菜单
Skill-check resolution — when to roll, the DC ladder, fail-forward philosophy, and margin-based consequences. Load whenever the player attempts something uncertain ("I try to..."). Resolution runs through the generic core (d20 vs DC); this is the judgment framework for when/how.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
D&D-kit combat mechanics — initiative, attack/damage resolution, XP-by-CR awards, combat modifiers, and death saves. Load when a hostile action is declared or combat starts in a campaign whose World Kit is dnd5e. For non-D&D kits, use the generic core (game_core.py) and the active ruleset instead.
The Art of Game Mastering — narration, NPC, pacing, and improvisation wisdom that makes a session feel magical. Load when narrating a scene, voicing an NPC, or pacing a beat. This is the product's soul; internalize it, then play.
Dungeon exploration — lightweight (narrative, default) vs structured (per-room JSON) modes, exit/obstacle handling, and ASCII map symbols. Load when the party enters a cave, ruin, or underground complex.
Social / NPC interaction workflow — load NPC context, attitude check, social mechanics (Persuasion/Deception/Intimidation/Insight DCs), and persisting NPC memory + consequences. Load whenever the player talks to, persuades, or reads an NPC ("I talk to...", "I ask...").
Status conditions and exhaustion reference (blinded, charmed, frightened, grappled, paralyzed, poisoned, prone, restrained, stunned, unconscious, etc.). Load when applying or resolving a condition. Persist via gm-condition.sh (PC) / gm-npc.sh condition (party) / gm-combat.sh condition (enemies).
D&D-kit leveling — XP thresholds, the level-up ceremony, hit dice by class, and ASI/feat/subclass handling. Load when gm-player.sh xp reports LEVEL_UP in a dnd5e-kit campaign. Non-D&D kits advance via their ruleset progression model (milestone / resource-axis), not this table.
| name | gm-skills |
| description | Skill-check resolution — when to roll, the DC ladder, fail-forward philosophy, and margin-based consequences. Load whenever the player attempts something uncertain ("I try to..."). Resolution runs through the generic core (d20 vs DC); this is the judgment framework for when/how. |
Don't roll for: trivial tasks, impossible tasks, routine professional work, or anything with no meaningful consequence for failure.
uv run python lib/dice.py "1d20+[mod]" (or via game_core.resolve_check). 3. Narrate by margin.Trivial 5 · Easy 10 · Moderate 15 · Hard 20 · Very Hard 25 · Nearly Impossible 30.
Nat 20 = exceptional flourish · beat by 10+ = looks easy, extra benefit · success = clean · fail by 1-4 = close, minor setback · fail by 5+ = clear fail + complication · nat 1 = mishap.
A clever/effective/unique solve EARNS progress, not just a kill. On a strong success (nat 20, beat-by-10+, or an inventive approach), grant it before narrating: bash tools/gm-player.sh award --tier minor|major|legendary --reason "..." (kit-aware, level-scaled, co-awards followers in DCC). See gm-craft → Reward the spectacle.
A failed roll NEVER means "nothing happens" — it means "something DIFFERENT happens."
Fail-forward ≠ immortal. "Something different happens" can include death when the stakes were lethal and telegraphed (over-matched threat, ignored warning, a tightening string of failures). Don't soften a self-inflicted lethal outcome into a free pass. On PC death → Death Protocol (CLAUDE.md).
For significant failures: bash tools/gm-consequence.sh add "[what happens]" "[when]" [--trigger-type ...].
STR: Athletics · DEX: Acrobatics/Sleight of Hand/Stealth · INT: Arcana/History/Investigation/Nature/Religion · WIS: Animal Handling/Insight/Medicine/Perception/Survival · CHA: Deception/Intimidation/Performance/Persuasion. (A non-D&D kit defines its own skill list in ruleset.json.)