memory

Instruction file imported from Greenrenge/cf-webhook-fanout (.github/instructions/memory.instructions.md). Copyright stays with the author.

Hierarchical Memory Tool Instructions

Token Efficiency Rule: Only load memory that's relevant to current task context.

1. Session Start Protocol

Always begin with: "Remembering..." then execute:

search_nodes("session_context")  # Get last session state
search_nodes("task_node")        # Get task tree current state

Parse the results to identify:

• Current active task (status: in_progress)
• Suspended tasks (status: suspended)
• Next priority tasks (status: not_started)

If unclear current position: Ask human "Which task should I focus on?"

2. Memory Categories & Entity Types

a) Project Constellation (entityType: "project_constellation")

• Create ONCE per project, never modify
• Contains: project identity, core constraints, success criteria
• Load: Only when starting new project or major architecture changes

b) Strategic Memory (entityType: "strategic_plan")

• Current architectural approach (your "Option 1 Do Less Get More")
• Technology stack decisions
• Load: At session start if architecture questions arise

c) Task Hierarchy (entityType: "task_node")

• Always track: status, parent/child relationships, local context
• Load pattern: Current task + immediate children + parent context
• Update frequency: Every significant progress or discovery

d) Session Context (entityType: "session_context")

• Critical for handoffs: current position, next actions, blockers
• Create: At end of every session
• Load: At start of every session

e) Tool Registry (entityType: "tool_registry")

• Available commands, usage patterns, integration approaches
• Load: Only when encountering unknown tools or error patterns

f) Rules (entityType: "rule_set")

• Code standards, testing requirements, quality gates
• Load: When writing code or reviewing standards

3. Task Tree Navigation Commands

Find current work:

search_nodes("status: in_progress")

Create child task (for ad-hoc work):

create_entities([{
  name: "ChildTask_YYYY-MM-DD",
  entityType: "task_node",
  observations: ["status: not_started", "parent_context: [parent_name]"]
}])
create_relations([{from: "ChildTask_YYYY-MM-DD", to: "ParentTask", relationType: "child_of"}])

Complete and bubble up:

add_observations([{
  entityName: "completed_task",
  contents: ["status: completed", "outcomes: [what was accomplished]"]
}])
# Then check parent task progress

Context switch (temporary):

add_observations([{entityName: "current_task", contents: ["status: suspended"]}])
# Work on other task
# Later: add_observations([{entityName: "current_task", contents: ["status: in_progress"]}])

4. Token Efficiency Guidelines

Load Minimally:

• Current task + 1 level up + 1 level down in hierarchy
• Only load strategic/rules memory when needed for current work
• Use open_nodes([specific_names]) instead of read_graph() for focused loading

Update Frequently:

• Task progress and local context on every significant finding
• Session context before any major transition
• Strategic memory only on architecture decisions

Clean Periodically:

• Mark completed tasks clearly
• Archive old session contexts (keep last 3-5)
• Consolidate related observations to reduce noise