inside the weapon itself rather than in the quiver, so the generic cache-driven ranged fallback (which only knows about quiver ammo) could never fire them at all - not a special-attack problem, a missing-primitive problem (BlowpipeAmmo). Wiki-verified envenom/scale-consumption chances included.
• Webweaver bow - entirely unregistered for normal attacks; the cache's chargebow category refuses to fire without a WeaponMap, so it just silently did nothing. Added the normal attack, its own ammo-generation (revenant ether), and its Wilderness passive (+50% accuracy/damage vs NPCs).
• Craw's bow - had zero implementation before this, not even a stub: no normal attack, no charging (activate/top-up/check/uncharge), nothing. Built from scratch, mirroring the Webweaver bow it upgrades into (same activation cost, max ammo, Wilderness passive).
• Tumeken's shadow - normal attack, uses the same shared ObjChargeManager charge primitive as the two bows above.
• Noxious halberd - had no normal-attack registration at all before this fix (only discovered live-testing): the envenom chance, and consuming the special attack's Virulence minimum-hit buff, which the special was already setting correctly but nothing was ever reading.
• Fang of the Hound, Crimson kisten, Sunspear, Thunder khopesh - each missing weaponCategory/attack_anim_stance*/attack_sound_stance* cache params entirely, so the client's combat-style tab and the swing animation both fell back to generic unarmed punch/kick. Cache overrides added where the client needs them (weaponCategory), animations played directly where the engine's generic cache-param lookup had nothing to read. Thunder khopesh and Fang of the Hound also needed their real passives added (delayed lightning bolt; Flames of Cerberus proc).
• Dogsword - shares the real Bandos/Saradomin/Zamorak godsword animation IDs, but this cache revision never gave them named aliases, so they have to be referenced by raw ID instead of the normal string param.
• Soulreaper axe - Soul Stack generation/cap/decay logic (a stack builds on every attack including misses, feeds a Strength bonus, and decays on inactivity).
• Dual macuahuitl, Scythe of Vitur, Glacial temotli/Sulphur blades/Earthbound tecpatl/Barrows' Torag's hammer (shared multi-hit family) - normal-attack implementations that didn't exist before.
• Trailblazer/Trailblazer Reloaded tools (all variants) - Wield sat on cache op1, which this engine's default equip handling only fires on op2, so the item could never be worn at all. Fixed by delegating op1 to the same equip entry point op2 uses internally.
• Belle's folly - had no weapon definition whatsoever (no weaponCategory, no animations/sounds); wiki confirms its attack animation is identical to the Ghrazi rapier, so it reuses that entry's exact data.
• Beyond the above, items.toml got real combat data (weaponCategory, attack animations, ammo categories) filled in for a broader set of weapons that had a working special attack but broken/missing normal-attack data: Bounty Hunter corrupted dragon weapons, Deadman mode variants, Dinh's bulwark ornament, Emberlight, and others.

Shared subsystems (new, used by multiple weapons each)

• Melee + ranged weapon poison (WeaponPoisonEffect) - didn't exist at all before; (p)/(p+)/(p++) weapons were silently doing nothing.
• Burn/venom/disease status effects (BurnEffectService, NpcPoisonEffectService, toxins package)
• BlowpipeAmmo - dart/scale charge tracking for the toxic blowpipe family
• ObjChargeManager - shared charge/uncharge primitive for ether/charge weapons (Webweaver bow, Craw's bow, Tumeken's shadow)
• BindEffectService - shove-stun/freeze-bind, shared by Dragon hasta's Shove and the godsword freeze specials
• PvPAreaAttackManager, ProtectionPrayerLockout, BypassProtectionPrayerPlayerHitModifier, PowerOfDeathMeleeProtection - PvP-specific combat rules several specials need

Real bugs fixed along the way

Not special-attack-specific, but found and fixed because a special attack or weapon exercised the broken path for the first time:

• ObjChargeManager was using the wrong RSCM type for 3 of its 4 functions (crashed any real caller, including Tumeken's shadow) - fixed at the source, twice: once for the RSCM lookup, once for a "0 charges" edge case that broke topping up an already-activated Webweaver bow.
• PlayerStatExtensions.statHeal crashed on an already-overhealed target (coerceIn given an empty range) - a shared fix covering every heal-based special, not just the one that surfaced it.
• Zulrah's scales are also a Herblore ingredient; a generic "any herblore ingredient used on anything" catch-all was silently swallowing "use scales on blowpipe/Webweaver bow" before dispatch could try the click order that actually works. Excluded scales from that catch-all.
• An uncaught exception from an unexpected inventory-transaction result disconnected the player instead of just failing the action - hardened to fail cleanly with a message.
• Dogsword's freeze effect was player-only, silently no-oping on NPCs despite the wiki confirming it's a real PvM tool.
• Morrigan's javelin was missing its attack range entirely (fell back to melee range).
• Nightmare staff (Volatile/Eldritch): resolveCombatAttack classifies the staff's normal attack as melee (its weaponCategory is the plain melee Staff, not PoweredStaff), so the special-activation dispatch always tried the melee path, found the weapon's registered magic special, rejected it as the wrong type, and silently cleared the special-attack toggle and threw a normal hit instead - with no error, so it looked "implemented" from reading the code alone. Same failure existed a second way: autocasting a spell with the staff resolves to CombatAttack.Spell, whose attack path never checked the special-attack toggle at all. Both paths now redirect into the magic special correctly. Only found via live testing - a pure damage-formula unit test can't catch a dispatch/classification bug like this, see "Testing" below.

Testing

Most of this PR was verified live, in-game, weapon by weapon: correct animation/spotanim/sound, correct damage range against wiki-documented worked examples, correct timing (hit delays, projectile sync), and no crashes across repeated server restarts through every change. A small number of weapons got comparatively lighter live testing - see "Known limitations" below - and the Nightmare staff bug above is a concrete example of why live testing mattered here: it read as fully implemented and correctly registered from the code alone, and only broke visibly once actually swung in-game.

This branch also carried unit tests during development, covering the pure damage-formula math only (max hit scaling, multiplier stacking, etc.) checked against wiki worked examples independent of live dice rolls. They were useful for that narrow slice - catching formula regressions without needing a live server - but they exercise the damage functions directly and never touch special-attack registration, dispatch, or the actual combat-type classification a real attack goes through. The Nightmare staff bug above lived entirely in that untested gap: the damage formula was correct and would have passed its unit test every time, while the special never fired in-game at all. Deliberately not included in this PR's diff - the formulas themselves are already re-verified live, and the priority here is the game content, not test scaffolding - but worth naming explicitly so reviewers know what "tested" does and doesn't cover here.

Known limitations

• A meaningful number of special-attack sound ids were sourced from a third-party reference implementation rather than the OSRS wiki, since the wiki only documents sound data for a subset of items. Those are lower-confidence than the wiki-sourced ones and could theoretically be wrong for a given weapon - flagged in code as unaliased raw ids, easy to grep for and revisit.
• Two weapons (Brine sabre, Vampyre flail) had comparatively light live testing versus the rest of the pack, since Brine sabre requires being underwater and Vampyre flail requires a vampyre target to see its full effect - both have since been spot-checked, but with less repetition than the others.

🤖 Generated with Claude Code