name	session-manager
description	Manages work sessions with pause/resume workflow. Triggers on "покажи сесії", "show sessions", "список сесій", "list sessions", "що далі", "what's next", "які плани", "what are the plans", "що по планам", "які цілі", "what are the goals", "що робимо", "what should we do", "де ми", "where are we", "продовжити", "continue", "resume", "давай", "let's go", "пауза", "pause", "зупинись", "stop", "статус", "status", "що зроблено", "what's done", "прогрес", "progress", "перемкнутись", "switch", "заархівувати", "archive". Replaces NEXT_SESSION_TODO.md with structured session files in planned/active/paused/completed dirs. Auto-saves progress, bilingual EN/UA.
allowed-tools	Read, Write, Glob, TodoWrite, Bash(git:*)

Session Manager

Overview

Session Manager provides automatic session state management and seamless pause/resume workflows. It replaces bulky orchestration skills (task-orchestrator, parallel-coordinator, epic-orchestrator) with a lightweight, auto-saving approach.

Key capabilities:

Auto-save: Captures session state after every TodoWrite completion
Explicit pause: User-triggered checkpoints with bilingual summaries
Context restoration: Seamless resume with full todo/agent/artifact state
Bilingual support: EN/UA commands and content
Compact format: Single markdown file per session (tables, checkboxes, admonitions)

Session storage: .claude/sessions/{active|paused|completed}/YYYY-MM-DD-{slug}.md

Core Workflow

User starts work
    ↓
Auto-save after each TodoWrite completion
    ↓
User says "pause" / "пауза"
    ↓
Generate summary → save to paused/
    ↓
[New conversation]
    ↓
User says "resume" / "продовжити" OR `claude --continue`
    ↓
Load session → restore TodoWrite → continue work
    ↓
Mark complete → move to completed/

When to Use This Skill

Session Manager автоматично активується на природній мові (EN/UA). Використовує .claude/sessions/{planned|active|paused|completed}/ для управління сесіями.

Trigger Phrases

1. Show Sessions (Показати сесії)

EN: "show sessions", "list sessions", "what sessions do we have", "available sessions", "all sessions", "what can we work on" UA: "покажи сесії", "список сесій", "які сесії є", "доступні сесії", "всі сесії", "над чим можна працювати", "що по сесіям"

Behavior: List all sessions by status (planned/active/paused/completed) with progress bars

2. What's Next / Plans (Що далі / Плани)

EN: "what's next", "what should we do", "what are the plans", "what's the plan", "next steps", "what to work on", "what are we doing", "what are the goals", "what's on the agenda", "what needs to be done", "what's pending", "what are our priorities" UA: "що далі", "що робимо", "які плани", "що по планам", "наступні кроки", "над чим працювати", "який план", "що робити", "які цілі", "що в планах", "що треба зробити", "що залишилось", "що у черзі", "які пріоритети"

Behavior:

If active session exists → show status
If no active session → show planned sessions menu

3. Session Status (Статус сесії)

EN: "session status", "where are we", "what's the progress", "current status", "how far are we", "show progress", "what have we done", "what's completed", "what's left" UA: "статус сесії", "де ми", "який прогрес", "поточний статус", "що зроблено", "як справи", "покажи прогрес", "що виконано", "що готово", "що залишилось"

Behavior: Show detailed status of active session with progress bars, completed/remaining tasks

4. Continue/Resume (Продовжити)

EN: "continue", "resume", "let's continue", "pick up where we left off", "resume session", "continue work", "let's go", "let's work", "start working", "pick up", "carry on" UA: "продовжити", "продовж", "давай продовжимо", "продовжуємо", "відновити роботу", "поїхали", "давай працювати", "починаємо", "давай далі", "давай"

Behavior:

If paused sessions exist → prompt to select
If no paused → show planned sessions
Auto-resume last active if only one

5. Continue Specific Session (Продовжити конкретну)

EN: "continue [name]", "resume [name]", "work on [name]", "let's do [name]", "start [name]", "do [name]" UA: "продовжити [назва]", "давай [назва]", "працювати над [назва]", "почати [назва]", "робити [назва]"

