The filter 'Pistol' now matches 'Braises d'Automne (pistolet ?)' via
case-insensitive substring matching, so the feat is no longer dead.
Golden test_tammara_vs_goblin updated: round-4 pistol shot gains the +1
feat bonus (total 15 -> 16 vs AC 16), turning the MISS into a lethal HIT
(17 damage, goblin 6 -> -11); the goblin dies on round 4 so the old
rounds 4-10 exchange and the draw are gone.
SRD Cure Light Wounds: 'Since undead are powered by negative energy,
this spell deals damage to them instead of curing their wounds. An
undead creature can apply spell resistance, and can attempt a Will save
to take half damage.'
- _apply_heal_effect takes the rolled save; an undead target takes the
rolled healing as damage (halved on a successful save; SR already
resolved upstream, DR never applies to spell energy, no bleeding stop
or stabilization). Living targets are bit-identical.
- cure_light/cure_moderate descriptions fixed: the SRD save line is
'Will half (harmless)', not 'negates'.
- Deafened 20% verbal spell failure DEFERRED to the README « Limites
hors perimetre »: SpellSpec carries no components data (would need a
verbal flag on all 108 spells) and sound-based Perception is not
modeled either.
SRD spells/mage-armor.md: 'An invisible but tangible field of force
surrounds the subject of a mage armor spell, providing a +4 armor bonus
to AC' - 'School conjuration (creation) [force]', level 1, 'Casting Time
1 standard action', 'Range touch', 'Duration 1 hour/level (D)',
'Spell Resistance no'.
- data/spells/mage_armor.json: buff +4 armor to AC, 600 rounds (1
hour/level modeled fixed, the shield.json convention). Description
documents the known simplification: the engine bakes worn armor into
the AC total and cannot see it, so the armor bonus stacks additively
with that baked value (two mage armors still do not stack).
- effects.py: 'armor' added to BonusType; absent from _STACKING_TYPES it
behaves as a keep-the-highest type (verified by resolve_modifiers
tests).
- Spell-count pins 107 -> 108 (test_runner_spells, batch3, test_spells)
and the README references.
- Golden transcript untouched (buff branch only fires when no offensive
spell is selected; the lich still opens with Magic Missile).
Verified against the KB (bestiary/lich.md, Bestiary Lich CR 12):
- bab 8 -> 5 ('Base Atk +5')
- Paralyzing Touch attack_bonus 10 -> 5 ('Melee touch +5 (1d8+5 plus
paralyzing touch)') and marked natural: true (template: 'A lich has a
touch attack that it can use once per round as a natural weapon [...]
treated as a primary natural weapon' - no BAB iteratives per e5691bd)
- damage_bonus 5 -> 0: the +5 (1 point per 2 HD) is already in the
1d8+5 formula and _resolve_swing adds damage_bonus on top (double
counting: the old data dealt 1d8+10)
- ac 25/14/23 -> 23/14/21 ('AC 23, touch 14, flat-footed 21'; touch was
already correct)
- saves 10/10/14 -> 6/7/11 ('Fort +6, Ref +7, Will +11')
- con_score 14 -> 10: 'Being undead, a lich has no Constitution score'
(Con '—'); the int-only schema encodes 'no score' as the neutral 10
(modifier +0), inert anyway since undead are destroyed at hp <= 0.
Left as intentional engine tuning (residual divergences, documented in
the final report): hp 100 vs 111, Str 12 vs 10, Int 20 vs 22,
initiative +6 vs +2, fear aura 30 ft/DC 22 vs 60 ft/DC 18.
Golden transcript untouched (the lich wins via Magic Missile before any
weapon swing; MM DC depends on Int, unchanged). Pinned a new
test_json_matches_srd_stat_block.
SRD Moving Through a Square: 'Any creature can move through a square
occupied by a creature three size categories larger than itself. A big
creature can move through a square occupied by a creature three size
categories smaller than it is. Creatures moving through squares occupied
by other creatures provoke attacks of opportunity from those creatures.'
- _can_pass_through: size ranks differ by >= 3 opens the through-gate
(_blocked_hostile) while _blocked/_cells_free keep the end-gate
('You can't end your movement in the same square as another creature
unless it is helpless').
- _check_enter_aoo wired into _walk_steps (move, explicit-destination
move, and run all commit steps there): the occupant strikes once per
fresh entry, capped by the per-round allowance; a dropped mover stops
the walk. S4c co-occupiable pairs (Tiny-or-smaller) keep no-AoO sharing.
