
cross-session-memory
by mathur-exe
SKILL.md
name: cross-session-memory description: Maintain project context across sessions and AI platforms
Cross-Session Memory Skill
Instructions for AI Agents
You are an AI coding assistant with the ability to maintain context across sessions using a file-based context system. This system is platform-agnostic and works across Cursor, Claude Code, Codex, and other AI coding tools.
Context File Locations
All context files live in docs/ (relative to the project root):
| File | Purpose |
|---|---|
AGENT.md | Instructions you read at every session start |
project.md | Static project overview and architecture |
session.md | Dynamic session state—start here |
memory.md | Accumulated learnings and decisions |
Session Start Protocol
When starting a new session:
-
Read context files in this order:
docs/AGENT.md(if exists)docs/project.md(if exists)docs/session.md(if exists)docs/memory.md(if exists)
-
Locate resume point: Find
## RESUME HEREsection indocs/session.md -
Summarize state: Tell the user:
- What task was in progress
- What was last completed
- Any blockers or pending items
-
Confirm before proceeding: Ask the user if the summary is accurate
During Session Protocol
Update docs/session.md after:
- Creating, modifying, or deleting files
- Completing significant tasks or milestones
- Encountering blockers or errors
- Making architectural decisions
- Changing direction or approach
Session End Protocol
Before the session ends:
-
Update
## RESUME HEREwith:- Current status (in-progress, blocked, completed)
- Last action attempted/completed
- Next logical step
- Any open questions or decisions needed
-
Update progress checkboxes if applicable
-
Archive session if switching contexts (optional)
Initialization Protocol
When user requests context setup (no docs/ context files exist):
- Create
docs/directory (if needed) - Create all four template files
- Populate
project.mdwith project information gathered from user - Initialize
session.mdwith first session entry
File Templates
AGENT.md Template
# Agent Instructions
This project uses cross-session context tracking. At the start of each session:
1. Read all files in `docs/` directory (focus on `AGENT.md`, `project.md`, `session.md`, `memory.md`)
2. Find `## RESUME HERE` in `session.md`
3. Summarize current state before proceeding
4. Update `session.md` after significant actions
## Project Conventions
<!-- Add project-specific conventions here -->
## Do Not
- Modify files without updating context
- Assume context from previous sessions
- Skip the resume summary step
project.md Template
# Project Overview
## Name
<!-- Project name -->
## Description
<!-- Brief description -->
## Goals
<!-- High-level objectives -->
## Architecture
<!-- System design, key components -->
## Tech Stack
<!-- Languages, frameworks, tools -->
## Key Files
<!-- Important files and their purposes -->
## Constraints
<!-- Limitations, requirements, boundaries -->
session.md Template
# Session Context
## RESUME HERE
**Status**: Not started
**Last Updated**: <!-- timestamp -->
**Platform**: <!-- cursor/claude-code/codex -->
### Current Task
<!-- What are we working on -->
### Last Action
<!-- What was just done -->
### Next Step
<!-- What should happen next -->
### Blockers
<!-- Any issues preventing progress -->
---
## Progress
### Completed
- [ ] <!-- Task 1 -->
- [ ] <!-- Task 2 -->
### In Progress
- [ ] <!-- Current task -->
### Pending
- [ ] <!-- Future task -->
---
## Session Log
### Session 1 - <!-- date -->
**Platform**: <!-- platform used -->
**Duration**: <!-- time spent -->
#### Accomplished
<!-- What was done -->
#### Decisions Made
<!-- Key decisions -->
#### Issues Encountered
<!-- Problems and solutions -->
memory.md Template
# Project Memory
Accumulated learnings and decisions that persist across sessions.
## Decisions
| Date | Decision | Rationale | Status |
|------|----------|-----------|--------|
<!-- Decisions log -->
## Learnings
### What Works
<!-- Successful patterns -->
### What Doesn't Work
<!-- Failed approaches to avoid -->
## Conventions Discovered
<!-- Project-specific patterns found during development -->
## External Dependencies
| Dependency | Purpose | Notes |
|------------|---------|-------|
<!-- External services, APIs, etc. -->
## Gotchas
<!-- Things to watch out for -->
Usage Examples
Initialize Context
User: "Set up context tracking for this project"
AI Response:
- Create
docs/directory (if needed) - Generate all four template files
- Ask user about project details
- Populate
project.mdwith gathered information - Confirm setup complete
Resume Session
User: "Let's continue where we left off"
AI Response:
- Read
docs/session.md - Find
## RESUME HEREsection - Summarize: "Last session you were working on [X]. You completed [Y] and were about to [Z]. There's a blocker regarding [W]. Is this accurate?"
- Wait for confirmation
- Proceed with next step
End Session
User: "I need to stop for now"
AI Response:
- Update
## RESUME HEREwith current state - Log session in Session Log section
- Update any memory items if applicable
- Confirm context saved
- Suggest: "Context saved. When you resume, I'll pick up from [current state]."
Transfer to Different Platform
User: "I'm switching to Cursor from Claude Code"
AI Response:
- Ensure
docs/session.mdis fully updated - Add note: "Transferred from Claude Code on [date]"
- Confirm all context files are saved
- Explain: "Context is saved in
docs/. When you open in Cursor, instruct it to readdocs/session.mdto resume."
Score
Total Score
Based on repository quality metrics
SKILL.mdファイルが含まれている
ライセンスが設定されている
100文字以上の説明がある
GitHub Stars 100以上
3ヶ月以内に更新がある
10回以上フォークされている
オープンIssueが50未満
プログラミング言語が設定されている
1つ以上のタグが設定されている
Reviews
Reviews coming soon