skills-for-ai
Get started
Productivity Development Skill

Happy Ending

Every session deserves a Happy Ending — the clean close for your AI session, without losing the thread.

Session closeContext handoffHandoffWrap-upMemory PatchProject memoryDecision logNext-session startMemory safetyReversal detectionCommit suggestion
Free
0 ↓ 0 v1.0.0 Updated 07.07.2026 by skills-for-ai

SKILL.md in the open agentskills.io standard — works directly in Claude, ChatGPT/Codex, Cursor, Copilot & more.

Happy Ending is an AI skill for closing AI work sessions cleanly, turning the chat into the valid project state: final decisions, open tasks, a memory patch and a copy-paste next-session prompt. Its core is a Memory Safety layer, so discarded, canceled or superseded work is not mistakenly saved as the valid state.

What this skill does

  • Extracts final decisions, open tasks, learnings and discarded points from the session
  • Protects your memory: canceled, reverted or superseded work is not saved as the valid state (Memory Safety)
  • Emits a memory patch (Add/Update/Ignore) instead of blind-rewriting whole memory files
  • Separates "don't adopt" (Do Not Carry Forward) from "don't try again" (Do Not Repeat)
  • Delivers a directly pasteable next-session prompt plus a compact Delta block
  • Suggests commit messages and commit groups for code — but executes nothing unasked
  • Two depths: Quick (compact default) and Deep (on request)

Description

Happy Ending is a skill for closing AI sessions — it turns a messy chat into a clean, reusable project state instead of a novel. Guiding line: Happy Ending doesn't save everything that happened, it secures what holds from now on. It runs in three phases (Look Back → Clean Up → Kiss Goodnight) and picks up final decisions, open tasks, learnings, user corrections, references and discarded points. Out of that comes a compact report: Result, Decisions, Open, Do Not Carry Forward, Do Not Repeat, a Memory Patch (Add/Update/Ignore), file/git notes when relevant, the single next best step, a directly pasteable next-session prompt, a Delta block (+ ~ − !) and a short closing line with counters. The core feature is the **Memory Safety layer**: every point gets a status (final, open, tentative, reverted, canceled, superseded, rejected, do-not-repeat, do-not-carry-forward, unclear) and a confidence — only final + high (or your explicit confirmation) may enter long-term memory. A reversal detector watches for undo signals ("actually no", "canceled", "take that out"), when the session contradicts itself the last confirmed state wins (final-state-wins), and instead of blind-rewriting whole memory files the skill emits a patch. Git and files are only ever suggested — Happy Ending never commits, pushes, deploys or overwrites without your explicit instruction. Two depths: Quick as the compact default, Deep on request.

Examples

Close the session quickly
Wrap up the session.
You July 17, 2026
Happy Ending

# Happy Ending

## Result

Landing-page hero copy and the primary CTA are settled. One asset is still open.

## Decisions

  • [final · high] Hero headline: "Ship faster, guess less."
  • [final · high] Primary CTA: "Start free" (secondary "See how it works").

## Open

  • [open] Export the hero background as WebP (1600px).

## Do Not Carry Forward

  • The purple-gradient hero — dropped mid-session, don't reintroduce.

## Memory Patch

+ Add

  • Preference: CTA voice is plain-imperative ("Start free"), not "Get started now".

– Ignore

  • The purple-gradient hero direction (discarded).

## Next best step

Export the WebP hero asset into /public/hero.

## Next-session prompt

> We're continuing the landing page. State: hero copy + CTAs final ("Ship faster, guess less." / "Start free"). Open: export the hero background as WebP (1600px). Don't reintroduce the purple-gradient hero — it was dropped.

## Delta

+ Saved: hero headline, CTA voice preference

– Done/removed: purple-gradient direction

! Open: WebP hero export

## Closing line

Happy Ending complete: 2 decisions saved · 1 task sorted · 1 unsafe carry-forward blocked · 1 clean start ready

Calibration: changed our mind twice
Close it out — but careful, we changed the DB decision twice.
You July 17, 2026
Happy Ending

Heads-up on the flip-flops — I'm saving only the last confirmed call, not the detours:

# Happy Ending

## Result

Auth is done; the storage decision changed twice and landed on Postgres.

## Decisions

  • [final · high] Database: Postgres (confirmed last, after SQLite then Mongo).

## Reverted / Canceled / Superseded

  • SQLite — superseded (early pick, replaced).
  • MongoDB — reverted ("actually let's not") → not the active state.

## Do Not Repeat

  • Don't re-propose Mongo for this project — explicitly rejected.

## Memory Patch

+ Add

  • [final · high] Project DB = Postgres.

~ Update

  • Replace any earlier "DB SQLite/Mongo" note with Postgres.

– Ignore

  • SQLite and Mongo as active choices.

## Next best step

Write the Postgres schema for the users and sessions tables.

## Next-session prompt

> Continuing the backend. State: DB is Postgres (final — SQLite and Mongo were tried and dropped, don't treat them as active). Auth is done. Next: Postgres schema for users + sessions.