Examples:

"давай спринт 1" → Load sprint-1-ux-fixes
"continue backend" → Load backend-code-quality
"work on testing" → Load testing-infrastructure

Behavior: Fuzzy match session name, confirm if ambiguous

6. Pause (Пауза)

EN: "pause", "stop", "that's it for now", "save progress", "stop for today", "break", "done for now", "that's all", "enough for now", "let's stop", "take a break", "that's it" UA: "пауза", "зупинись", "все на сьогодні", "зберегти прогрес", "перерва", "готово на сьогодні", "все", "достатньо", "хватит", "зупинимось", "зробимо перерву"

Behavior: Save active session to paused/, show resume instructions

7. Switch Session (Перемкнутись)

EN: "switch to [name]", "work on [name] instead", "let's do something else", "change session", "do something different" UA: "перемкнутись на [назва]", "краще попрацюємо над [назва]", "давай інше", "змінити сесію", "щось інше", "давай по-іншому"

Behavior: Pause current session, resume target session

8. Archive (Архівувати)

EN: "archive this session", "this is done", "mark as complete", "finish this session", "complete session", "we're done", "session complete" UA: "заархівувати сесію", "це готово", "позначити завершеним", "закінчити сесію", "завершити сесію", "ми закінчили", "сесія готова"

Behavior: Move session to completed/, mark status ✅

9. Overview (Огляд)

EN: "overview", "big picture", "overall status", "where do we stand", "summary", "what's the situation" UA: "огляд", "загальна картина", "загальний статус", "де ми стоїмо", "підсумок", "яка ситуація", "що по-загалом"

Behavior: Show high-level summary across all sessions

Auto-Trigger Scenarios

After TodoWrite updates (automatic):

When task marked as completed
When new tasks added
Every 5-7 task completions (checkpoint)

On conversation end (automatic):

User says goodbye: "bye", "пока", "до побачення", "все"
Long inactivity (if detectable)
SessionEnd hook triggers

On context switch (automatic):

User mentions different session name → offer to switch
Example: working on Sprint 1, user says "давай backend" → auto-switch prompt

Do NOT Use This Skill For

Trivial tasks (complexity score 0-2, <3 todos)
Read-only operations (no state to save)
Simple Q&A (no workflow to resume)

Session Name Matching

Smart fuzzy keyword matching allows users to reference sessions naturally:

User says	Matches
"спринт" / "sprint"	All sprint-* sessions
"спринт 1" / "sprint 1"	sprint-1-ux-fixes
"спринт 2" / "sprint 2"	sprint-2-ux-improvements
"бекенд" / "backend"	backend-code-quality
"тести" / "testing"	testing-infrastructure
"ux"	sprint-1, sprint-2, sprint-3

Ambiguity resolution: When multiple matches found, prompt user to select:

Found 3 sessions matching "sprint":
1. Sprint 1: UX Fixes [10/14 done, 71%, 11.5h left]
2. Sprint 2: UX Improvements [0/7 done, 18h]
3. Sprint 3: UX Polish [0/4 done, 7h]

Which one? (type number or more specific name)

Response Templates

List Sessions Response

📊 Доступні сесії:

🔴 АКТИВНІ (1):
  → Sprint 1: UX Fixes [10/14, 71%, 11.5h left]

📅 ЗАПЛАНОВАНІ (4):
  1. Sprint 2: UX Improvements [0/7, 18h]
  2. Sprint 3: UX Polish [0/4, 7h]
  3. Backend Code Quality [0/10, 25h]
  4. Testing Infrastructure [0/4, 20h]

⏸️ ПРИЗУПИНЕНІ (2):
  • Sprint1 Audit (Oct 31)
  • Old Sprint1 Work (Oct 30)

✅ ЗАВЕРШЕНІ (0)

Щоб продовжити: "давай спринт 1" або "продовжити backend"

Session Status Response

📊 Sprint 1: UX Fixes

Ціль: Критичні UX виправлення
Прогрес: 10/14 tasks (71%) ━━━━━━━━━⚊⚊⚊⚊
Час: 6h витрачено / 11.5h залишилось

✅ Зроблено (10):
  ✓ Dashboard duplication (7d395e2)
  ✓ Sidebar auto-expansion (c777c9e)
  ✓ Status badge system (4348208)
  ✓ Keyboard navigation docs (6bd9c99)
  ...

