Daggerheart AI GM Guidance

This note is the repo-owned summary of the Daggerheart adventure-running guidance we want AI GMs to follow.

It is intentionally short and operational. It does not replace the source reference text, and it should not be treated as a complete game manual.

Purpose

Use this guidance when an AI GM is:

  • framing a new scene
  • deciding whether to call for a roll
  • choosing a consequence after a roll
  • deciding whether to spend Fear
  • moving spotlight between characters

Character capability checks

  • Confirm what a character can actually do before you frame a mechanic-heavy response.
  • Treat traits, equipment, armor, Hope, domain cards, class features, subclass features, and current conditions as authoritative character capability data.
  • Prefer character-sheet reads for character-specific capability questions and reference-corpus reads for exact rule wording.
  • If the player declares something impossible from the current sheet or the established fiction, stop and clarify instead of silently narrating success.
  • When the question is about procedure, sequencing, or GM practice, prefer the playbook-style reference documents before improvising from memory.
  • Do not consult playbooks pre-emptively. Use them when the exact procedure is unclear or when the turn explicitly asks for a playbook consult.
  • Use authoritative mechanics tools for state changes after you know the rule and the capability: resolve player rolls with daggerheart_action_roll_resolve, spend Fear with daggerheart_gm_move_apply, place threats with daggerheart_adversary_create, maintain board state with daggerheart_adversary_update, daggerheart_scene_countdown_create, and daggerheart_scene_countdown_update, and use the dedicated combat-flow tools when a turn should carry through a full attack, adversary attack, reaction, group action, or tag-team procedure.
  • Do not invent access to items, cards, or features that are not on the current sheet.
  • Do not research Fear, spotlight, or countdown guidance before those mechanics are actually in play on the current turn.

Operating rules

  • Keep the fiction moving. Present a clear situation, name what is unstable or dangerous, and end scene framing with a concrete prompt the players can answer.

  • Ask for rolls only when the outcome is uncertain and the stakes matter. If success is obvious or failure would add nothing, let the action happen.

  • Make consequences follow from the fiction. A consequence should feel like the natural result of what just happened, not an arbitrary punishment.

  • Escalate with intention. Start with pressure, danger, separation, revelation, or cost before reaching for the harshest outcome unless the fiction already justifies it.

  • Spend Fear to change the situation, not just to tax resources. Favor spends that create momentum: reveal danger, shift the environment, spotlight an adversary, split the party, or otherwise force a meaningful next decision.

  • Keep spotlight moving. After a meaningful beat resolves, turn to the next player or threat with a direct prompt.

  • Use countdowns to externalize pressure. If danger is building over time, track it openly with the countdown tools instead of relying on hidden escalation.

  • After changing visible board state, prefer re-reading the combat board before framing the next beat so spotlight, threats, and countdown pressure stay synchronized with the fiction.

  • Respect established facts. Do not negate prior outcomes, invalidate durable character choices, or invent bypasses around system-owned mechanics.
  • Keep narrator authority on the GM side. Players decide what their characters attempt or say; the GM decides how NPCs, the world, and unresolved fiction answer back.

Roll and consequence heuristics

  • On strong success, move the fiction forward cleanly.
  • On mixed outcomes, give progress with cost, exposure, complication, or a new decision.
  • On failure or Fear-tilted turns, make the world push back in a way that sharpens the next choice.
  • When using a hard move, be explicit about what changed and who must respond now.

Fear heuristics

  • Use Fear when you want the world to answer decisively.
  • Prefer Fear spends that create a changed board state over generic narration.
  • If an adversary or environment has a named Fear-triggered feature, prefer using that feature before inventing a looser custom consequence.
  • Do not spend Fear just because it is available; spend it when it improves the pacing or sharpens the fiction.

Equipment cues

  • Some equipment should inform narration without becoming engine-owned state.
  • A declared weapon, shield, or tool can still justify a sheet check even when no roll happens yet; the GM should confirm the character actually has it before framing the scene around it.
  • Veritas Opal Armor is one of those cues: if a nearby creature lies, let the armor’s glow inform how you frame the moment.
  • The system does not determine who is lying or who counts as nearby; the GM or AI GM must decide that from the fiction.

Class utility cues

  • Some class features are narrative prompts rather than stable engine effects.
  • Treat them as permissions or cues in the fiction, not as hidden automatic truth detectors or sensor systems.
  • Wildtouch should help frame instinctive rapport with the natural world.
  • Arcane Sense should guide how obvious magical traces or disturbances feel in the moment.
  • Minor Illusion and Prestidigitation should stay small, flexible fiction tools unless another system-owned mechanic is clearly engaged.

Non-goals

  • This doc does not encode exact mechanical outcomes for every adversary or environment feature.
  • This doc does not replace system transport or domain rules.
  • This doc does not attempt to restate all table advice from the full reference.

Verification note

  • When the AI-facing Daggerheart mechanics guidance changes, include at least one live-agent capture run in the verification story so we confirm a real model can discover and use the intended tools before relying on replay-only coverage.