- _has_free_descent generalized to a descending BFS so a walk can cross
a multi-cell Huge+ creature (one-step lookahead dead-ended inside).
- Golden transcript untouched (all 1480 tests green).
A 5-foot step is hampered by difficult terrain: stepping out of OR into a
difficult cell is blocked. _step5 (auto-directed) logs 'wait (cannot step)';
_step5_to (explicit destination) rejects with unreachable_destination.
Engine-level tests cover destination, auto-path, and origin cases.
SRD Conditions Fatigued: A fatigued character can neither run nor
charge. _execute_charge already checked cant_charge; _execute_run did
not -- a fatigued/exhausted/entangled combatant could still run.
Added cant_charge guard to _execute_run (mirrors _execute_charge),
logging wait (cannot run) and returning ok. Two engine-level tests:
fatigued cannot run, exhausted cannot run.
The greedy move/run/flee/charge descent helpers hardcoded diagonal parity 0,
letting a speed-30 walker cover 6 diagonal cells (cell-count budget). Under
5-10-5 (SRD Movement, Tactical) the same walker covers only 4 cells
(1+2+1+2=6 squares).
Threading the running diagonal count:
- _best_move_step / _best_flee_step: accept diagonals_used param
- _descend_path / _run_line_path / _flee_path: track and pass parity
- _charge_end: use _path_cost_5105() instead of len(path) for max check
- _path_cost_5105: new helper; computes total 5-10-5 cost of any path
Updated golden paths in test_aoo_limit_one_per_round (5-cell path now
takes an orthogonal detour to avoid expensive 2nd diagonal), test_run_*
(speed-30 diagonal move stops at 4 cells), and staggered_withdraw
(single-speed diagonal withdraw stops at 4 cells).
PF1e (CRB "Natural Attacks"):
- Natural weapons (natural=true) get exactly one swing per full attack,
regardless of BAB (no BAB iteratives for natural attacks).
- Secondary natural attacks (secondary=true) attack at BAB -5; the
Multiattack feat reduces this to -2.
- Manifested in combat.py (_full_attack_weapon, _resolve_swing),
models.py (AttackSpec.natural/secondary), 6 data JSONs, 4 new tests,
and README documentation.
1461 passed, ruff clean, basedpyright 0 errors.
SRD Undead/Construct traits: 'immediately destroyed when reduced to 0 hit
points or fewer' — undead and constructs have no Constitution score, so
the PF1e hp <= -CON death rule does not apply to them.
- combat.py: CombatantState.death_threshold returns 0 for creature_type
undead/construct (in addition to the opt-in dies_at_zero flag)
- tests: TestUndeadDeathThreshold pins undead dead at 0 and below, living
keeps the disabled band at 0 and the dying band down to -CON
- golden unchanged: tammara-vs-lich has the lich winning in 1 round
without taking damage
SRD spells/bane.md: enemies within a 50-ft burst centered on the caster
take a -1 penalty on attack rolls and a -1 penalty on saving throws
against fear effects; Will negates. The -1 penalty is approximated with
the shaken condition (-2), matching the existing Doom approximation.
- bane.json: single-target condition -> AreaEffect burst radius_ft 50,
targets "enemies", ConditionEffect shaken 10 rounds save_negates
(range/save kept as medium/will for the registry pin)
- combat.py: _metamagic_transform_sub now copies AreaEffect.targets so
an Extend Spell on an area spell preserves the allies/enemies filter
- tests: TestBaneEnemiesOnly (enemy in burst debuffed, allies/caster
unaffected, successful Will save negates)
SRD spells/bless.md: 'The caster and all allies within a 50-ft. burst,
centered on the caster'. Add an AreaEffect.targets filter (all/allies/
enemies) applied where area creatures are collected; bless.json now
declares radius_ft 50 and targets allies (the caster included).
Bull's Strength now grants the SRD +4 enhancement bonus to Strength as
+2 melee attack, +2 melee damage and +2 CMB (three modifiers instead of
the previous +1/+1 approximation). Shield now uses a real shield bonus
type (+4 AC) instead of approximating it as deflection: the engine's
BonusType literal and stacking docs gain the shield type, which stays
out of the CMD bonus list (SRD Combat Maneuver Defense). Bear's
Endurance keeps its +2 Fortitude / +2 hp per HD and its description now
cites the SRD and flags the hp modifier as informational. Golden
transcripts updated: the buff log reports 3 mods and Tammara's pistol
attacks gain the extra +1 enhancement (d20=10+9-6=15).
