Memory Keeper

Memory Keeper Skill

Maintains and updates Aaron's Core Memory file (~/.claude/CLAUDE.md) with project updates, session notes, and observed development patterns.

Purpose

This skill automates the maintenance of Aaron's persistent context by:

• Analyzing recent work and conversations
• Updating project status sections
• Recording major developments in session notes
• Tracking development patterns and observations
• Ensuring project metadata stays current

When to Use

Invoke this skill when:

1. After major project milestones - Phase completions, releases, significant commits
2. End of work sessions - Capture what was accomplished
3. Project status changes - Version bumps, status updates, new phases
4. User explicitly requests - "Update my memory", "Record this in CLAUDE.md"
5. Significant infrastructure changes - New tools, scripts, workflows

What Gets Updated

Project Sections

For each active project (xai, cc-flow, stoch, screener, etc.):

• Version/Status: Current version tags, phase completions
• Recent Achievements: Latest completed work (last 3-7 days)
• Key Metrics: Test coverage, documentation stats, quality scores
• File Locations: Important files added or moved
• Next Milestones: Upcoming phases or goals

Session Notes

Updates the "Session Notes" section with:

• Major Developments: Date-stamped significant events
• Patterns Observed: New workflows, productivity insights, technical patterns
• Project Status Summary: Current state of all active projects
• Infrastructure Updates: New scripts, tools, configurations

Metadata

• Last Updated: Current date timestamp
• Quick Commands: New or updated command patterns
• Recovery Procedures: Lessons learned from incidents

Analysis Process

Step 1: Gather Context

# Analyze recent git activity
git -C <project-path> log --oneline --since="7 days ago"
git -C <project-path> describe --tags

# Check project status files
find <project-path> -name "STATUS*.md" -o -name "PHASE_*.md" -o -name "README.md"

# Analyze conversation history
python ~/cc-flow/tools/workspace-manager/src/claude-conversation-viewer.py stats
python ~/cc-flow/tools/workspace-manager/src/claude-conversation-viewer.py list --limit 10 --sort date

Step 2: Identify Changes

• New tags/versions (git describe output changes)
• Completed phases (new PHASE_*_COMPLETE.md files)
• Recent commits (git log analysis)
• Large conversations (indicates major work sessions)
• New files/directories (project structure changes)

Step 3: Update CLAUDE.md

For Project Updates:

• Read current project section from CLAUDE.md
• Compare with latest git status, tags, and documentation
• Update version, status, achievements, and key files
• Preserve all existing information, only augment

For Session Notes:

• Read current Session Notes section
• Add new major developments with date stamps
• Update patterns if new workflows emerged
• Update project status summary
• Add infrastructure updates if tools/scripts created

For Metadata:

• Update "Last Updated" timestamp
• Add new quick commands if applicable

Step 4: Formatting Rules

Date Formats:

• Session notes header: Session Notes (Nov 12-20, 2025)
• Major developments: **Nov 20**: Description
• Project sections: (2025-11-20) or Nov 20, 2025

Markdown Conventions:

• Use ✅ for completed items
• Use **bold** for emphasis
• Use code blocks for commands
• Use - for bullet lists
• Preserve existing structure and formatting

Conciseness:

• Keep entries factual and brief
• Focus on outcomes, not process
• Include metrics when relevant (LOC, test count, file sizes)
• Use present tense for current status

Example Updates

Example 1: After stoch beta1 release

**Current Status** (2025-11-18 - Beta1 Release):
- ✅ Beta1 released (v2.0.0-beta1) - Nov 18, 2025
- ✅ Complete documentation (~4,500 lines: README, API ref, migration guide)
- ✅ 149 tests passing, zero warnings
- ✅ Git tagged: v2.0.0-beta1

Example 2: Session note addition

**Major Developments**:
...
- **Nov 20**: sudo askpass helper created - SSH/Termux compatible, kdialog GUI, /dev/tty fallback

Example 3: Pattern observation

**Patterns Observed**:
...
- Rapid multi-model integration: 3 AI backends integrated in single 4-hour session

Safety Guidelines

1. Always Read First: Never update CLAUDE.md without reading current content
2. Preserve Everything: Only add/augment, never delete unless explicitly instructed
3. Maintain Structure: Keep existing section headers and organization
4. Verify Dates: Ensure date stamps are accurate and consistent
5. Check References: Verify file paths and references exist before adding
6. No Assumptions: Base updates only on verifiable evidence (git logs, files, conversation history)

Integration with Slash Command

When invoked via /update-memory:

1. Analyze recent work (last 7 days unless specified)
2. Identify significant changes across all projects
3. Update CLAUDE.md with findings
4. Report what was updated to user

Workflow Example

# User invokes: /update-memory

# Skill executes:
# 1. Read current CLAUDE.md
# 2. Check git activity for all active projects
# 3. Analyze recent conversations
# 4. Find new documentation files
# 5. Update relevant sections
# 6. Report changes made

Output:
"Updated CLAUDE.md with:
- stoch: Version updated to v2.0.0-beta1+28 (28 post-beta commits)
- Session notes: Added Nov 20 developments (askpass helper, cc-term deletion)
- Patterns: Added project pruning observation
- Last Updated: 2025-11-20"

Error Handling

• If CLAUDE.md is corrupted, back up to ~/.claude/CLAUDE.md.backup.$(date +%s) before editing
• If uncertain about project status, ask user for clarification
• If dates are ambiguous, default to current date
• If sections are missing, add them following existing structure

Quality Checks

Before finalizing updates:

• ✅ All file paths referenced actually exist
• ✅ Version tags match git describe output
• ✅ Dates are chronologically consistent
• ✅ Markdown formatting is valid
• ✅ No duplicate entries
• ✅ All active projects have current status