create-behavior-tree

1. Overview

Generates formal BT specs for NPC/AI decision-making. Translates agent motivations into hierarchical decision trees. Output: behavior_tree_spec.md — discrete actions, conditional checks, prioritized decision hierarchies.

Scenarios: NPC/enemy/allies AI behavior, hierarchical trees from qualitative descriptions, reusable AI capabilities for multiple agent types.

2. Quick Reference

Output: {TARGET_FOLDER}/docs/behavior_tree_spec.md.

BT node types:

• Selector — tries children in priority order; succeeds on first child success
• Sequence — executes children left-to-right; fails on first child failure
• Decorator — modifies a single child's behavior (Inverter, Repeat, UntilFailure, etc.)
• Action — leaf node that performs a task (succeed/fail)
• Condition — leaf node that checks a predicate (true/false)

Workflow: Role & Goal Discovery → Capability Extraction (Actions/Conditions) → Behavior Mapping → Tree Construction & Validation → Export. Use - indentation only (no Unicode tree chars). Every branch reaches a terminal node.

3. Workflow

1. Role & Goal Discovery: Identify AI Role (Guard, Merchant), Type (Hostile/Friendly/Neutral), primary motivations.
2. Capability Extraction: Define available Actions and Conditions without assuming implementation details — these become leaf nodes in the tree.
3. Behavior Mapping: Translate natural language rules into prioritized BT node hierarchy (Selectors, Sequences, Decorators). Each node has a single responsibility; compose complex behavior by arranging simple nodes into a tree structure.
4. Tree Construction & Validation: Strict formatting using - indentation only (avoid Unicode tree chars or nested code blocks). Every branch leads to an action or valid exit. Validate that each subtree reaches a terminal node.
5. Export: Write to {TARGET_FOLDER}/docs/behavior_tree_spec.md.

3. Intercepts

• [Implementation Detail]: Engine-specific code detected → "Translating [Detail] to behavioral description (e.g., 'Play animation')."
• [Vague Logic]: Ambiguous rules → "Logic for [Behavior] too vague. Use concrete conditions like is_in_range or has_line_of_sight?"

4. Quality Gates

• Behavioral Descriptions: All nodes use abstract behavioral descriptions exclusively
• Testable Conditions: Every condition is concrete and verifiable (e.g., is_in_range, has_line_of_sight)
• Complete Branches: Every branch leads to a terminal action or valid exit
• Progressive Logic: Each decision advances the tree forward without repeating checks
• Hierarchical Structure: Use proper Selector/Sequence nesting instead of flat lists