• Context: Loaded via ContextProvider Interface
• Memory: Accessed via MemoryStore Interface
• Schemas: Validated against context_metadata.schema.json and memory_entry.schema.json

IMPORTANT: Execute ALL steps in order. Do not skip any step.

• Gather user intent (what task or problem they want to solve)
• Determine the scope: skill discovery, agent recommendation, command lookup, or full workflow orchestration
• Identify whether the user needs a single tool or a multi-tool workflow

• Use

  memoryStore.getSkillMemory("divine", "{project-name}")

  to load existing project memory. See MemoryStore Interface.
• Load previously recommended workflows and their outcomes
• Retrieve user preference patterns for tool selection

• Load engineering domain context via

  contextProvider.getIndex("engineering")

  . See ContextProvider Interface.
• Load the skill registry (all available SKILL.md files)
• Load the agent registry (all agent .config.json files)
• Load the command registry (all command .md files)

• Enumerate all skills, agents, and commands in The Forge
• Classify each by category, domain, and capability
• Build an indexed catalog with triggers, descriptions, and prerequisites

• Analyze the user's task against the catalog
• Score each tool for relevance based on:
  • Domain alignment (language, framework, infrastructure)
  • Task type match (review, generate, plan, analyze, deploy)
  • Trigger keyword matches
  • Historical success from memory
• Rank recommendations by confidence

• For simple tasks: recommend a single skill, agent, or command
• For complex tasks: compose a multi-step workflow with:
  • Ordered skill invocations via

    skillInvoker

  • Agent delegation assignments
  • Command sequences
  • Data flow between steps
• Include estimated effort and prerequisites for each step

• Save output to

  /claudedocs/divine_{project}_{YYYY-MM-DD}.md

• Follow naming conventions in

  ../OUTPUT_CONVENTIONS.md

• Present recommendations with rationale and confidence scores

• Use

  memoryStore.update(layer="skill-specific", skill="divine", project="{project-name}", ...)

  to store insights. See MemoryStore Interface.
• Record which tools were recommended and for what tasks
• Track user feedback on recommendation quality
• Update preference patterns for future recommendations

• All mandatory workflow steps executed in order
• Memory loaded via

  memoryStore.getSkillMemory()

  (Step 2)
• Context loaded via

  contextProvider.getIndex()

  (Step 3)
• Tool discovery completed with full catalog (Step 4)
• Task-to-tool matching performed with scoring (Step 5)
• Output saved with standard naming convention
• Memory updated via

  memoryStore.update()

  (Step 8)