Releases: SukinShetty/Nemp-memory
/nemp:health — Memory Diagnostics Engine
What's New
/nemp:health — A self-diagnosing memory system for AI agents.
One command. 10 diagnostic checks. A health score out of 100. Auto-fix for common issues.
The Problem
Your Nemp memory silently degrades over time, orphaned keys, broken CLAUDE.md sync, missing logs, stale memories piling up. You don't notice until the agent starts making bad decisions.
The Solution
/nemp:health
Returns a weighted health score (0-100) across 10 checks:
| Check | What It Catches | Weight |
|---|---|---|
| Memory file integrity | Corrupted or missing memories.json | 15 |
| CLAUDE.md sync | Memory ↔ CLAUDE.md out of sync | 15 |
| Access log health | Missing or corrupted access.log | 10 |
| Stale memory detection | Memories not accessed in 30+ days | 10 |
| Duplicate detection | Semantically overlapping memories | 10 |
| Key naming quality | Vague keys like "misc", "temp", "stuff" | 10 |
| Value quality | Empty, too short, or too long values | 10 |
| Memory count balance | Too few or too many memories | 5 |
| Config integrity | Missing or invalid config.json | 5 |
| Token budget | Total memory token usage | 10 |
Three Modes
Standard — Quick diagnostic report
/nemp:health
Verbose — Detailed breakdown per check with specific issues listed
/nemp:health --verbose
Auto-fix — Automatically repair what it can (regenerate MEMORY.md, create missing access.log, fix sync issues)
/nemp:health --fix
Why This Matters
Full Changelog
- New command:
/nemp:healthwith 10 weighted diagnostic checks - New command:
/nemp:health --verbosefor detailed breakdown - New command:
/nemp:health --fixfor auto-repair - New file:
commands/health.md(command specification) - Score-based health reporting (0-100 weighted scale)
- Auto-fix capabilities: MEMORY.md regeneration, access.log creation, sync repair
v0.3.0 - Nemp Foresight
What's New
🔮 Nemp Foresight: Predictive Memory Pre-loading
Foresight works like a personal assistant. It hears your request, grabs only the relevant memories, and loads them into the agent's brain before it even starts thinking.
The agent doesn't search memory anymore. Memory finds the agent.
Instead of dumping all memories or making the agent search, Foresight reads your prompt and predicts which memories are relevant.
/nemp:foresight "Fix the login bug where tokens expire too quickly"
- Scores all memories against your prompt
- Returns only relevant ones (39 → 12 in testing, 69% token savings)
- Auto-adjusts based on intent: high-action prompts load more, simple queries load less
- 13 domain detectors: auth, database, api, frontend, backend, testing, deployment, payment, design, state, security, architecture, feature
- Core keys (framework, stack, language) always included
v0.2.0 - Agent Tracking, Access Logs & Token Optimization
What's New
Agent ID Tracking
Every memory write now records an agent_id field showing which agent wrote it — useful for multi-agent workflows where multiple Claude agents collaborate on the same project.
Access Logs
All read, write, and delete operations are logged to .nemp/access.log with timestamps, agent names, and operation details.
New /nemp:log Command
View the audit trail of all memory operations. Filter by agent, tail recent entries, or clear the log.
Token Compression
/nemp:save now compresses values before storing — removing filler words, redundant phrases, and keeping values under 200 characters. ~70% smaller stored values.
/nemp:init Optimization
Single bash scan, reads only package.json, writes all memories in one operation. Significantly fewer tool calls and less token usage.
MEMORY.md Index
Auto-generated index file at .nemp/MEMORY.md provides a quick overview of all stored memories with agent attribution.
Updated Commands
save.md— token compression, agent_id, access logginginit.md— single-scan optimization, MEMORY.md generationrecall.md— agent display, read logginglist.md— agent column, MEMORY.md regenerationforget.md— delete logging, MEMORY.md update
New Commands
log.md— audit trail viewer (/nemp:log)
Compatibility
Backward compatible with v0.1.0 memory files. Existing memories will work without migration.
v0.1.0 — First official release of Nemp Memory
Nemp Memory v0.1.0
The memory plugin for Claude Code that remembers everything. 100% local, zero cloud, zero API keys.
🚀 Agent Teams Compatible - Tested with Claude Code's new agent teams feature (launched Feb 5, 2026). Multiple teammates can pull project decisions from shared Nemp Memory in parallel.
Core Features
/nemp:init— Auto-detect project tech stack/nemp:save— Save architectural decisions and project context/nemp:recall— Recall specific memories/nemp:context— Search memories by keyword/nemp:sync— Two-way sync between Nemp and CLAUDE.md with conflict detection/nemp:auto-sync— Continuous sync as you work/nemp:suggest— AI-suggested memories based on activity/nemp:list— List all memories/nemp:forget— Remove a memory/nemp:export— Export to CLAUDE.md
Install
/plugin marketplace add https://github.com/SukinShetty/Nemp-memory
/plugin install nemp
Links
- 🌐 Website: nemp.dev
- 📝 Blog: substack.com/@sukinshetty
Full Changelog: https://github.com/SukinShetty/Nemp-memory/commits/v0.1.0