JSON Memory

Structured JSON memory system with efficient jq-based updates and action logging.

Installation

/plugin install json-memory@abilityai

Skills

Skill	Description
/json-memory:setup-memory	Initialize the memory system
/json-memory:load-memory	Load context at session start
/json-memory:update-memory	Save changes after a session
/json-memory:memory-jq <op> <path> [value]	Direct jq-based field updates

Setup

Run /json-memory:setup-memory to initialize. Creates:

memory/
├── memory_index.json   # Primary structured memory
├── memory_map.yaml     # Index of memory components
└── action_log.txt      # Reverse-chronological action history

Memory Structure

The memory_index.json contains these sections:

{
  "metadata": { "schema_version", "last_updated", "total_interactions" },
  "profile": { "name", "role", "context", "objectives" },
  "preferences": { "communication_style", "tools" },
  "memory": { "key_facts", "decisions", "project_overviews" },
  "context": { "active_topics", "recent_work", "scratchpad" },
  "entities": [{ "type", "name", "relationship", "importance" }],
  "tasks": { "active": [], "completed": [] },
  "custom": {}
}

Key Principles

Selective Loading

Never load the entire memory file. Use jq queries to load only what you need:

jq '{profile, context}' memory/memory_index.json

Atomic Updates

Use the tmp file pattern to avoid corruption:

jq '.field = "value"' file.json > tmp.json && mv tmp.json file.json

Common JQ Patterns

# Set a field
/json-memory:memory-jq set .profile.name "Agent Name"

# Append to array
/json-memory:memory-jq append .memory.key_facts "Important fact"

# Increment counter
/json-memory:memory-jq increment .metadata.total_interactions

# Read a field
/json-memory:memory-jq get .profile.name

Workflow

1.Session Start — Run /json-memory:load-memory

2.During Session — Use /json-memory:memory-jq for quick updates

3.Session End — Run /json-memory:update-memory to persist changes

Install Cornelius Playbook Builder