🔲 Залишилось (4):
  1. Badge tooltips (0.5h)
  2. ARIA labels (3h)
  3. Color contrast (0.5h)
  4. Empty states (4h)

Наступне: Add tooltips to notification badges

Готовий продовжити? Скажи "давай" або "продовжити"

Resume Prompt

✅ Відновлюю сесію: Sprint 2 UX Improvements

Ціль: Оптимізація таблиць та мобільна версія
Прогрес: 0/7 tasks (0%)
Заплановано: 18h

Задачі:
  1. Messages Table optimization (2h)
  2. Mobile sidebar (4h)
  3. Mobile tables (3h)
  ...

Почнемо з: Messages Table column widths

Готовий? Скажи "готовий" або "поїхали"

Pause Confirmation

💾 Сесію збережено: .claude/sessions/paused/sprint-1-ux-fixes.md

Прогрес: 10/14 tasks (71%) • 6h витрачено / 11.5h залишилось
Наступне: Badge tooltips

Щоб продовжити: "продовжити спринт 1" або "resume sprint 1"

Auto-Save Logic

When Auto-Save Triggers

After TodoWrite completion (primary trigger):

Task marked as completed
Batch completion (3-5 tasks done)
Significant milestone reached

On conversation end (secondary trigger):

User signals goodbye
Inactivity detected
Natural conclusion

What Auto-Save Captures

High-priority (always):

Current todo list state
Last completed action
Next pending task

Medium-priority (if changed):

Session goal/approach
Active blockers
Agent outputs

Low-priority (on explicit pause only):

Full artifact list
Decision history
Detailed notes

Auto-Save vs Explicit Pause

Trigger	Location	Detail Level	User Prompt
Auto-save	`active/`	High-priority only	Silent
Explicit `/pause`	`paused/`	Full detail	Bilingual message

Rationale: Auto-save is lightweight (don't interrupt flow), explicit pause is comprehensive (user expects checkpoint).

TodoWrite Integration

Bi-Directional Sync

TodoWrite → Session (save):

On task completion → update session file
On new tasks → expand todo section
On status change → reflect in session

Session → TodoWrite (restore):

On /resume → recreate todo list with exact statuses
Verify counts match (completed vs pending)
Mark first pending task as in_progress

State Validation

Before pause:

Check: Only 0-1 tasks should be in_progress
If multiple in_progress: Complete current or mark as blocked
If inconsistent: Add warning to session file

After resume:

Verify: TodoWrite state matches session file
If mismatch: Session file is source of truth
Output warning: "Todo state diverged. Restored from session."

Session State Format

All sessions use the same markdown template with EN headers and bilingual content.

Template: See references/SUMMARY_TEMPLATES.md

Key sections:

Context - 2-word snippets in table format
Todo - Checkboxes synced with TodoWrite
Agents Used - Table with artifact links
Next Actions - Immediate steps + blockers + questions
Artifacts - File list with paths and metrics

Admonitions:

> [!NOTE] - Session metadata, auto-sync indicators
> [!TIP] - Todo list headers, workflow hints
> [!WARNING] - Blockers, resume instructions
> [!IMPORTANT] - Key decisions, constraints

Example (compact):

# Session: BaseCRUD Refactor

**Status**: ⏸️ Paused

## Context

| What | State |
|------|-------|
| Goal | CRUD inheritance pattern |
| Approach | Migrate 8 services to BaseCRUD |
| Blocker | None |

## Todo

- [x] atom_crud.py (-50 LOC)
- [x] agent_crud.py (-18 LOC)
- [ ] message_crud.py (est. 20 min)

