session-wrap - SKILL.md Agent Skill

name: session-wrap

description: This skill should be used when the user asks to "wrap up session", "end session", "session wrap", "/wrap", "document learnings", "what should I commit", or wants to analyze completed work before ending a coding session.

version: 2.2.0

Session Wrap Skill

Comprehensive session wrap-up workflow with multi-agent analysis.

Execution Flow


┌─────────────────────────────────────────────────────┐

│  1. Check Git Status                                │

├─────────────────────────────────────────────────────┤

│  2. Phase 1: 4 Analysis Agents (Parallel)           │

│     ┌─────────────────┬─────────────────┐           │

│     │  doc-updater    │  automation-    │           │

│     │  (docs update)  │  scout          │           │

│     ├─────────────────┼─────────────────┤           │

│     │  learning-      │  followup-      │           │

│     │  extractor      │  suggester      │           │

│     └─────────────────┴─────────────────┘           │

├─────────────────────────────────────────────────────┤

│  3. Phase 2: Validation Agent (Sequential)          │

│     ┌───────────────────────────────────┐           │

│     │       duplicate-checker           │           │

│     │  (Validate Phase 1 proposals)     │           │

│     └───────────────────────────────────┘           │

├─────────────────────────────────────────────────────┤

│  4. Integrate Results & AskUserQuestion             │

├─────────────────────────────────────────────────────┤

│  5. Execute Selected Actions                        │

├─────────────────────────────────────────────────────┤

│  7. Update Documentation Files (Auto, MANDATORY)    │

│     ├─ claude.md  (project guidelines)              │

│     └─ wrapup.md  (session progress)                │

├─────────────────────────────────────────────────────┤

│  8. Update Obsidian Daily Note (Auto, MANDATORY)    │

│     └─ Claude Session Log section                   │

├─────────────────────────────────────────────────────┤

│  9. Recommend Azure DevOps Board Sync (Optional)    │

│     └─ Conditional: azure-boards skill installed +  │

│        repo in ~/.claude/devops-defaults.json       │

│        → AskUserQuestion                            │

└─────────────────────────────────────────────────────┘

Step 1: Check Git Status


git status --short

git diff --stat HEAD~3 2>/dev/null || git diff --stat

Step 2: Phase 1 - Analysis Agents (Parallel)

Execute 4 agents in parallel (single message with 4 Task calls).

Session Summary (Provide to all agents)


Session Summary:

- Work: [Main tasks performed in session]

- Files: [Created/modified files]

- Decisions: [Key decisions made]

Parallel Execution


Task(

    subagent_type="doc-updater",

    description="Document update analysis",

    prompt="[Session Summary]\n\nAnalyze if CLAUDE.md, context.md need updates."

)



Task(

    subagent_type="automation-scout",

    description="Automation pattern analysis",

    prompt="[Session Summary]\n\nAnalyze repetitive patterns or automation opportunities."

)



Task(

    subagent_type="learning-extractor",

    description="Learning points extraction",

    prompt="[Session Summary]\n\nExtract learnings, mistakes, and new discoveries."

)



Task(

    subagent_type="followup-suggester",

    description="Follow-up task suggestions",

    prompt="[Session Summary]\n\nSuggest incomplete tasks and next session priorities."

)

Agent Roles

| Agent | Role | Output |

|-------|------|--------|

| doc-updater | Analyze CLAUDE.md/context.md updates | Specific content to add |

| automation-scout | Detect automation patterns | skill/command/agent suggestions |

| learning-extractor | Extract learning points | TIL format summary |

| followup-suggester | Suggest follow-up tasks | Prioritized task list |

Step 3: Phase 2 - Validation Agent (Sequential)

Run after Phase 1 completes (dependency on Phase 1 results).


Task(

    subagent_type="duplicate-checker",

    description="Phase 1 proposal validation",

    prompt="""

Validate Phase 1 analysis results.