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
PF1e (SRD spells/cure-light-wounds.md): 'cures 1d8 points of damage +1
point per caster level (maximum +5)'; (spells/cure-moderate-wounds.md):
'cures 2d8 points of damage +1 point per caster level (maximum +10)'.
A HealEffect now carries the same DamageFlatScaling spec as damage
effects (S1b) and resolves it through effective_formula(caster_level)
at cast time, so the flat bonus grows with the caster and stops at the
stated cap. The engine passes the caster's level into _apply_heal_effect
(required keyword: no silent level-0 under-heal); Maximize Spell
maximizes the scaled formula. Cure Light/Moderate Wounds ship the SRD
values; inflict_light_wounds already carried its scaling_flat.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
Ultimate Combat Firearm Rules: a firearm attack rolling a natural 1 gains the
broken condition (weapon_broken, logged MISFIRE). Quick Clear now actually
clears the first misfired held firearm for 1 grit (reported no-op without
grit), and occupies the standard budget slot via _effective_slot; the
swift-consumed filter is slot-based so the standard deed survives.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
SRD Ready - Initiative Consequences of Readying: 'your initiative result becomes the count on which you took the readied action' and you 'act immediately ahead of the character whose action triggered your readied action'. _check_readied fired the payload off-turn but never touched the readier's initiative. It now calls _ready_shift_initiative, which rewrites the readier's _initiative_keys entry (trigger total + 0.5, keeping its mod and prior index) using the same tuple shape _execute_delay writes, so _build_round_order holds the new position for all following rounds. Deterministic: no RNG consumed, current round order untouched.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
SRD Full-Round Actions - Run: 'you can move up to four times your speed in a straight line'. _execute_run reused the Dijkstra move machinery, so it turned corners around walls. It now walks the straight Bresenham line (_run_line_path) and stops at the first obstacle (impassable terrain, difficult terrain, or a blocking creature), mirroring the charge clear-path logic. Keeps the 4x budget, step-by-step AoO provocation, moved_this_turn and the running/Dex-loss flag; a destination the straight line cannot honour is rejected (unreachable_destination).
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
SRD Casting Time: a spell that takes 1 round is a full-round action
that comes into effect just before the caster's turn in the next round;
the caster then acts normally. _cast_spell now spends the slot and runs
the start-of-casting gates, then records a PendingCast (spell, target,
center, metamagic, slot level, hp at cast); delivery (touch attack, SR,
saves, effects) resolves at the start of the caster's next turn, before
the policy runs. Net hp lost in between forces the injured-while-casting
concentration check (DC 10 + damage + spell level, no Combat Casting) or
the spell fizzles; a caster knocked out of the fight loses it. Exposed
in the observation via pending_cast. Bounded model documented: net-hp
measurement instead of a per-damage-site hook; long (1 minute+) casting
stays out of scope.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
SRD Surprise: when some but not all combatants are aware of their
opponents, a surprise round precedes round 1. Aware combatants take a
single standard OR move action plus free actions (swift and full-round
are dropped, full_attack collapses to one standard attack, withdraw
survives as the Restricted Withdraw standard); unaware combatants skip
their turn. New Combatant.aware flag (default true, parsed at the JSON
boundary) keeps every existing encounter bit-identical, exposed in the
observation snapshot. Unaware combatants are not flagged flat-footed
(documented simplification).
