원클릭으로
session-notes-writer
Proactively suggests and writes session notes at end of work sessions for multi-day task continuity
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
메뉴
Proactively suggests and writes session notes at end of work sessions for multi-day task continuity
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Use when deciding how to manage, audit, sync, enable, disable, or troubleshoot skills and agents with the skillshare CLI across Claude Code, Codex, Gemini, OpenCode, Kiro, Goose, Pi, Antigravity, or project-local .skillshare setups. Especially use for questions about global vs project mode, disabled-by-default policies, .skillignore/.agentignore, target drift, broken symlinks, duplicate skills, sync dry-runs, or whether to use skillshare versus a runtime-native tool such as Codex skill-manager.
Use when creating git commits or reviewing commit messages. Ensures messages follow the Conventional Commits specification v1.0.0, which enables automated changelogs and semantic versioning. Invoked when the user wants to commit changes, craft a commit message, or validate an existing one.
Use when user references previous work phases, tasks, or discussions from prior sessions not present in current conversation context.
Use when designing or reviewing Python classes, deciding where logic belongs, or recognizing antipatterns like helper classes, utility bags, and god objects.
Use when crafting prompts that require factual accuracy, working with long documents, generating claims that need verification, or when outputs must be auditable and grounded in source material
Use when transitioning between work sessions and wanting to pick up where the previous session left off.
| name | session-notes-writer |
| description | Proactively suggests and writes session notes at end of work sessions for multi-day task continuity |
This skill teaches Claude to proactively write session notes at the end of complex work sessions to enable seamless continuation in future sessions.
1. End of work session (user signals)
2. Work left incomplete
3. Complex multi-session work
Skip session notes for:
Before this skill:
With this skill:
When user signals end of session, evaluate:
IF (task incomplete OR multi-day work OR blocker encountered)
AND (NOT trivial change)
AND (NOT fully documented in git/CLAUDE.md)
THEN
Suggest: "Should I write a session note for when you continue this work?"
Factors to consider:
✅ Write session note IF:
❌ Skip session note IF:
Watch for signals:
Ask yourself:
Offer proactively:
"This looks like a good stopping point. Should I write a session note
capturing what we completed and what to continue next session? This
will help pick up the work seamlessly tomorrow."
If user confirms, proceed to Step 4.
If user declines, respect it:
User: "No need, I've got it"
Claude: "Understood. Feel free to ask if you change your mind."
Gather context:
git status, git diff --stat)Create .claude-sessions/ directory if needed:
mkdir -p .claude-sessions
Generate filename:
.claude-sessions/YYYY-MM-DD-<topic>.md
Use template from session-notes convention:
# Session: <Topic> - <Date>
## Summary
[2-3 sentences: What was worked on this session]
## Completed This Session
- [Specific accomplishment 1]
- [Specific accomplishment 2]
- [Commit references if applicable]
## Decisions Made
- [Decision 1 with rationale]
- [Decision 2 with rationale]
## Next Session
- [ ] [Specific task to continue]
- [ ] [Another task]
- [ ] [Follow-up item]
## Blockers / Open Questions
- [Blocker 1: description, potential solutions]
- [Question 1: what needs clarification]
## Notes / Discoveries
- [Informal observation 1]
- [Project quirk discovered]
- [Useful reference link]
Write the file:
# Use Write tool to create session note
Tell user what you created:
"Session note written to .claude-sessions/2025-12-21-context-recovery.md
Summary:
- Completed: git-context-recovery skill, context-recovery convention
- Next session: Create session-notes-writer skill, enhance lessons-learned
- No blockers
This will help us pick up exactly where we left off next time."
User: "I need to stop here, I'll continue tomorrow"
Claude evaluates:
Verdict: Write session note
Claude suggests:
"This looks like a good stopping point. Should I write a session note
capturing what we completed (csv_parser, monitor modules) and what to
continue tomorrow (cli.py with Click)?"
User: "Yes please"
Claude writes:
.claude-sessions/2025-12-18-modular-refactor.md
# Session: Modular Architecture Refactor - 2025-12-18
## Summary
Refactored monolithic sniffer.py into modular structure. Created csv_parser.py and monitor.py with core classes. CLI migration pending.
## Completed This Session
- Created csv_parser.py with RekordboxCSVParser class
- Created monitor.py with RekordboxMIDISniffer class
- Updated imports in sniffer.py
- All existing functionality preserved
## Decisions Made
- **Use Click for CLI instead of argparse**
- Rationale: Better UX, cleaner command structure, used in modern Python CLIs
- Reference: https://click.palletsprojects.com/
## Next Session
- [ ] Create cli.py with Click command structure
- [ ] Migrate all CLI commands from sniffer.py
- [ ] Update tests to import from new modules
- [ ] Update CLAUDE.md with new architecture
## Blockers / Open Questions
- None currently
## Notes / Discoveries
- Separation makes testing much easier (can mock CSV parser independently)
- monitor.py is ~200 lines, good size for a focused module
Claude confirms:
"Session note written to .claude-sessions/2025-12-18-modular-refactor.md
Tomorrow we'll pick up with creating cli.py using Click."
