You are saving the current work context so it can be resumed later.

Purpose

This skill creates a compact snapshot of:

•What you're currently working on
•Progress made so far
•Important decisions and context
•Next steps to take
•Relevant file paths and code locations

This allows another AI session (or the same session later) to pick up exactly where you left off.

Process

1. Gather Context

Collect information about the current work:

bash

# Check git status to see what's changed
git status

# See recent changes
git diff --stat

# Check recent bash history if relevant
# Look at task list if tasks were created

2. Identify Key Information

Determine:

•Task description: What problem are you solving?
•Current state: What's been completed? What's in progress?
•Important context: Key decisions, architectural choices, dependencies
•File locations: Which files are being modified or are relevant
•Next steps: What needs to happen next (specific and actionable)
•Blockers: Any issues or open questions

3. Create Compact Summary

Write a concise but complete context document. Use this structure:

markdown

# Task Context

## Objective
[One sentence: What is being built/fixed/improved]

## Current Status
- ✅ Completed: [List completed items]
- 🔄 In Progress: [Current work]
- ⏳ Next Steps: [Specific next actions]

## Key Context
[Important background, decisions, or constraints]

## Files Modified/Relevant
- `path/to/file.ts:42` - [Brief note about why this matters]
- `path/to/other.py` - [Brief note]

## Technical Notes
[Any important technical decisions, patterns being used, or gotchas]

## Questions/Blockers
[Any open questions or blockers, if applicable]

---
Last updated: [Current date/time]

4. Write to TASK.md

Save the context to TASK.md in the current working directory:

bash

# Get current directory
pwd

Then use the Write tool to create ./TASK.md with your summary.

5. Confirm with User

Show the user what was saved and confirm:

•File location (full path to TASK.md)
•Brief summary of what was captured
•Reminder that they can use /resume-context to load this later

Best Practices

Be concise but complete:

•Don't include unnecessary details
•Focus on what's needed to resume work
•Be specific (file paths, line numbers, function names)

Make next steps actionable:

•❌ Bad: "Continue working on authentication"
•✅ Good: "Implement password reset flow in auth/reset.ts, following OAuth pattern from auth/login.ts:42-78"

Capture decisions:

•Document why certain approaches were chosen
•Note any trade-offs or alternatives considered
•Include relevant context from CLAUDE.md or discussions

Keep it current:

•Update TASK.md as work progresses
•Don't let stale context accumulate

Example Output

markdown

# Task Context

## Objective
Add user authentication with JWT tokens to the Express API

## Current Status
- ✅ Completed:
  - Set up JWT library and middleware
  - Created User model with password hashing
  - Implemented /register endpoint
- 🔄 In Progress:
  - Working on /login endpoint validation
- ⏳ Next Steps:
  1. Complete login endpoint with proper error handling
  2. Add refresh token mechanism
  3. Write tests for auth endpoints

## Key Context
- Using bcrypt for password hashing (cost factor 12)
- JWT secret stored in .env (not committed)
- Following REST API patterns from existing /users endpoints
- Auth middleware should be reusable for all protected routes

## Files Modified/Relevant
- `src/routes/auth.ts:1-45` - Auth endpoints (register complete, login in progress)
- `src/models/User.ts:12-34` - Password hashing methods
- `src/middleware/auth.ts:1-28` - JWT verification middleware
- `.env.example:8-9` - JWT_SECRET configuration

## Technical Notes
- Using jsonwebtoken library (already in package.json)
- Tokens expire after 24 hours
- Password must be min 8 chars, validated in User model
- Following existing error response format: { error: "message" }

## Questions/Blockers
- Should we implement refresh tokens now or in a follow-up?
- Need to confirm if we want rate limiting on login attempts

---
Last updated: 2026-02-14 15:30

Important Notes

•Privacy: Don't save sensitive data (passwords, API keys, secrets)
•Location: Always save to ./TASK.md in the current working directory
•Version control: Users may want to add TASK.md to .gitignore if it's personal notes
•Overwriting: Warn user if TASK.md already exists before overwriting

Remember: Good context saves time. Make it easy for future you (or another AI) to jump right back in.