**Progress**: 2/8 tasks (25%)

## Next Actions

> [!WARNING]
> Resume: `/resume` або `/продовжити`

1. message_crud.py migration (20 min)
2. topic_crud.py migration (25 min)

Session Lifecycle

1. Session Creation

Trigger: First TodoWrite call OR user starts complex task

Actions:

Generate slug from task description
Create session file in active/
Initialize with goal, approach, empty todo list

Location: .claude/sessions/active/YYYY-MM-DD-{slug}.md

2. Active Session

State: Work in progress, auto-save enabled

Updates:

After each TodoWrite completion → update progress stats
After agent delegation → add to "Agents Used" table
On file changes → update "Artifacts" section
Every 3-5 completions → full checkpoint

Location: Stays in active/

3. Paused Session

Trigger: Explicit /pause OR conversation end

Actions:

Generate comprehensive summary (all sections)
Move file: active/{slug}.md → paused/{slug}.md
Output bilingual resume instruction

Location: .claude/sessions/paused/YYYY-MM-DD-{slug}.md

4. Resumed Session

Trigger: Explicit /resume OR claude --continue with paused session

Actions:

Load session file from paused/
Restore TodoWrite state
Verify artifacts exist
Move back: paused/{slug}.md → active/{slug}.md
Output context summary + readiness check

Location: Back to active/

5. Completed Session

Trigger: All todos completed OR user marks as done

Actions:

Final summary generation
Move file: active/{slug}.md → completed/{slug}.md
Mark status as ✅ Complete
Archive artifacts (optional)

Location: .claude/sessions/completed/YYYY-MM-DD-{slug}.md

Error Handling

Session File Not Found

On /resume:

No paused session found.

Checked:
- .claude/sessions/paused/ (0 files)
- NEXT_SESSION_TODO.md (not found)
- .claude/sessions/active/ (0 files)

Would you like to start new work?

Session File Corrupted

On /resume:

Session file partially corrupted.

Readable context:
- Goal: OAuth2 migration
- Progress: 5/12 tasks
- Last action: Token refresh implemented

Unable to parse: Agents Used section

Continue with partial context? [y/n]

TodoWrite State Mismatch

On /resume:

⚠️ Warning: Todo state diverged from session file.

Session says: 5/12 tasks done
TodoWrite shows: 3/12 tasks done

Restored from session file (source of truth).
Verify progress and update if needed.

Multiple Active Sessions

On auto-save OR /pause:

⚠️ Warning: Multiple active sessions found:
1. 2025-10-30-oauth2-migration.md (updated 2h ago)
2. 2025-10-30-basecrud-refactor.md (updated just now)

Which session are you working on?
I'll pause the other automatically.

Artifacts Missing

On /resume:

⚠️ Warning: Some artifacts were modified/deleted since pause:

Missing:
- backend/app/auth/jwt_handler.py (deleted?)

Modified:
- backend/app/auth/token.py (23 changes since pause)

Continue anyway? Context may be stale.

Integration with Hooks

Session Manager integrates with Claude Code's SessionStart/SessionEnd hooks for automation.

SessionStart Hook

Purpose: Auto-resume on conversation start

Script: .claude/hooks/resume-session.sh

Behavior:

Check for paused sessions in .claude/sessions/paused/
If found: Prompt "Resume session {name}? [y/n]"
If yes: Trigger session-manager /resume workflow
If no: Continue without session

Configuration (.claude/settings.json):

{
  "hooks": {
    "SessionStart": [{
      "matcher": "resume",
      "hooks": [{
        "type": "command",
        "command": "$CLAUDE_PROJECT_DIR/.claude/hooks/resume-session.sh"
      }]
    }]
  }
}

SessionEnd Hook

Purpose: Auto-save on conversation end

Script: .claude/hooks/save-session-state.sh

Behavior:

Check for active sessions in .claude/sessions/active/
If found: Trigger lightweight auto-save
Move to paused/ if explicit goodbye detected

Configuration (.claude/settings.json):