Context: Claude spent session researching bidirectional MIDI monitoring, hit technical blocker.
User: "This is getting complex, let's pause and think about it"
Claude evaluates:
Verdict: Write session note
Claude suggests:
"Good idea to pause. Should I write a session note capturing what we
learned about bidirectional monitoring and the technical limitations we
discovered? This will help us make an informed decision next session."
User: "Yes, that'd be helpful"
Claude writes:
.claude-sessions/2025-12-19-bidirectional-research.md
# Session: Bidirectional MIDI Monitoring - 2025-12-19
## Summary
Researched bidirectional MIDI monitoring for capturing Rekordbox→controller LED feedback. Discovered standard mido/rtmidi cannot capture these messages due to macOS CoreMIDI architecture.
## Completed This Session
- Analyzed mido and rtmidi source code for monitoring capabilities
- Tested IOPort for bidirectional capture (unsuccessful)
- Researched CoreMIDI spy techniques and virtual through-ports
- Documented findings in tasks/121925_bidirectional_research/
## Decisions Made
- **Not implementing bidirectional in Phase 1**
- Rationale: Requires custom CoreMIDI driver or complex virtual port architecture
- Complexity doesn't justify benefit for current use case
- Controller→computer monitoring (IOPort) is sufficient for primary functionality
## Next Session
- [ ] Decide: Drop bidirectional entirely OR defer to future version?
- [ ] Continue with message grouping (Phase 2) if proceeding
- [ ] Update CLAUDE.md with architecture decision
## Blockers / Open Questions
- **Technical blocker: Standard MIDI monitoring can't capture Rekordbox→controller**
- Potential solutions:
1. Custom CoreMIDI driver (high complexity)
2. Virtual through-port requiring user setup (poor UX)
3. Scope out bidirectional (simplest, meets 90% use case)
- Recommendation: Option 3 for v1, revisit for v2 if users request it
## Notes / Discoveries
- IOPort.get_count() is reliable for input device enumeration
- "MIDI monitoring" ≠ "MIDI sniffing" in technical terms
- Rekordbox uses proprietary LED protocol (not standard MIDI feedback)
- macOS CoreMIDI doesn't expose "spy" capability to applications
Context: Fixed typo in CLAUDE.md
User: "Okay, that's done"
Claude evaluates:
Verdict: Skip session note
Claude response:
"Change committed. No session note needed for this simple fix."
Users can ask Claude to update session notes:
User: "Add to today's session note that we also discovered X"
Claude:
# Find today's session note
ls .claude-sessions/2025-12-21-*.md
# Read current content
# Use Edit tool to add discovery to "Notes / Discoveries" section
Confirm:
"Updated .claude-sessions/2025-12-21-context-recovery.md with discovery about X"
DO:
DON'T:
.claude-sessions/ directory if missingWhen uncertain if session note is warranted, ask:
Session notes feed into the context-recovery protocol:
User (next session): "Continue where we left off"
Claude (with git-context-recovery skill):
# Priority 1: Check git
git log --oneline -5
# Finds some commits but no phase documentation
# Priority 4: Check session notes
ls -lt .claude-sessions/ | head -1
# Shows: 2025-12-20-bidirectional-research.md
cat .claude-sessions/2025-12-20-bidirectional-research.md
# Reads "Next Session" section
Claude responds:
"Picking up from yesterday's session. We decided to defer bidirectional
monitoring (technical blocker). Next step is implementing message grouping
(Phase 2). Should I proceed?"
Result: Seamless continuation with full context
Session notes convention is project-level (not global). To enable:
mkdir -p .claude-sessions
cat > .claude-sessions/README.md <<'EOF'
# Claude Session Notes
Session notes for multi-day work continuity.
Convention: ~/.claude/conventions/session-notes.md
EOF
## Session Notes
- Uses: ~/.claude/conventions/session-notes.md
- Location: .claude-sessions/
- Claude proactively writes session notes for complex multi-day work
.claude-sessions/
(Session notes are ephemeral; commit if you want history)
~/.claude/conventions/session-notes.md - Session note structure and when to use~/.claude/conventions/context-recovery.md - How session notes fit context recovery~/.claude/skills/git-context-recovery/SKILL.md - Checks session notes as part of recoveryThis skill makes session notes happen automatically, ensuring multi-session work never loses context.