The orchestrator prompt generated by hub generate is now significantly more complete. New sections are injected into every editor’s orchestrator output, covering behavior patterns that leading AI coding tools use internally. Skills are now listed with descriptions and repo associations.

What changed

AGENTS.md for all editors

All editor generators now write AGENTS.md at the workspace root:

• generateCursor writes AGENTS.md alongside .cursor/rules/orchestrator.mdc
• generateClaudeCode writes AGENTS.md alongside CLAUDE.md
• generateOpenCode writes AGENTS.md alongside .opencode/agents/orchestrator.md
• generateKiro no longer writes .kiro/steering/orchestrator.md (was duplicate)

Core behavior sections

New shared buildCoreBehaviorSections() adds 6 sections to all orchestrator prompts:

• Core Behavior — conciseness, accuracy over reassurance, no invented facts, no unnecessary apologies
• Working Style — simplest solution, avoid over-engineering, bias toward self-discovery
• Search, Reading, and Investigation — read before editing, gather context first
• Code Changes — runnable code, preserve patterns, max 3 retry attempts, check dependencies first
• Security and Safety — no hardcoded secrets, flag risks, avoid common vulnerability patterns
• Git and Operational Discipline — inspect before committing, no destructive operations without authorization

Skills listing

New buildSkillsSection() reads all SKILL.md files from the skills directory and generates a table in the orchestrator prompt with skill name, description, and associated repositories. Includes guidance on when to consult skills.

Why

The previous orchestrator prompt focused on pipeline mechanics but lacked guidance on how the agent should behave when actually working. Skills were copied to the editor but never referenced in the orchestrator prompt, so agents didn’t know they existed.

Who is affected

All workspaces using RHM. After upgrading, run hub generate to regenerate prompts for your editor.

Upgrade

npx @arvoretech/hub@0.15.0 generate