## doc-updater proposals:

[doc-updater results]



## automation-scout proposals:

[automation-scout results]



Check if proposals duplicate existing docs/automation:

1. Complete duplicate: Recommend skip

2. Partial duplicate: Suggest merge approach

3. No duplicate: Approve for addition

"""

)

Step 4: Integrate Results


## Wrap Analysis Results



### Documentation Updates

[doc-updater summary]

- Duplicate check: [duplicate-checker feedback]



### Automation Suggestions

[automation-scout summary]

- Duplicate check: [duplicate-checker feedback]



### Learning Points

[learning-extractor summary]



### Follow-up Tasks

[followup-suggester summary]

Step 5: Action Selection


AskUserQuestion(

    questions=[{

        "question": "Which actions would you like to perform?",

        "header": "Wrap Options",

        "multiSelect": true,

        "options": [

            {"label": "Create commit (Recommended)", "description": "Commit staged changes"},

            {"label": "Create automation", "description": "Generate skill/command/agent"},

            {"label": "Skip", "description": "End without action"}

        ]

    }]

)

Note: wrapup.md and claude.md are ALWAYS updated automatically in Step 7 (not optional).

Step 6: Execute Selected Actions

Execute only the actions selected by user.

Step 7: Update Documentation Files (Automatic)

ALWAYS execute this step after completing the session wrap. This is NOT optional.

Both files are MANDATORY updates — do NOT skip either one.

This step uses outputs from:

doc-updater: claude.md and wrapup.md specific proposals
learning-extractor: TIL items
followup-suggester: Next tasks

7-A: Update/Create claude.md

File Location

Check for claude.md in order:

.claude/claude.md (preferred)
CLAUDE.md (project root)
Create .claude/claude.md if neither exists

claude.md Purpose

Project-wide guidelines and instructions that persist across all sessions:

Project overview and architecture
Development environment setup (branches, package manager, Docker, etc.)
API endpoints and key interfaces
Coding conventions and standards
Important caveats and gotchas discovered during development
Reference to wrapup.md for session continuity

Update Rules

Read existing file (if exists) and preserve its structure
Merge doc-updater proposals for claude.md — add new sections or update existing ones
Add newly discovered project guidelines from this session (e.g., new endpoints, env vars, workflow caveats)
Do NOT add session-specific details (those go to wrapup.md)
Keep concise — focus on information Claude needs in ALL future sessions

Implementation


1. Read existing claude.md (or initialize template if not exists)

2. Apply doc-updater claude.md proposals (add/update sections)

3. Add any new project-wide knowledge discovered this session

4. Write updated content using Write or Edit tool

5. Log: "Updated .claude/claude.md"

7-B: Update/Create wrapup.md

File Location

Check for wrapup.md in order:

.claude/wrapup.md (preferred)
wrapup.md (project root)
Create .claude/wrapup.md if not exists

wrapup.md Purpose

Session progress tracking file containing:

Completed work and commits
Troubleshooting records (errors + solutions)
Learnings (TIL)
Pending/uncommitted work
Next tasks/TODO
Remote server status

NOT for: Project-wide guidelines (those go to CLAUDE.md)

Update Content Structure


# Session Progress Summary



> Last Updated: [Current Date/Time]



## Branch Info



- **Current Branch**: [branch name]

- **Remote Tracking**: [origin/branch]



---



## Completed Work



### [Commit Hash] [Commit Message]

- [Brief description]

- Files: [changed files]



---



## Troubleshooting Log



### [Issue Title]

**Error:**

[Error message]




**Cause:**

[Root cause]



**Solution:**

[How resolved]



---



## Learnings (TIL)



[learning-extractor output]



---



## Pending Work



| File | Status | Notes |

|------|--------|-------|

| [file] | [modified/untracked] | [description] |



---



## Next Tasks



[followup-suggester output - prioritized list]



