session-start-protocol-skill

name	session-start-protocol-skill
description	Initialize Claude Code session following Anthropic best practices for multi-session continuity. Use when: (1) starting any new Claude Code session, (2) resuming work after context compaction, (3) uncertain about current project state. Source: Anthropic Claude 4 Best Practices.

Session Start Protocol

Purpose: Multi-session continuity and progress tracking Source: Anthropic's Claude 4 Best Practices + Agent Harness patterns Success: Enables unlimited context through automatic summarization

---

Usage Scenarios

(1) When starting any new Claude Code session

• Every morning
• After breaks
• When switching between projects

(2) When resuming work after context compaction

• Claude Code automatically compacts context when approaching limits
• You can continue working indefinitely
• Use this protocol to orient yourself

(3) When uncertain about current project state

• What was I working on?
• What's completed vs in-progress?
• What are the priorities?

---

Failed Attempts

Attempt	Why It Failed	Lesson Learned
Relying on memory alone	Context gets compacted, lose progress	Always check git status
Starting random tasks	Forgot what was incomplete	Check system-status.json first
Skipping session initialization	Missed important updates, duplicated work	2-min protocol saves hours

---

Quick Start (< 2 min)

# Step 1: Check current state (30 sec)
git status && git branch --show-current
git log --oneline -5

# Step 2: Check feature status (30 sec)
cat memory-bank/always/system-status.json | jq '.features[] | {name, status, passes}'

# Step 3: Find incomplete work (30 sec)
cat memory-bank/always/system-status.json | jq '.features[] | select(.passes == false)'

# Step 4: Review recent activity (30 sec)
git log --oneline -10
cat memory-bank/always/system-status.json | jq '.recent_fixes[-3:]'

Output: List of incomplete features, recent commits, current priority

---

Complete Protocol

On Every New Session

1. Check Git State

# What branch am I on?
git status
git branch --show-current

# What changed recently?
git log --oneline -5
git diff HEAD~1  # See last commit changes

2. Read System Status

# What features are incomplete?
cat memory-bank/always/system-status.json | jq '.features[] | select(.passes == false) | .name'

# What's the current priority?
cat memory-bank/always/system-status.json | jq '.current_sprint'

# Any blockers?
cat memory-bank/always/system-status.json | jq '.active_blockers'

3. Select Next Task

• Focus on ONE incomplete feature (passes: false)
• Check if there are active blockers
• Read relevant Entry #X from memory-bank/learned/
• Understand context before starting

4. Set Session Goal

• Pick specific, achievable goal
• "Complete Feature X" or "Fix Issue Y"
• NOT "Work on everything" (violates Anthropic incremental progress principle)

---

When Context is Fresh (After Compaction)

Don't Panic: Context compaction is normal and expected

Discovery Protocol:

1. Filesystem Discovery (more reliable than compaction summary):

   git status              # Current changes
   git log --oneline -10   # Recent work
   cat memory-bank/always/system-status.json  # Feature status
   

2. Re-read Critical Files:

   • memory-bank/always/CORE-PATTERNS.md
   • Relevant Entry from memory-bank/learned/
   • Feature-specific files if needed

3. Trust the System:

   • Git history is complete
   • system-status.json is up-to-date
   • Compaction summaries provide context

---

Evidence

Source: Anthropic's Claude 4 Engineering Blog + Best Practices Used In: production (162+ entries, multi-month project) Success: Multi-session continuity maintained across 100+ sessions Time Cost: 2 min per session start Time Saved: 10-30 min per session (vs getting oriented randomly)

Key Insight from Anthropic:

"Focus on incremental progress—making steady advances on a few things at a time rather than attempting everything at once."

---

Integration

Used With:

• system-status.json - Feature tracking
• /session-end protocol - Checkpointing before closing
• memory-bank/learned/ - Entry references
• Git commits - Progress tracking

Complements:

• session-end-checkpoint-skill (if you create it)
• troubleshooting-decision-tree-skill
• Project patterns from CORE-PATTERNS.md

---

Success Criteria

You've mastered this protocol when:

• Can start any session in < 2 min
• Always know what you were working on
• Never duplicate completed work
• Can resume after context compaction without confusion
• Focus on ONE task at a time (incremental progress)

---

Example Session Start

$ git status
On branch feature/new-auth
Changes not staged for commit:
  modified:   src/auth/controller.js

$ cat memory-bank/always/system-status.json | jq '.features[] | select(.passes == false)'
{
  "name": "Authentication_System",
  "status": "implementing",
  "passes": false,
  "entry": 15
}

$ git log --oneline -3
abc1234 feat: Add JWT token validation
def5678 feat: Create auth middleware
ghi9012 feat: Setup auth routes

👉 Goal for this session: Complete JWT token refresh logic
👉 Reference: Entry #15 in memory-bank/learned/
👉 Expected: Mark Authentication_System passes: true by end of session

---

Quick Reference

# Quick session start
git status && \
cat memory-bank/always/system-status.json | jq '.features[] | select(.passes == false)' && \
git log --oneline -5

# If you have a /session-start command configured:
/session-start