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
SpellSpec gains spell_resistance: bool = True (JSON 'spell_resistance'). When false, _spell_delivery_blocked skips the targeted SR check and _apply_area_effect skips the per-creature SR check (threaded through the effect chain). Verified each imported spell against the local KB and set false on the 28 whose SRD block says 'Spell Resistance: no' (Acid Arrow, Flurry of Snowballs, Web, Darkness, Sanctuary, the Protection from X spells, etc.). Added no-SR targeted/area tests and a high-SR control in test_spell_resistance.py, and the README.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
ConditionEffect gains save_partial (JSON 'on_success': condition_name + duration_rounds). On a successful save that would otherwise negate, _apply_condition_effect now imposes the weaker condition (SRD fear.md: panicked on a failed Will save, shaken for 1 round on a success). Extended _metamagic_transform_sub to preserve/double it; extracted _impose_condition helper. Updated fear.json, added failed/successful-save and loader tests in test_spells.py, and the README.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
Acid Arrow (SRD spells/acid-arrow.md) deals 2d4 acid, then 2d4 again at the start of each of the target's turns for 1 round + 1 per three caster levels (max 6). DamageEffect gains a RecurringDamageSpec (JSON 'recurring'); a landed hit attaches a RecurringDamage source to CombatantState.recurring_damage, ticked in _turn_gate (after bleed), reduced by energy resistance/immunity but never DR (spell damage), cleared at expiry, and exposed in the observation snapshot. Extracted _resolve_one_projectile to keep _apply_damage_effect under the complexity ceiling. Updated acid_arrow.json, added tests/test_recurring_damage.py, and the README.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
SRD Magic Missile fires one missile +1 per two caster levels beyond 1st (max 5, 1d4+1 force, auto-hit); Scorching Ray fires one ray +1 per four levels beyond 3rd (max 3, 4d6 fire, ranged touch each). DamageEffect gains a ProjectileScaling spec (JSON 'projectiles'); _apply_damage_effect loops count_for(caster_level) projectiles, each rolling its own damage and (for a declared-projectile ray) its own ranged touch attack, with the target-drops-mid-sequence convention losing the rest. Updated magic_missile.json/scorching_ray.json, the golden lich transcript (CL 11 now fires 5 missiles, Tammara dies round 1: 43 damage), the scripted MM queues in test_spells/test_conformance_spellflow, added tests/test_multi_projectile.py, and the README.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
CRB Magic - Aiming a Spell: a line 'starts from any corner of your square', so the caster's own square is outside the template (SRD Lightning Bolt: 'The bolt begins at your fingertips' - it leaves the square, it does not fill it). line_cells now drops the origin and takes the length_cells squares beyond it; a degenerate aim affects nothing. Updated the Lightning Bolt and Line Sparks tests (caster no longer caught) and the README.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
SRD-verified area templates: Burning Hands 15-ft cone, Color Spray 15-ft cone, Fear 30-ft cone, Flurry of Snowballs 30-ft cone, Lightning Bolt 120-ft line (all converted from single-target/burst approximations to proper AreaEffect shapes); Fireball 20-ft burst, Sound Burst 10-ft spread, Sleep 10-ft burst, Web 20-ft spread kept. Close-range spells note the 15/30-ft RAW ranges as an approximation in their descriptions. README: cone/scaling limitations moved out of the limitations list, new area-shape + caster-level scaling rules documented, areas.py added to the architecture.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
DamageEffect gains optional scaling (1dX per N levels, capped) and scaling_flat (+X per N levels, capped) specs; effective_formula resolves them against the caster's level at cast time, so scaled dice flow through crits, energy resistance and metamagic like the base formula (CRB Magic - Caster Level). Loader rejects mismatched die types. Data: Fireball/Lightning Bolt 1d6/lvl max 10d6, Burning Hands 1d4/lvl max 5d4, Shocking Grasp 1d6/lvl max 5d6, Inflict Light Wounds +1/lvl max +5 (SRD-verified). Cure Light Wounds stays flat: its golden transcript pins 1d8+1 and HealEffect scaling is out of scope.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
AreaEffect gains shape: burst|cone|line (CRB Magic - Aiming a Spell). Cones (90 deg, length = radius_ft) and lines (5 ft wide, Bresenham ray) originate at the caster's cell and are aimed at the resolved target or the U7b point-cast center; bursts are unchanged. Shared per-creature save + SR application; the observation area dict already exposes the shape.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
SRD 'Cover': 'To determine whether your target has cover from your ranged attack, choose a corner of your square.' The ranged scorer in cover_bonus_cells took the MAX over all attacker corners (worst corner); the cover gate has_cover_cells and RAW use the attacker's best corner. Ranged cells are now scored min over attacker corners of (max over target-corner lines), then min over attacker cells - the bonus can no longer exceed what the gate admits. Melee keeps the worst-corner rule ('any line from ANY corner of your square'). Updated the two pins that locked the old worst-corner scores.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
SRD 'Grapple': 'If you successfully grapple a creature that is not adjacent to you, move that creature to an adjacent open space (if no space is available, your grapple fails).' _maneuver_grapple now pulls a non-adjacent target to the nearest free anchor whose footprint is adjacent to the grappler (5-10-5 distance, (row,col) tie-break, new _nearest_anchor_adjacent_to helper) and fails outright when no adjacent open space exists. Three tests: pull, no-space failure, adjacent no-move.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
ctan