{
  "hooks": {
    "SessionEnd": [{
      "hooks": [{
        "type": "command",
        "command": "$CLAUDE_PROJECT_DIR/.claude/hooks/save-session-state.sh"
      }]
    }]
  }
}

Bilingual Support

Command Aliases

All commands support EN and UA:

EN	UA
`/pause`	`/пауза`
`/resume`	`/продовжити`
`/session`	`/сесія`

Trigger phrases (see "When to Use This Skill" section for full list)

Content Format

Headers: Always EN (## Context, ## Todo, ## Next Actions)

Content: Mixed EN/UA as natural

User-facing messages: Always bilingual

Examples:

Goal: Рефакторинг CRUD services
Approach: BaseCRUD inheritance pattern

- [x] Migrate atom_crud.py (-50 LOC)
- [ ] Мігрувати message_crud.py (est. 20 min)

Resume with: `/resume` або `/продовжити`
Say "ready" / "готовий" to proceed.

Progressive Disclosure

This SKILL.md provides:

High-level workflow overview
Command descriptions
Core behavior patterns

For detailed procedural steps, reference files are available:

`references/PAUSE_WORKFLOW.md`

Detailed checklist for pause operation:

Metadata generation steps
Todo state capture logic
Agent history documentation
Next actions identification
Session file generation
Error handling procedures

Load when: Implementing pause logic or debugging pause issues.

`references/RESUME_WORKFLOW.md`

Detailed checklist for resume operation:

Session file location priority
Context parsing steps
TodoWrite restoration logic
Artifact verification
Session file movement
Context reconstruction best practices

Load when: Implementing resume logic or debugging context restoration.

`references/SUMMARY_TEMPLATES.md`

Complete templates with examples:

Standard template (full session)
Minimal template (quick tasks)
Epic template (multi-day sessions)
Admonition usage guide
Bilingual content guidelines
Compact context rules
Session slug generation

Load when: Generating session summaries or validating format.

Best Practices

Compact Context

Use 2-word snippets:

✅ "Inheritance pattern", "API refactor"
❌ "We are using an inheritance-based pattern to..."

Use tables over prose:

✅ | Blocker | TypeScript errors |
❌ "Currently blocked by TypeScript errors..."

Use metrics over descriptions:

✅ "(-50 LOC)", "(+18 tests)", "(3.2s → 0.8s)"
❌ "Reduced code size", "Added tests", "Improved performance"

Session File Management

Active sessions:

Keep only 1-2 active at a time
Auto-save every 3-5 completions
Minimal detail (high-priority only)

Paused sessions:

Full detail (all sections)
Clear resume instructions
Verify before moving

Completed sessions:

Archive after 30 days
Keep for historical reference
Link from project docs if significant

TodoWrite Sync

Before pause:

Complete or block in_progress tasks
Verify counts match
Clean up stale todos

After resume:

Recreate exact state
Mark first pending as in_progress
Validate against session file

Artifact Links

Use relative paths:

✅ backend/app/crud/atom_crud.py
❌ /Users/maks/project/backend/app/crud/atom_crud.py

Include metrics:

✅ atom_crud.py (230→180 LOC)
❌ atom_crud.py

Link artifacts:

✅ [migration report](.artifacts/basecrud/phase-1.md)
❌ "See the migration report in artifacts"

Agent Output Integration

Purpose

When agents complete delegated work, they should integrate their findings directly into the active session instead of creating orphaned artifact files in the project root.

Problem this solves:

❌ Before: Agents create SPRINT1_TEST_FIX_VERIFICATION.md, ANALYSIS_REPORT.md etc.
✅ After: Agent findings automatically appended to .claude/sessions/active/{slug}.md

For Agents (Subagents)

When you complete work as a delegated agent, follow this workflow:

Step 1: Check for Active Session

# Find active session file
active_session=$(ls .claude/sessions/active/*.md 2>/dev/null | head -1)

Step 2: Append Your Report

If session exists:

if [ -n "$active_session" ]; then
  # Add separator and timestamp header
  echo -e "\n---\n" >> "$active_session"
  echo "## Agent Report: $(date +'%Y-%m-%d %H:%M') - [Your Agent Name]" >> "$active_session"
  echo "" >> "$active_session"