---



## Remote Server Status (if applicable)



- Server: [name]

- Last Run: [timestamp]

- Status: [success/failure]

- Data Location: [path]

Update Rules

Preserve existing content - Merge, don't overwrite
Add new commits to "Completed Work" (prepend, newest first)
Append troubleshooting entries (keep history)
Update "Pending Work" from current git status
Refresh "Next Tasks" with followup-suggester output
Add learnings from learning-extractor
Update timestamp in header

Implementation


1. Read existing .claude/wrapup.md (or create if not exists)

2. Collect data from:

   - git log (recent commits)

   - git status (pending work)

   - Session conversation (troubleshooting)

   - doc-updater output (wrapup.md proposals)

   - learning-extractor output (TIL)

   - followup-suggester output (next tasks)

3. Merge into existing structure

4. Write updated content

5. Notify user: "Updated .claude/wrapup.md"

Step 8: Update Obsidian Daily Note (Automatic)

ALWAYS execute this step after Step 7. This is NOT optional.

Update the ## 🤖 Claude Session Log section in today's Obsidian daily note with a concise summary of what was accomplished in this session.

File Location

The daily note path follows this pattern:


/mnt/c/Users/yyoo029/Documents/Obsidian/유용상/Daily/YYYY-MM-DD.md

Use today's date (from the system) to construct the file path.

Target Section

Find the ## 🤖 Claude Session Log section and append the session log between the heading and the --- separator that follows it.

Content Format


### [Project Name] — YYYY-MM-DD HH:MM



**Branch:** `branch-name`



**작업 내역:**

- [Completed task 1 — brief description]

- [Completed task 2 — brief description]

- ...



**주요 결정/발견:**

- [Key decision or discovery, if any]



**다음 작업:**

- [Top 2-3 next tasks from followup-suggester]

Content Rules

Read the existing daily note — preserve all existing content
Append to the Claude Session Log section (do NOT overwrite other session logs from the same day)
Keep it concise — max 10-15 lines per session entry
Use bullet points — no paragraphs, no code blocks unless essential
Include branch name — critical for context when reviewing later
Korean preferred for descriptions, English OK for technical terms
If the daily note file does not exist, skip this step silently (do not create the file)

Implementation


1. Construct today's daily note path: /mnt/c/Users/yyoo029/Documents/Obsidian/유용상/Daily/YYYY-MM-DD.md

2. Check if file exists — if not, skip with log message

3. Read the file and locate "## 🤖 Claude Session Log" section

4. Build session summary from:

   - Completed work (from wrapup.md or session context)

   - Key decisions/discoveries (from learning-extractor)

   - Next tasks (from followup-suggester, top 2-3 only)

5. Append summary after the "## 🤖 Claude Session Log" heading

6. Write updated content

7. Notify user: "Updated Obsidian daily note"

Step 9: Recommend Azure DevOps Board Sync (Optional, Conditional)

Run only if ALL conditions are true:

azure-boards skill is installed (genaikit plugin — ${CLAUDE_PLUGIN_ROOT}/skills/azure-boards/SKILL.md resolvable via Claude Code's plugin loader).
Current repo is mapped in ~/.claude/devops-defaults.json.
PAT is available (AZURE_DEVOPS_EXT_PAT env var set, or file-stored PAT at ~/.azure/azuredevops/personalAccessTokens).

If any are false → skip silently. Do NOT call check_auth.sh here — that script requires AAD login (az account show) which is unnecessary for PAT-only users and would wrongly skip the entire step.

