behavior-tree-composer

Core Capabilities

1. Generate syntactically correct py_trees behavior trees
2. Use domain-specific actions from action library
3. Validate generated BTs before returning to user
4. Iteratively refine based on validation feedback

Workflow

Step 1: Discover Available Actions

Invoke the bt-action-discovery skill (user-level skill in ~/.claude/skills/) to get available robot actions, parameters, and constraints from the project's action_library.

This returns:

• Primitive actions (GoToPose, SetPen, etc.)
• Composite actions (DrawShape, PatrolWaypoints, etc.)
• Workspace constraints
• Parameter ranges

Step 2: Understand Task

• Parse natural language task description
• Identify required actions from discovered action library
• Plan behavior tree structure

Step 3: Generate Behavior Tree

Create a Python file in generated_bts/ with this exact structure:

"""
[Description of what this BT does]
"""

import py_trees
import sys
import os

# Add parent directory to path
sys.path.insert(0, os.path.abspath(os.path.join(os.path.dirname(__file__), '../..')))

from py_trees_nodes.primitives import GoToPose, SetPen, PenUp, PenDown, MoveDistance, GetPose, CheckBounds
from py_trees_nodes.composites import DrawShape, PatrolWaypoints


def create_root() -> py_trees.behaviour.Behaviour:
    """
    Create the behavior tree root node.

    Returns:
        The root node of the behavior tree
    """

    # Build your tree here
    root = py_trees.composites.Sequence(name="MainTask", memory=True)

    # Add children actions...

    return root


if __name__ == '__main__':
    """Direct execution for testing."""
    print("Creating behavior tree...")
    root = create_root()
    print("Tree structure:")
    print(py_trees.display.unicode_tree(root, show_status=True))

CRITICAL Requirements:

• Function MUST be named create_root() (not create_behavior_tree())
• Must return py_trees.behaviour.Behaviour object
• Save to generated_bts/{descriptive_name}.py
• Only import actions that are actually used

Step 4: Validate Before Returning

CRITICAL: Always validate before presenting BT to user

Run validation orchestrator:

python validation/orchestrator.py generated_bts/{filename}.py

This runs all validation tiers and returns JSON with results.

Step 5: Handle Validation Results

Parse the JSON output and take action based on tier:

• Tier 1 Critical Errors: MUST fix - regenerate the BT
• Tier 2 Warnings: Should fix but not blocking
• Tier 3 Advisory: Informational only
• Max 3 regeneration attempts before giving up

Validation Decision Tree:

Tier 1 Pass?
  ├─ No  → MUST regenerate (up to 3 attempts)
  └─ Yes → Tier 2 Pass?
            ├─ No  → Should fix if possible, or warn user
            └─ Yes → Present to user

If still failing after 3 attempts:

• Explain limitation to user
• Show what errors remain
• Suggest simplifying the task

Detailed Patterns and Examples

For detailed generation patterns, composite structures, common patterns, and working examples, see:

generation-patterns.md (in the same directory as this SKILL.md)

This external reference includes:

• py_trees composite patterns (Sequence, Selector, Parallel)
• Common BT patterns (setup-execute, conditional, multi-step)
• Validation tier details
• Common generation errors and fixes
• Working examples to reference
• Best practices

Output Format

Save generated BTs to:

generated_bts/{descriptive_name}.py

After validation passes:

Present the file to user with:

• Summary of what the BT does
• How to execute it: python execution/run_bt.py generated_bts/{filename}.py
• Tree structure visualization

Notes

• Always validate before returning - This is mandatory
• Use bt-action-discovery to get available actions - Don't hardcode or invent actions
• Follow the template exactly - Especially imports and create_root() name
• Use memory=True for Sequences - This is the standard
• Degree of freedom: Medium - Structured generation with validation loop

Related Skills

• bt-action-discovery - Discovers available robot actions
• bt-validator - Standalone validation and debugging