  # Append your findings
  cat your_report.md >> "$active_session"

  echo "✅ Findings appended to: $active_session"
else
  echo "⚠️  No active session - creating standalone artifact"
  # Save to project root or .artifacts/
fi

Alternative (using helper script):

# Use the helper script (if available)
.claude/scripts/update-active-session.sh "your-agent-name" your_report.md

Step 3: Update TodoWrite (if needed)

If you discovered new tasks during your work:

Use TodoWrite tool to add:
- [ ] New task discovered by agent
- [ ] Another follow-up action

This triggers auto-save automatically.

Step 4: Report Back

Include session update status in your final output:

✅ Verification complete. Findings appended to active session.
📄 Session file: .claude/sessions/active/2025-11-01-sprint1-consolidated.md

For Coordinator

When delegating to agents, provide session context:

Good delegation:

Task: Verify test fixes from Sprint 1

**Context:** Active session: sprint-1-ux-fixes
**Session file:** .claude/sessions/active/2025-11-01-sprint1-consolidated.md

Please append your verification report to the active session.

Agent will automatically:

Check for active session
Append findings to session file
Update TodoWrite if new tasks found
Trigger auto-save

Coordinator doesn't need to:

Manually merge agent outputs
Link artifact files
Update session manually

Auto-Save Behavior

Agent session updates trigger auto-save:

Event	Auto-Save Type	Location
Agent appends to session	Lightweight checkpoint	`active/`
TodoWrite update	Lightweight checkpoint	`active/`
Explicit `/pause`	Comprehensive save	`paused/`

Rationale: Auto-save is lightweight (don't interrupt flow), explicit pause is comprehensive (user expects full checkpoint).

Session Update Template

Recommended format for agent reports:

---

## Agent Report: 2025-11-01 14:30 - pytest-test-master

### Summary
Brief 1-2 sentence summary of findings.

### Verification Results
- ✅ Test X passed
- ❌ Test Y failed (reason)
- ⚠️  Test Z requires attention

### Recommendations
1. Action item 1
2. Action item 2

### Files Analyzed
- `backend/tests/test_foo.py`
- `backend/app/service.py`

---

Handling Orphaned Artifacts

If you find orphaned artifact files:

Use the cleanup script during pause workflow:

# In PAUSE_WORKFLOW.md, step 6.5:
# Find orphaned agent artifacts
orphaned=$(find . -maxdepth 1 -name "*REPORT*.md" -o -name "*VERIFICATION*.md" -o -name "*SUMMARY*.md")

# Append to session before pausing
for artifact in $orphaned; do
  echo -e "\n## Orphaned Artifact: $(basename $artifact)\n" >> "$session_file"
  cat "$artifact" >> "$session_file"
  rm "$artifact"  # Clean up
done

Examples

Example 1: pytest-test-master Integration

Before (creates orphaned file):

# Agent creates: SPRINT1_TEST_FIX_VERIFICATION.md
# Coordinator must manually merge into session

After (automatic integration):

# Agent checks: ls .claude/sessions/active/
# Finds: 2025-11-01-sprint1-consolidated.md
# Appends verification report to session
# Updates TodoWrite with new tasks
# Auto-save triggered
# Result: Zero orphaned files

Example 2: react-frontend-architect Integration

Agent completes UX audit:

active_session=$(ls .claude/sessions/active/*.md | head -1)

if [ -n "$active_session" ]; then
  echo -e "\n---\n" >> "$active_session"
  echo "## Agent Report: $(date +'%Y-%m-%d %H:%M') - react-frontend-architect" >> "$active_session"
  echo "" >> "$active_session"
  echo "### UX Audit Results" >> "$active_session"
  echo "" >> "$active_session"
  echo "- ✅ Keyboard navigation: Fully accessible" >> "$active_session"
  echo "- ✅ WCAG 2.1 AA: All checks passed" >> "$active_session"
  echo "- ⚠️  Mobile sidebar: Needs improvement" >> "$active_session"
  echo "" >> "$active_session"
  echo "**Recommendation:** Implement hamburger menu for mobile (4h)" >> "$active_session"
fi

Result: Audit findings in session, TodoWrite updated with mobile task, auto-save triggered.