Detection (cheap, no API calls)


# 1. azure-boards skill present (genaikit)?

ls ~/.claude/plugins/cache/*/genaikit/*/skills/azure-boards/SKILL.md >/dev/null 2>&1 || skip



# 2. Repo mapped? Check repo from git remote against devops-defaults.json.

python3 -c "

import json, sys, subprocess

from pathlib import Path

url = subprocess.run(['git','config','--get','remote.origin.url'],

    capture_output=True, text=True).stdout.strip()

if 'dev.azure.com' not in url: sys.exit(2)

repo = url.rstrip('/').rsplit('/',1)[-1]

mapping_file = Path.home() / '.claude/devops-defaults.json'

mapping = json.loads(mapping_file.read_text()) if mapping_file.exists() else {}

sys.exit(0 if repo in mapping else 1)

"



# 3. PAT reachable?

[ -n "$AZURE_DEVOPS_EXT_PAT" ] || [ -s ~/.azure/azuredevops/personalAccessTokens ] || skip

If exit code is 0 → proceed. Non-zero → skip.

AskUserQuestion (after Step 8, NEVER auto-execute)

This step is purely a recommendation. Do not push to the board without explicit user opt-in.


AskUserQuestion(

    questions=[{

        "question": "Sync this session's work to the Azure DevOps board for {team_name}?",

        "header": "Board Sync (optional)",

        "multiSelect": true,

        "options": [

            {"label": "Create work items for next tasks",

             "description": "Turn followup-suggester output into new Tasks on the board"},

            {"label": "Update existing work items",

             "description": "Apply state/comment changes to items mentioned this session"},

            {"label": "Skip",

             "description": "Do not touch the board"}

        ]

    }]

)

Execution rules

Default to Skip when in doubt — never assume the user wants writes.
Show preview before each write: list the exact create/update operations and ask one final confirmation if more than 2 items will change.

Resolve team defaults from the mapping file (~/.claude/devops-defaults.json). Entry shape: {"team": "...", "area_path": "...", "iteration_path": "..."}. area_path is required for create; iteration_path is optional and, if omitted, resolve the current sprint at runtime:


GENAIKIT=$(ls -d ~/.claude/plugins/cache/*/genaikit/*/skills/azure-boards | head -1)

team=$(python3 -c "import json,pathlib; print(json.loads(pathlib.Path.home().joinpath('.claude/devops-defaults.json').read_text())['$repo']['team'])")

current_iter=$(az boards iteration team list --team "$team" --timeframe current --output json 2>/dev/null | jq -r '.[0].path // empty' | sed -E 's#\\Iteration\\#\\#')

Use genaikit azure-boards scripts — never call the Azure DevOps REST API directly:


# Create

GENAIKIT=$(ls -d ~/.claude/plugins/cache/*/genaikit/*/skills/azure-boards | head -1)

tmp=$(mktemp); printf '%s' "$rationale" > "$tmp"

bash "$GENAIKIT/scripts/create_work_item.sh" \

  --type Task --title "$title" \

  --area "$area_path" --iteration "$current_iter" \

  --description-file "$tmp"

rm -f "$tmp"



# Update (state transition)

bash "$GENAIKIT/scripts/update_work_item.sh" --id "$id" --state "$state"

Source data to map:
- "Create work items for next tasks" → use the followup-suggester output verbatim (one work item per suggested task; title = task line, description = rationale if present)
- "Update existing work items" → only if the session conversation explicitly referenced work item IDs (e.g., #AB1234 patterns in commit messages or user requests). Do not invent IDs.
Failure handling: on non-zero exit from create_work_item.sh / update_work_item.sh, surface stderr verbatim and stop. Common causes: expired PAT, missing Work Items scope on PAT, wrong iteration path, locked state transition.

After the recommendation

Whether the user accepted or skipped, conclude the wrap normally. Step 9 does not affect Steps 7/8 (those have already run).

Quick Reference

When to Use

End of significant work session
Before switching to different project
After completing a feature or fixing a bug

When to Skip

Very short session with trivial changes
Only reading/exploring code
Quick one-off question answered

Arguments

Empty: Proceed interactively (full workflow)
Message provided: Use as commit message and commit directly

Additional Resources

See references/multi-agent-patterns.md for detailed orchestration patterns.