---

name: session-continuity-assistant description: This skill should be used when the user asks to "continue session", "resume work", "what was I doing last time", "pick up where I left off", "check recent activity", "what happened since last session", or needs context about previous sessions. version: 1.0.0

Session Continuity Assistant

Purpose

Help users seamlessly resume work by leveraging the framework's multi-resolution memory system. This skill provides deep context about recent activity, project progress, and suggested next actions.

When to Use

• User returns after time away
• User forgot what they were working on
• User wants to understand recent progress
• User needs context from previous sessions
• User asks what they were doing
• User wants to know what changed since last session

Core Workflow

Step 1: Check Current State

Read .session-state.md from the providers directory to understand:

• Last session date and topic
• Active projects
• Current focus
• Pending tasks
• Aggregation status (weekly/monthly logs)

Location: ~/.claude-memory/providers/claude/session-state.md

Step 2: Load Recent Activity

Read appropriate memory level based on time elapsed:

Same day:

• Load daily log: ~/.claude-memory/providers/claude/logs/daily/YYYY.MM.DD.md
• Contains full session details (~150 lines)

Same week (1-7 days ago):

• Load weekly summary: ~/.claude-memory/providers/claude/logs/weekly/YYYY.MM.weekN.md
• Contains aggregated week activities (~100 lines)

Same month (8-30 days ago):

• Load monthly summary: ~/.claude-memory/providers/claude/logs/monthly/YYYY.MM.md
• Contains high-level monthly overview (~30 lines)

Older than 30 days:

• Check aggregation status first
• Inform user if logs need aggregation
• Suggest running /aggregate week or /aggregate month

Step 3: Project Context

If user was working on specific project:

1. Read project context: .projects/[project-name]/.context.md

   • Full project memory
   • Recent decisions
   • Technical details

2. Read project status: .projects/[project-name]/.status.md

   • Current phase/status
   • Roadmap
   • Next actions
   • Blockers

3. Highlight progress: Compare with previous state if available

Step 4: Cross-Provider Check

Check if other providers (LMStudio, Gemini, etc.) have recent activity:

Read: ~/.claude-memory/integration/provider-activities.quick.md

• Shows recent activity from all providers
• Prevents duplicated work
• Enables seamless handoff between providers

Step 5: Present Summary

Format the continuation summary clearly:

Copy**Resuming from**: [Last session date] - [Time elapsed]

**Last Topic**: [What you were doing]

**Progress Made**:
- Key achievement 1
- Key achievement 2
- Key achievement 3

**Decisions/Blockers**:
- Decision 1
- Blocker 1 (if any)

**Current State**:
- Project: [Project name]
- Status: [Status emoji + state]
- Files changed: [Count]

**Suggested Next Steps** (Top 3):
1. [Priority action 1]
2. [Priority action 2]
3. [Priority action 3]

---

Ready to continue? Or would you like to start something new?

Step 6: Offer Choices

After presenting summary, offer clear options:

Option A: Continue where you left off

• Load full project context
• Focus on next priority action

Option B: Check other projects

• Show active projects dashboard
• Allow context switch with /switch [project-name]

Option C: Start new activity

• Suggest using /new command
• Begin fresh context

Memory Structure Reference

Hierarchical Organization

Working Memory (Always loaded):

• .session-state.md - Current session context (~50 lines)
• Always in providers/claude/ directory

Quick Memories (Startup only):

• global-memory.quick.md - User profile condensed (~50 lines)
• project.context.quick.md - Project essentials (~30 lines)
• provider-activities.quick.md - Cross-provider timeline (~10 lines)

Full Memories (On-demand):

• global-memory.safe.md - Complete profile, PII redacted (~165 lines)
• project.context.md - Full project memory
• Daily/weekly/monthly logs

Cloud Memory (If configured):

• GitHub repository: claude-memory-cloud
• Synced across devices
• Contains all logs, projects, global memory

Reading Strategy

To continue session from:

