- Home
- Skills
- Productivity
- wrap-up — End-of-session routine that saves tokens
Works with the AI Agents you already use
wrap-up — End-of-session routine that saves tokens
Say "wrap up" and your coding agent compresses its memory (fewer tokens every future session), prunes merged branches, verifies the deploy is really live, and writes tomorrow's first step.
Free
wrap-up — End-of-session routine that saves tokens
Example session with this skill installed
wrap up
- Read your context and instructions
- Compiled the wrap-up
Wrap-up report
Memory — 3 entries this session touched. Merged two notes on the checkout bug into one current-state line; cut the 14-step investigation. Kept one fact that was hiding in it ("payment webhook only accepts HTTPS from the allowlisted IP") and moved it into deployment.md. Index updated, 1 dead link fixed.
Branches — 4 merged branches deleted (0 ahead). 2 kept, waiting for your call:
| Branch | Ahead | Last commit | What it is |
|---|---|---|---|
feat/csv-export | 3 | 12 days ago | CSV export for the orders page |
spike/redis-cache | 1 | 26 days ago | caching experiment |
Continue, merge or abandon? I'll list them again next wrap-up if you skip.
Publish — committed fix: retry webhook on 502 and docs: deployment notes separately. Secret scan: clean. Pushed to main.
Release status: live (verified) — health check OK, new container started 2 min after push, bundle contains the new retry string.
Kept on purpose
data/exports/not committed (its own.gitignoreexcludes everything; contains real order data)devbranch kept (long-lived integration branch)
Tomorrow's first step (written to the top of MEMORY.md):
⏭️ Next first step (wrapped up 2026-11-08): decide on
feat/csv-export, then run the refund flow on staging
Illustrative example.
Connects securely to your tools. The creator never sees your data.
About this skill
Everyone has a routine for starting a session with an AI coding agent. Almost nobody has one for ending it. The result, a few months in:
- Memory bloat. Every session adds notes to the file your agent auto-loads. You pay for all of it at the start of every session, and half of it is old debugging stories.
- Branch rot. A dozen branches, nobody sure which are already merged.
- Fake "done". The agent reports "shipped" after a push, but the deploy never went live.
- Morning amnesia. The first part of each session goes to working out where you left off.
wrap-up fixes all four with one phrase.
The four steps
- Compress memory — only the entries this session touched (no full rescan), checked against a keep/cut table: keep current state, environment facts, open todos; cut investigation history and one-off war stories. The subtle rule: environment facts often hide inside war stories, so it extracts the fact before deleting the story.
- Prune branches — labels lie, so it checks reality:
git rev-list --count origin/main..<branch>. Zero ahead means merged and safe to delete. Anything ahead goes into a standing table with its age and gets a continue / merge / abandon call. Nothing unmerged is deleted without your say-so. - Publish and verify — never
git add ., a one-line secret scan on staged content, optional migration-drift and version-bump checks, and a real deployment check. Version strings and timestamps don't count as proof; "live" is only reported after verifying. - Tomorrow's first step — one line, written as an action, at the top of the memory index. Your next session opens with it and gets straight to work instead of spending tokens reconstructing context.
Why it saves tokens
The memory index is loaded at the start of every session, so every stale line costs you again and again. Scoped compression keeps it small, and the first-step line cuts the "where was I?" exploration at the start of each session.
What it won't do
Delete unmerged branches, push, or delete remote branches without asking. It reports what it kept on purpose and why.
Single SKILL.md, no dependencies. Open source (MIT), original in Traditional Chinese: https://github.com/youllook/wrap-up
Unmerged branches require explicit manual approval. Windows worktree deletion may be blocked by file locks. Requires git and shell environment.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 9 days ago
- Free with an Agensi account
Requires
Permissions
Terminal: runs git commands (fetch, branch, rev-list, commit, push) and a grep secret scan. Read/Write files: reads and edits the agent's memory files. Network: only git push/fetch to your own remote and, if your project deploys on push, a health check against your own deployment. Anything irreversible (push, deleting remote branches) asks first.