.agent/skills/memory-bank/SKILL.md
Persistent cross-session memory protocol. Load this skill whenever the user asks to initialize, read, update, or review the project Memory Bank, or whenever context about architecture, progress, or current state is needed.
npx skillsauth add a2mus/smart-da3m memory-bankInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
The Memory Bank is a structured set of Markdown files stored in memory-bank/
at the project root. It is the single source of truth for all cross-session
context. Because Antigravity agents start each session without memory of previous
ones, every session MUST begin by reading the Memory Bank files in order.
Read these files in order before any other action:
memory-bank/core/current-state.md — 🎯 THE NOW — Active tasks, current phase, blockers.memory-bank/core/projectBrief.md — 📋 THE MISSION — What we are building and why.memory-bank/core/productContext.md — 👥 THE USER — Target audience, UX goals, constraints.memory-bank/core/techContext.md — 🔧 THE STACK — Languages, versions, dependencies, config.memory-bank/core/patterns.md — 🏗️ THE PATTERNS — Architecture decisions and code conventions.memory-bank/core/progress.md — 📊 THE HISTORY — Completed milestones, known issues.memory-bank/core/next-session.md — 📝 THE HANDOVER — Notes left by the previous session.If any file is missing, create it with a minimal template and inform the user.
Before ending a session, always update:
current-state.md — Reflect what changed this session (tasks done, new blockers).progress.md — Append completed milestones with date (use YYYY-MM-DD format — always check system date first).next-session.md — Write clear, specific instructions for the next agent session:
current-state.md# Current State
**Phase:** [e.g., "MVP Development – Sprint 2"]
**Last updated:** YYYY-MM-DD
## Active Tasks
- [ ] Task description
## Blockers
- None
## Context Notes
(Anything the next agent needs to know immediately)
projectBrief.md# Project Brief
## What We Are Building
(One paragraph max — clear, specific)
## Goals
- Goal 1
- Goal 2
## Non-Goals
- What we are explicitly NOT doing
## Success Criteria
- Measurable outcomes
productContext.md# Product Context
## Target Users
(Who uses this, their level, their needs)
## Key User Flows
1. Flow description
## UX Constraints
- (e.g., mobile-first, offline support, RTL for Arabic)
techContext.md# Technical Context
## Stack
| Layer | Technology | Version |
|-------------|-------------------|---------|
| Language | | |
| Framework | | |
| Database | | |
| Deploy | | |
## Key Dependencies
(Libraries that define architecture choices)
## Environment
- Dev: (setup instructions)
- Prod: (deployment target)
## Configuration
(Env vars, secrets structure — no actual values)
patterns.md# Architecture & Code Patterns
## Architecture Decisions
- **ADR-001:** [Decision] — [Rationale]
## Code Conventions
- Naming:
- File structure:
- Error handling:
- Testing approach:
## Patterns to Enforce
(Patterns the agent must always follow)
## Anti-Patterns to Avoid
(What we explicitly decided not to do)
progress.md# Progress Log
## Milestones Completed
| Date | Milestone | Notes |
|------------|-------------------------------|-------|
| YYYY-MM-DD | Initial scaffolding | |
## Known Issues
- Issue description (priority: high/medium/low)
## Deferred Items
- Items moved to backlog
next-session.md# Next Session Handover
**Written by:** Agent on YYYY-MM-DD
## Start Here
(First thing to do next session)
## Open Decisions
- Decision needed from user: ...
## Files Last Modified
- `path/to/file.ext` — what changed
## Warnings
(Anything tricky or incomplete)
Date verification first — Before writing any date to a Memory Bank file, check the system date. Never assume or hallucinate dates.
Read before write — Never update a Memory Bank file without first reading its current content. Preserve existing entries.
Append, don't overwrite progress — progress.md entries are append-only. Old entries must never be deleted.
Keep files lean — If any file exceeds ~150 lines, split content into linked sub-files and update the index.
Self-maintenance (The Gardener) — When a new architectural pattern or decision is made during a session:
patterns.md immediately.Memory Bank is sacred — Never delete or significantly restructure the memory-bank/ folder without explicit user confirmation. All edits are reversible via Git.
When the user says "initialize memory bank" or "create memory bank":
memory-bank/core/ directory.projectBrief.md and techContext.md first.memory-bank/core/projectBrief.md and fill in your project details."project-root/
└── memory-bank/
└── core/
├── current-state.md ← Read FIRST every session
├── projectBrief.md
├── productContext.md
├── techContext.md
├── patterns.md
├── progress.md
└── next-session.md ← Read LAST before starting work
testing
Perform cross-artifact consistency analysis across spec.md, plan.md, and tasks.md. Use after task generation to identify gaps, duplications, and inconsistencies before implementation.
development
Run comprehensive verification on current codebase state.
testing
Intelligently sync your fork with the upstream spec-kit repository. Reads both versions of every changed file, compares quality, and produces the richest possible result by blending the best of each.
development
Impeccable UI design workflow — create distinctive, production-grade interfaces or enhance existing ones. Integrates design context gathering, anti-pattern detection, heuristic scoring, and systematic polish. Works after speckit-brainstorm (new design) or on existing UI code (enhancement mode).