gemini-memory-sync（Gemini 记忆 sync）

Gemini Memory Sync

Documentation Delegation

Documentation Source: For authoritative memory/import syntax and current features, query gemini-cli-docs skill. This skill provides sync patterns; gemini-cli-docs provides official Gemini CLI documentation.

Overview

This skill provides patterns for keeping Claude Code (CLAUDE.md) and Gemini CLI (GEMINI.md) memory files synchronized. The core principle is CLAUDE.md as source of truth with GEMINI.md importing and adding overrides.

When to Use This Skill

Keywords: sync memory, sync context, claude.md gemini.md, memory import, context drift, @import, memport

Use this skill when:

• Setting up GEMINI.md for a new project
• Detecting drift between memory files
• Understanding the import syntax
• Troubleshooting sync issues

Core Principle: Single Source of Truth

CLAUDE.md (Source of Truth)
    │
    │ @import
    ▼
GEMINI.md (Imports + Overrides)

Why CLAUDE.md is the source:

• Claude Code is the primary development environment
• CLAUDE.md is already established in most projects
• Single point of update reduces maintenance burden
• Git history shows context evolution in one place

GEMINI.md Structure

Recommended Template

# GEMINI.md

@CLAUDE.md

## Gemini-Specific Overrides

You are Gemini CLI. Your unique capabilities:
- Large context window (Flash) / Very large (Pro)
- Interactive PTY shell (vim, git rebase -i, htop)
- Checkpointing with instant rollback
- Policy engine for tool control
- Native Google Cloud authentication

### When to Use Your Strengths

- **Bulk analysis**: Use your large context for codebase-wide exploration
- **Interactive tools**: Handle vim, git interactive commands
- **Risky operations**: Use sandbox and checkpointing
- **Second opinions**: Provide independent validation

### Model Selection

- Use **Flash** for bulk analysis and simple tasks
- Use **Pro** for complex reasoning and very large contexts

Import Syntax

Gemini CLI uses @ prefix for imports (memport):

# Import entire file
@CLAUDE.md

# Import relative path
@./docs/conventions.md

# Import from parent
@../shared/COMMON.md

Note: Unlike CLAUDE.md's flexible import, GEMINI.md's memport has:

• Maximum import depth: 5 levels
• Circular import detection
• File access validation

Drift Detection

Manual Detection

# Quick diff (ignoring Gemini-specific sections)
diff <(grep -v "^## Gemini-Specific" CLAUDE.md) <(grep -v "^## Gemini-Specific\|^@" GEMINI.md)

Hash-Based Detection

# Store hash of CLAUDE.md
claude_hash=$(md5sum CLAUDE.md | cut -d' ' -f1)

# Store in sync state
echo "{\"claude_hash\": \"$claude_hash\", \"last_sync\": \"$(date -u +%Y-%m-%dT%H:%M:%SZ)\"}" > .claude/temp/sync-state.json

Check for Drift

# Compare current hash to stored
current_hash=$(md5sum CLAUDE.md | cut -d' ' -f1)
stored_hash=$(cat .claude/temp/sync-state.json 2>/dev/null | jq -r '.claude_hash // ""')

if [ "$current_hash" != "$stored_hash" ]; then
  echo "CLAUDE.md has changed since last sync"
fi

Sync Patterns

Pattern 1: Import-Based (Recommended)

GEMINI.md simply imports CLAUDE.md - no sync needed:

# GEMINI.md
@CLAUDE.md

## Gemini-Specific
{overrides here}

Pros:

• No sync maintenance
• Always up-to-date
• Single source of truth enforced

Cons:

• GEMINI.md must start with @import
• Can't selectively import sections

Pattern 2: Section-Based Sync

Copy specific sections from CLAUDE.md:

# Extract specific sections
conventions=$(sed -n '/^## Conventions/,/^## /p' CLAUDE.md | head -n -1)
build_commands=$(sed -n '/^## Build/,/^## /p' CLAUDE.md | head -n -1)

# Rebuild GEMINI.md
cat > GEMINI.md << EOF
# GEMINI.md

## Conventions (synced from CLAUDE.md)
$conventions

## Build Commands (synced from CLAUDE.md)
$build_commands

## Gemini-Specific Overrides
{your overrides}
EOF

Pros:

• Selective control
• Can transform content

Cons:

• Requires manual sync
• Can drift easily

Pattern 3: Template Generation

Generate GEMINI.md from CLAUDE.md with transformations:

# Transform CLAUDE.md to GEMINI.md
cat CLAUDE.md | \
  sed 's/Claude Code/Gemini CLI/g' | \
  sed 's/claude/gemini/g' > GEMINI.md

# Append Gemini-specific section
cat >> GEMINI.md << 'EOF'

## Gemini-Specific Overrides
{overrides}
EOF

Common Issues

Issue: Import Not Working

Symptom: Gemini doesn't see CLAUDE.md content

Fix: Ensure correct path syntax

# Correct
@CLAUDE.md
@./CLAUDE.md

# Incorrect
@/CLAUDE.md  (absolute paths may fail)

Issue: Circular Import

Symptom: Error about circular references

Fix: Don't have CLAUDE.md import GEMINI.md

Issue: Import Depth Exceeded

Symptom: Nested imports not loading

Fix: Memport has max depth of 5. Flatten import chain.

Issue: Context Drift

Symptom: Gemini behaves differently than Claude

Fix:

1. Use /sync-context command
2. Or rebuild GEMINI.md with @import pattern

Best Practices

1. Use @Import Pattern

Always prefer import over copy:

# GEMINI.md - Good
@CLAUDE.md

## Gemini-Specific
...

2. Keep Overrides Minimal

Only override what's truly Gemini-specific:

• Model selection guidance
• Interactive shell instructions
• Sandbox usage patterns

3. Document What's Synced

If using section-based sync, note the source:

## Conventions (synced from CLAUDE.md on 2025-11-30)

4. Validate After Sync

Test that Gemini understands the context:

gemini "What are the project conventions?" --output-format json

5. Regular Drift Checks

Include in CI or pre-commit:

# In CI
./scripts/check-memory-drift.sh

Sync Workflow

Initial Setup

# 1. Ensure CLAUDE.md exists
if [ ! -f "CLAUDE.md" ]; then
  echo "CLAUDE.md not found. Create it first."
  exit 1
fi

# 2. Create GEMINI.md with import
cat > GEMINI.md << 'EOF'
# GEMINI.md

@CLAUDE.md

## Gemini-Specific Overrides

You are Gemini CLI with unique capabilities:
- Large context window (exceeds typical LLM limits)
- Interactive PTY shell
- Checkpointing with rollback
- Policy engine

Prioritize tasks that leverage these strengths.
EOF

# 3. Initialize sync state
mkdir -p .claude/temp
echo "{\"claude_hash\": \"$(md5sum CLAUDE.md | cut -d' ' -f1)\", \"last_sync\": \"$(date -u +%Y-%m-%dT%H:%M:%SZ)\"}" > .claude/temp/sync-state.json

echo "GEMINI.md created with @import to CLAUDE.md"

Manual Sync

# Check if sync needed
if [ "$(md5sum CLAUDE.md | cut -d' ' -f1)" != "$(cat .claude/temp/sync-state.json | jq -r '.claude_hash')" ]; then
  echo "CLAUDE.md has changed. If using @import, no action needed."
  echo "If using section-based sync, rebuild GEMINI.md sections."

  # Update sync state
  echo "{\"claude_hash\": \"$(md5sum CLAUDE.md | cut -d' ' -f1)\", \"last_sync\": \"$(date -u +%Y-%m-%dT%H:%M:%SZ)\"}" > .claude/temp/sync-state.json
fi

Related Skills

• gemini-workspace-bridge - Overall workspace architecture
• gemini-context-bridge - Legacy context sharing

Related Commands

• /sync-context - Trigger manual sync check

Test Scenarios

Scenario 1: Initial Setup

Query: "How do I set up GEMINI.md to use CLAUDE.md?" Expected Behavior:

• Skill activates on "sync memory" or "claude.md gemini.md"
• Provides @import syntax and template Success Criteria: User receives working GEMINI.md template with @CLAUDE.md import

Scenario 2: Drift Detection

Query: "How do I check if my memory files are out of sync?" Expected Behavior:

• Skill activates on "context drift" or "sync"
• Provides hash-based detection method Success Criteria: User receives drift detection script

Scenario 3: Import Issues

Query: "My GEMINI.md @import isn't working" Expected Behavior:

• Skill activates on "import" troubleshooting
• Provides common issues and fixes Success Criteria: User receives troubleshooting steps for path syntax

Version History

• v1.1.0 (2025-12-01): Added MANDATORY section, Test Scenarios, Version History
• v1.0.0 (2025-11-25): Initial release