Best Practices

Efficient Context Loading

1. Start minimal: Read session-state first (50 lines)
2. Load progressively: Only load detailed logs if needed
3. Respect token budget: Use quick versions when available
4. Aggregate first: If logs not aggregated, suggest user run /aggregate

Temporal Awareness

Check current date vs last session:

• Same day: "Continuing from earlier today"
• Yesterday: "Continuing from yesterday"
• This week: "Continuing from [day] this week"
• Older: "It's been [N] days since last session"

Suggest aggregation when needed:

• Friday afternoon: "Would you like to run /aggregate week before weekend?"
• End of month: "Month ending soon, consider /aggregate month"

Project Awareness

Understand project state transitions:

• PLANNING → IMPLEMENTING → TESTING → COMPLETED
• Different next actions per state
• Highlight blockers if project stuck

Multi-project scenarios:

• If user worked on multiple projects, prioritize by "last touched"
• Show top 3 active projects
• Allow quick switching with /switch

Resources

For detailed information, see:

• references/memory-structure.md - Complete framework memory layout
• references/continuation-patterns.md - Common resume scenarios and templates

Integration with Commands

This skill complements framework commands:

/continue: Uses quick memories (84-88% token savings) /switch [project]: Changes project context /projects: Lists all active projects /end: Creates detailed session log for continuity

Use this skill when user asks conversationally; use commands when explicit control needed.

Examples

Example 1: Same-Day Continuation

User: "What was I doing?"

Skill loads:

1. session-state.md (last session today, 3 hours ago)
2. Daily log for today
3. Current project context

Response:

Copy**Resuming from**: Today, 3:12 PM (3 hours ago)

**Last Topic**: Implementing cloud memory sync (v2.3)

**Progress Made**:
- Created setup-cloud-memory.sh script
- Documented multi-device workflow
- Added provider-agnostic design

**Next Steps**:
1. Test setup script on second device
2. Merge branch to main
3. Update README with cloud sync docs

Ready to continue testing on second device?

Example 2: Weekly Continuation

User: "What did I work on this week?"

Skill loads:

1. session-state.md (last session 3 days ago)
2. Weekly summary for current week
3. Provider activities (check other providers)

Response:

Copy**Resuming from**: Monday (3 days ago)

**This Week's Focus**:
- Golfleet L2 reorganization (Monday)
- Cloud memory infrastructure planning (Tuesday-Wednesday)
- No activity Thursday-Friday

**Key Achievements**:
- Mapped 3 operational cores for Golfleet
- Designed cloud memory architecture (v2.3)
- Created comprehensive planning docs

**Open Items**:
- [ ] Complete Golfleet RACI matrix
- [ ] Test cloud memory on multiple devices

Which would you like to continue: Golfleet or Cloud Memory?

Example 3: Cross-Provider Handoff

User: "Continue where I left off"

Skill loads:

1. session-state.md (last Claude session 2 days ago)
2. provider-activities.quick.md (shows LMStudio session yesterday!)
3. Load LMStudio logs to see what was done

Response:

Copy**Note**: Your last session was with **LMStudio** yesterday, not Claude.

**LMStudio Session** (Yesterday, 10:45 AM):
- Worked on data analysis pipeline
- Processed 3 datasets
- Generated visualization scripts

**Claude Session** (2 days ago):
- Framework enhancement planning

Would you like to:
1. Continue LMStudio work (data analysis)
2. Return to Claude work (framework)
3. See full provider timeline

Troubleshooting

Issue: Can't find recent logs Solution: Check aggregation status in session-state.md; suggest /aggregate

Issue: Project context missing Solution: User may not have used /switch [project] yet; guide to create project context

Issue: Memory files not found Solution: Framework may not be fully set up; suggest running /start first

Issue: Cloud memory out of sync Solution: Check if cloud sync configured; suggest pulling latest with git pull in ~/.claude-memory-cloud/

Version History

v1.0.0 (2025-12-28)

• Initial release
• Multi-provider awareness
• Cloud memory support
• Progressive loading strategy