copilot-memories

Files (1)

• SKILL.md

SKILL.md (first lines)

# Copilot Memories - Persistent Memory Architecture

## Overview

Three-layer memory system that persists across sessions. Every interaction follows: **LOAD → USE → SYNC**.

```
Session Start → LOAD all layers → USE for context → SYNC updates → Session End
```

## Architecture

### Layer 1: User Memory

**Location:** `memories\root-memories.md`

**Purpose:** Personal/global preferences that follow the user across ALL workspaces.

| Stores | Examples |
|--------|----------|
| Coding style preferences | "Use tabs not spaces", "Prefer early returns" |
| Naming conventions | "camelCase for variables, PascalCase for types" |
| Tool preferences | "Use pnpm, not npm" |
| Workflow preferences | "Always write tests first" |
| Communication style | "Be concise, no fluff" |

### Layer 2: Repo Memory

**Location:** `memories\repo\<topic>.md`

**Purpose:** Project-scoped facts that persist across sessions inside this workspace.

| Stores | Examples |
|--------|----------|
| Architecture facts `architecture.md` | "We use React Query for data fetching" |
| Build commands `command.md` | "Run `npm run build` to compile the project" |
| Business rules `business.md` | "Free tier users can only create 3 projects" |
| Rules/conventions `rules.md` | "All API routes must be defined in `pages/api`" |
| Security vulnerabilities `security.md` | "Don't forget to sanitize user input in the search endpoint" |

### Layer 3: Session Memory

**Location:** `memories\<session-id>\<purpose>.md`

**Purpose:** Temporary task-specific notes for the CURRENT conversation only.

| Stores | Examples |
|--------|----------|
| Investigation notes `investigation.md` | "Tried X, got Y error, seems related to Z" |
| Current task goals `goals.md` | "Implementing JWT auth" |
| Plan/next steps `plan.md` | "1. Research JWT libraries 2. Choose one 3. Implement" |
| Multi-phase based on plan `phase-1-research.md`, `phase-2-implementation.md` | "Phase 1: Research complete. Phase 2: Implementation in progress." |
| Findings `findings.md` | "Passport.js is popular but heavy, JWT-Auth is lightweight and sufficient" |
| Implementation notes `implementation.md` | "Added auth middleware, updated user model, wrote tests" |
| Review notes `review.md` | "Code works but needs better error handling" |
| Test cases `test-cases.md` | "Tested login, logout, token refresh flows" |
| Rules/assumptions `rules.md` | "Assuming we need to support both web and mobile clients" |

## Decision Tree

```
Is it personal/global preference?
  ├── YES → User Memory (root-memories.md)
  └── NO  → Is it related to the current project/repo?
              ├── YES → Is it temporary/task-specific?
              │         ├── YES → Session Memory (memories/session/)
              │         └── NO  → Repo Memory (memories/repo/)
              └── NO  → User Memory (root-memories.md)
```

### Decision Flow (in order):

1. **First check:** Is this fact permanent and personal to the developer? → User Memory
2. **Second check:** Is this a project-scoped fact that future sessions need? → Repo Memory
3. **Third check:** Is this only useful for the current task/conversation? → Session Memory
4. **Default:** When in doubt, use Repo Memory (it's the safest middle ground)

## File Format & Templates

### User Memory (`root-memories.md`)

```markdown
# User Preferences

## Coding Style
-
-

## Preferred Tools
-
-

## Workflow
-
-

## Communication
-
-
```

### Repo Memory (`memories/repo/`)

Each file is one `.md` file per topic:

```markdown
# <Topic Name>

## Fact
<concise statement>

## Source
<how we learned this>

## Related
<links to other memory files or ADRs>
```

Naming: `naming-conventions.md` — e.g. `business.md`, `commands.md`, `folder-structure.md`

From tainn03/oh-my-copilot, MIT.

More from tainn03/oh-my-copilot