Benefits

For Agents:

✅ Clear instructions on where to save outputs
✅ Automatic integration (no manual steps)
✅ Context preservation guaranteed

For Coordinator:

✅ Zero manual artifact merging
✅ All findings in one place (session file)
✅ Seamless pause/resume workflow

For Project:

✅ Zero orphaned files in project root
✅ Clean git status
✅ Easy to find historical work (session files)

Comparison with Old Orchestration Skills

What session-manager Replaces

task-orchestrator (930 lines):

Over-engineered routing logic
Manual session state JSON files
Verbose coordination warnings

parallel-coordinator (657 lines):

Medium-sized batching (15-25min)
Manual checkpointing
Session continuity (244 lines of docs)

epic-orchestrator (560 lines):

Multi-week coordination
Manual progress.md updates
Git-centric team workflows

Total: 2,147 lines → 450 lines (79% reduction)

Key Improvements

Simplicity:

Single markdown file (not 8-12 files per session)
Auto-save (not manual state management)
Lightweight checkpoints (not comprehensive batches)

Usability:

"It just works" (auto-save after TodoWrite)
Clear commands (/pause, /resume)
Bilingual support (EN/UA)

Context efficiency:

One file to load (not scattered across artifacts)
2-word snippets (not essays)
Progressive disclosure (load details on demand)

Notes

Session files are markdown (easy to read/edit manually)
Use .gitignore for active/ and paused/ dirs (personal state)
Keep completed/ in git for historical reference (team visibility)
Session slug max 40 chars total
Always output bilingual prompts for user-facing messages
Prefer tables over prose, metrics over descriptions
Load reference files only when implementing or debugging

Install Skill

SKILL.md

Session Manager

Overview

Core Workflow

When to Use This Skill

Trigger Phrases

1. Show Sessions (Показати сесії)

2. What's Next / Plans (Що далі / Плани)

3. Session Status (Статус сесії)

4. Continue/Resume (Продовжити)

5. Continue Specific Session (Продовжити конкретну)

6. Pause (Пауза)

7. Switch Session (Перемкнутись)

8. Archive (Архівувати)

9. Overview (Огляд)

Auto-Trigger Scenarios

Do NOT Use This Skill For

Session Name Matching

Response Templates

List Sessions Response

Session Status Response

Resume Prompt

Pause Confirmation

Auto-Save Logic

When Auto-Save Triggers

What Auto-Save Captures

Auto-Save vs Explicit Pause

TodoWrite Integration

Bi-Directional Sync

State Validation

Session State Format

Session Lifecycle

1. Session Creation

2. Active Session

3. Paused Session

4. Resumed Session

5. Completed Session

Error Handling

Session File Not Found

Session File Corrupted

TodoWrite State Mismatch

Multiple Active Sessions

Artifacts Missing

Integration with Hooks

SessionStart Hook

SessionEnd Hook

Bilingual Support

Command Aliases

Content Format

Progressive Disclosure

references/PAUSE_WORKFLOW.md

references/RESUME_WORKFLOW.md

references/SUMMARY_TEMPLATES.md

Best Practices

Compact Context

Session File Management

TodoWrite Sync

Artifact Links

Agent Output Integration

Purpose

For Agents (Subagents)

Step 1: Check for Active Session

Step 2: Append Your Report

Step 3: Update TodoWrite (if needed)

Step 4: Report Back

For Coordinator

Auto-Save Behavior

Session Update Template

Handling Orphaned Artifacts

Examples

Example 1: pytest-test-master Integration

Example 2: react-frontend-architect Integration

Benefits

Comparison with Old Orchestration Skills

What session-manager Replaces

Key Improvements

Notes

`references/PAUSE_WORKFLOW.md`

`references/RESUME_WORKFLOW.md`

`references/SUMMARY_TEMPLATES.md`