memory-bank

Maintains a structured project knowledge base in .memory-bank/ with machine-parseable update suggestions.

4|Updated Aug 8, 2026
One-click install
npx skills add https://github.com/alexeyshishin/as-skill --skill memory-bank-alexeyshishin
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: memory-bank
Source: https://github.com/alexeyshishin/as-skill/tree/main/domains/core/skills/memory-bank
Command: npx skills add https://github.com/alexeyshishin/as-skill --skill memory-bank-alexeyshishin

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Project knowledge — architecture decisions, conventions, API contracts, active tasks — gets lost between chat sessions, forcing every new contributor or AI session to rediscover context from scratch. ## Core Features & Use Cases - Importance-gated evaluation: After each interaction, classifies changes on a 0-3 importance scale and only suggests updates for level 2+ (new conventions, contracts, architectural changes), skipping noise like typo fixes. - Structured suggestion format: Emits machine-parseable MEMORY_SUGGESTION blocks with path, operation (create/update/append), content format (md/yaml/mermaid/diff), and validation flags, or an explicit NO_MEMORY_UPDATE notice. - Standardized knowledge layout: Defines a .memory-bank/ structure with index, product overview, steerings, tech details, and tasks, plus a bootstrap flow for new projects. - Use Case: After a chat where you define a new public API contract, the AI proposes a concise YAML contract entry for .memory-bank/tech-details/ instead of letting the decision vanish. ## Quick Start Ask the AI to maintain a memory bank for this project and suggest an update whenever an important architectural decision or convention changes.

Frequently Asked Questions about memory-bank

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I keep project context persistent across AI chat sessions?

Maintain a .memory-bank/ directory with structured Markdown files covering architecture, conventions, and tasks. After each interaction, evaluate whether stable project knowledge changed and record it as a concise update, so future sessions start with full context.

What should go into a project memory bank?

Store stable project knowledge: architecture decisions, public APIs, contracts, schemas, conventions, stack details, CI/CD setup, and current tasks. Exclude temporary fixes, local helpers, one-off experiments, and anything containing secrets or credentials.

How do I avoid noisy or excessive memory bank updates?

Use an importance scale from 0 to 3 and only suggest updates at level 2 or higher, such as new conventions or architectural changes. Typo fixes, formatting, and internal refactors stay below the threshold and are skipped.

Does the memory bank skill work with Claude Code, Cursor, and Codex?

Yes, the skill's frontmatter explicitly lists Claude Code, Cursor, and Codex as applicable environments. It is a prompt-based workflow with no external dependencies, so it runs anywhere SKILL.md instructions are supported.

When should I not update a project memory bank?

Skip updates for local implementation details, temporary experiments, minor refactors, and style-only changes. When unsure, do not suggest an update, since a wrong or noisy fact erodes trust in the knowledge base.