---
description: End-of-session checklist for shipping, memory, and self-improvement
allowed-tools: Read, Write, Edit, Glob, Grep, Bash(git:*), Bash(gh:*), Bash(ls:*), Bash(npm run:*), Bash(yarn:*), Bash(pnpm:*), Bash(make:*), Bash(deploy:*), Bash(mv:*), Bash(mkdir:*)
---

# Session Wrap-Up

Run four phases in order. Each phase is conversational and inline — no
separate documents. All phases auto-apply without asking; present a
consolidated report at the end.

## Phase 1: Ship It

**Commit:**
1. Run `git status` in each repo directory that was touched during the session
2. If uncommitted changes exist, auto-commit to main with a descriptive message
3. Push to remote

**File placement check:**
4. If any files were created or saved during this session:
   - Verify they follow your naming convention
   - Auto-fix naming violations (rename the file)
   - Verify they're in the correct subfolder per your project structure
   - Auto-move misplaced files to their correct location
5. If any document-type files (.md, .docx, .pdf, .xlsx, .pptx) were created
   at the workspace root or in code directories, move them to the docs folder
   if they belong there

**Deploy:**
6. Check if the project has a deploy skill or script
7. If one exists, run it
8. If not, skip deployment entirely — do not ask about manual deployment

**Task cleanup:**
9. Check the task list for in-progress or stale items
10. Mark completed tasks as done, flag orphaned ones

## Phase 2: Remember It

Review what was learned during the session. Decide where each piece of
knowledge belongs in the memory hierarchy:

**Memory placement guide:**
- **Auto memory** (Claude writes for itself) — Debugging insights, patterns
  discovered during the session, project quirks. Save these to the auto memory
  directory at `~/.claude/projects/` for the relevant project.
- **CLAUDE.md** (instructions for Claude) — Permanent project rules,
  conventions, commands, architecture decisions that should guide all future
  sessions
- **`.claude/rules/`** (modular project rules) — Topic-specific instructions
  that apply to certain file types or areas. Use `paths:` frontmatter to scope
  rules to relevant files (e.g., testing rules scoped to `tests/**`)
- **`CLAUDE.local.md`** (private per-project notes) — Personal WIP context,
  local URLs, sandbox credentials, current focus areas that shouldn't be
  committed
- **`@import` references** — When a CLAUDE.md would benefit from referencing
  another file rather than duplicating its content

**Decision framework:**
- Is it a permanent project convention? → CLAUDE.md or `.claude/rules/`
- Is it scoped to specific file types? → `.claude/rules/` with `paths:`
  frontmatter
- Is it a pattern or insight Claude discovered? → Auto memory
- Is it personal/ephemeral context? → `CLAUDE.local.md`
- Is it duplicating content from another file? → Use `@import` instead

Note anything important in the appropriate location.

## Phase 3: Review & Apply

Analyze the conversation for self-improvement findings. If the session was
short or routine with nothing notable, say "Nothing to improve" and proceed
to Phase 4.

**Auto-apply all actionable findings immediately** — do not ask for approval
on each one. Apply the changes, commit them, then present a summary of what
was done.

**Finding categories:**
- **Skill gap** — Things Claude struggled with, got wrong, or needed multiple
  attempts
- **Friction** — Repeated manual steps, things user had to ask for explicitly
  that should have been automatic
- **Knowledge** — Facts about projects, preferences, or setup that Claude
  didn't know but should have
- **Automation** — Repetitive patterns that could become skills, hooks, or
  scripts

**Action types:**
- **CLAUDE.md** — Edit the relevant project or global CLAUDE.md
- **Rules** — Create or update a `.claude/rules/` file
- **Auto memory** — Save an insight for future sessions
- **Skill / Hook** — Document a new skill or hook spec for implementation
- **CLAUDE.local.md** — Create or update per-project local memory

## Phase 4: Report

Present a consolidated summary with two sections — applied items first,
then no-action items:

### Format:

```
## Phase 1: Ship It
- [commit status]
- [file placement status]
- [deploy status]
- [task cleanup status]

## Phase 2: Remember It
- [what was saved and where]

## Phase 3: Review & Apply

Findings (applied):

1. ✅ [category]: [description]
   → [action type] [what was done]

2. ✅ [category]: [description]
   → [action type] [what was done]

---
No action needed:

3. [category]: [description]
   Already documented / not actionable
```