long-agent（长 Agent）

Long Agent Sessions

Manage extended agent sessions with progress tracking and context management.

Session Management

Initialize Session

# Create session file
SESSION_ID=$(date +%Y%m%d-%H%M%S)
SESSION_FILE=~/.claude/sessions/$SESSION_ID.md

mkdir -p ~/.claude/sessions

cat > $SESSION_FILE << EOF
# Session: $SESSION_ID
Started: $(date)
Task: [Describe task]

## Progress
- [ ] Step 1
- [ ] Step 2
- [ ] Step 3

## Context
[Important context to remember]

## Notes
[Running notes]
EOF

echo "Session: $SESSION_FILE"

Track Progress

# Update progress
echo "- [x] Completed: [description]" >> $SESSION_FILE

# Add note
echo "### $(date +%H:%M) - [title]" >> $SESSION_FILE
echo "[notes]" >> $SESSION_FILE

Resume Session

# List recent sessions
ls -lt ~/.claude/sessions/ | head -10

# View session
cat ~/.claude/sessions/SESSION_ID.md

# Continue from last checkpoint
tail -50 ~/.claude/sessions/SESSION_ID.md

Progress Patterns

Checkpoint System

#!/bin/bash
# checkpoint.sh SESSION_ID DESCRIPTION

SESSION_FILE=~/.claude/sessions/$1.md
CHECKPOINT=$2

echo "" >> $SESSION_FILE
echo "## Checkpoint: $(date +%H:%M)" >> $SESSION_FILE
echo "Status: $CHECKPOINT" >> $SESSION_FILE
echo "Can resume from here." >> $SESSION_FILE

Milestone Tracking

# Session Progress

## Milestones
- [x] M1: Environment setup
- [x] M2: Core implementation
- [ ] M3: Testing
- [ ] M4: Documentation

## Current: M3 - Testing

### Completed
- Unit tests for auth module
- Integration test setup

### In Progress
- API endpoint tests

### Blocked
- Need test fixtures for edge cases

Context Management

Save Important Context

#!/bin/bash
# save-context.sh SESSION_ID

SESSION_FILE=~/.claude/sessions/$1.md

cat >> $SESSION_FILE << 'CONTEXT'

## Critical Context (Don't Lose)

### File Locations
- Main entry: src/index.ts
- Config: src/config.ts
- Tests: tests/

### Key Decisions
- Using JWT for auth (not sessions)
- PostgreSQL for persistence
- Redis for caching

### Current State
- Auth module: complete
- API routes: 80% done
- Tests: 40% coverage
CONTEXT

Context Recovery

# Get context for resumption
CONTEXT=$(grep -A 50 "## Critical Context" ~/.claude/sessions/$SESSION_ID.md)

gemini -m pro -o text -e "" "I'm resuming a long coding session. Here's the context:

$CONTEXT

Help me remember:
1. Where was I?
2. What's the next step?
3. Any gotchas to remember?"

Bearings Check

Get Your Bearings

#!/bin/bash
# bearings.sh - Where am I in this task?

echo "=== Current Session Status ==="
echo ""

# Git status
echo "### Git Status"
git status --short

echo ""
echo "### Recent Commits"
git log --oneline -5

echo ""
echo "### Modified Files"
git diff --stat HEAD~3

echo ""
echo "### TODO Items"
grep -r "TODO\|FIXME" src/ --include="*.ts" | head -10

Progress Summary

# Generate progress summary
gemini -m pro -o text -e "" "Summarize progress on this session:

Session log:
$(tail -100 ~/.claude/sessions/$SESSION_ID.md)

Git activity:
$(git log --oneline -10)

Provide:
1. What's been accomplished
2. Current status
3. Remaining work
4. Estimated completion"

Long Task Patterns

Multi-Day Task

# Multi-Day Task: [Name]

## Day 1 - [Date]
### Accomplished
- [items]

### Blocked
- [items]

### Tomorrow
- [items]

---

## Day 2 - [Date]
### Context from Yesterday
[summary]

### Accomplished
- [items]

Handoff Document

When handing off to another session or agent:

# Task Handoff

## What This Is
[Brief description]

## Current State
- [Bullet points of where things stand]

## What's Working
- [Completed items]

## What's Not Working
- [Issues/blockers]

## Next Steps
1. [Specific step]
2. [Specific step]

## Key Files
- `src/main.ts` - Entry point
- `src/auth/` - Auth module (done)
- `src/api/` - API routes (in progress)

## Commands to Know

npm run dev # Start development npm test # Run tests npm run build # Build for production

## Gotchas

- [Things that might trip you up]

Session Templates

Feature Development

cat > ~/.claude/sessions/feature-template.md << 'EOF'
# Feature: [Name]

## Requirements
- [ ] Requirement 1
- [ ] Requirement 2

## Design
[Architecture notes]

## Implementation Progress
- [ ] Step 1
- [ ] Step 2
- [ ] Step 3

## Testing
- [ ] Unit tests
- [ ] Integration tests
- [ ] Manual testing

## Documentation
- [ ] Code comments
- [ ] README update
- [ ] API docs
EOF

Bug Fix

cat > ~/.claude/sessions/bugfix-template.md << 'EOF'
# Bug Fix: [Issue ID/Description]

## Symptom
[What's happening]

## Expected
[What should happen]

## Investigation
- [ ] Reproduced locally
- [ ] Found root cause
- [ ] Identified fix

## Fix
[Description of fix]

## Verification
- [ ] Fix works
- [ ] No regressions
- [ ] Tests added
EOF

Best Practices

1. Log frequently - Update progress often
2. Save context - What would you need to resume?
3. Create checkpoints - Mark safe resume points
4. Summarize before breaks - Capture state of mind
5. Use templates - Consistent structure helps
6. Track blockers - Note what's waiting
7. Handoff cleanly - Assume fresh context next time