## Delta

+ Saved: DB = Postgres (final)

– Done/removed: SQLite (superseded), Mongo (reverted)

! Open: Postgres schema

## Closing line

Happy Ending complete: 1 decision saved · 1 task sorted · 2 unsafe carry-forwards blocked · 1 clean start ready

I didn't save Mongo/SQLite as the state — the final call wins. Want a commit message for the auth work too?

1 / 2

What it does not do

  • Not a chat summary and not a chronological retelling — it secures the valid state
  • Not an auto memory-dump — only final + high (or your confirmation) is saved for good
  • Never commits, pushes, deploys or overwrites without an explicit instruction — it suggests
  • Not a full project-management system and no substitute for tests, review or release
  • Doesn't impose a memory layout — it patches your real memory (CLAUDE.md, a memory folder, Cursor rules …)

Compatibility & tech

ClaudeChatGPT/CodexCursorCopilotGemini CLIWindsurfCline
Tested (internal)
4 scenarios
Recommended runtime
Claude Opus or Sonnet; with git/file access for commit/file suggestions, otherwise pure copy-paste suggestions
Modes
Quick (default) · Deep · Handoff · Git suggestion (optional) · Skill notes (optional)
Inputs
The session's chat history · Git status (optional) · Existing project memory (optional)
Output format
Compact Markdown report: Result · Decisions · Open · Do-not lists · Memory Patch · Next-session prompt · Delta · Closing line
Subcategory
Session close & handoff
License
Proprietär

Security profile

Local

Runs entirely on your machine with your own AI — no external runtime, no running costs.

Instruction-only

Only instructions, templates and references — no executable scripts.

No network access

Works offline with what you provide — does not call external services on its own.

What do these badges mean? →

What you get

  • happy-ending-1.0.0/5 files
    • .claude-plugin/marketplace.json
    • plugins/happy-ending/skills/happy-ending/5 files
      • SKILL.md
      • manifest.json
      • references/3 files
        • happy-ending-examples.md
        • happy-ending-memory-safety.md
        • happy-ending-templates.md
    • .agents/skills/happy-ending/→ universal — same content (Codex, Cursor, Copilot, Gemini, Windsurf, Cline)
    • LICENSE.txt

Installation

After unlocking, you install with a single command — it auto-detects your AI tool.

Runs inClaude CodeGitHub CopilotGemini CLICursorCodex CLIWindsurfCline

Also works as a chat prompt

Full in chat

No AI tool? Paste it into Claude, ChatGPT or Gemini and use the method right away.

This skill ships no scripts, so the prompt carries its full method.

Installing is the full version — it triggers automatically, runs its scripts and loads references as needed. As a chat prompt you drive the method by hand.

Unlock to copy the ready-to-paste prompt — then in “My Skills”.

Reviews

No reviews yet – be the first.

Note

Start it manually by default ("make a Happy Ending" or "/happy-ending"). The skill only suggests memory, git and file changes — it saves, commits or overwrites nothing without your explicit approval. When the final state is unclear, it asks one short question instead of guessing.

Changelog

  • v1.0.007.07.2026Initial release: three-phase session close (Look Back → Clean Up → Kiss Goodnight) with a Memory Safety layer (status + confidence, reversal detector, final-state-wins, patch-not-rewrite), separate Do Not Carry Forward / Do Not Repeat, a memory patch, a copy-paste next-session prompt, a compact Delta block and a closing line. Two depths (Quick/Deep) + optional Git/skill blocks; git and files are only ever suggested. References for templates, memory-safety and examples.

Frequently asked questions

What does Happy Ending do?

Happy Ending is an AI skill for closing AI work sessions cleanly, turning the chat into the valid project state: final decisions, open tasks, a memory patch and a copy-paste next-session prompt. Its core is a Memory Safety layer, so discarded, canceled or superseded work is not mistakenly saved as the valid state.

How do I close an AI session cleanly?

With Happy Ending — it scans the session for final decisions, open tasks and discarded points, builds a memory patch and a copy-paste next-session prompt, compact rather than a novel. Trigger it with "/happy-ending" or "wrap up the session".

Is Happy Ending just a summary?

No. A summary tells you what happened — Happy Ending secures what holds from now on, meaning valid decisions, the current state, open tasks and the next entry point. Discarded, canceled or superseded points are deliberately not saved as an active direction.

How does Happy Ending keep wrong entries out of memory?

Through its Memory Safety layer — every point gets a status (final, reverted, canceled, superseded …) and a confidence, and only final + high (or your explicit confirmation) may enter long-term memory. A reversal detector catches undo signals, and when the session contradicts itself the last confirmed state wins (final-state-wins).

Can Happy Ending commit or push?

Happy Ending suggests commit messages and commit groups but never executes anything unasked — no commit, push, deploy or overwrite without your explicit instruction.

What's the difference between Do Not Carry Forward and Do Not Repeat?

Do Not Carry Forward means not adopting something into memory, project state or the next session. Do Not Repeat means not proposing or attempting it again. Happy Ending keeps the two separate.

Which AI tools does Happy Ending work with?

Claude · ChatGPT/Codex · Cursor · Copilot · Gemini CLI · Windsurf · Cline

How do I use Happy Ending?

Happy Ending is a SKILL.md in the open agentskills.io standard: install it with one command (npx) or download it and add it to your AI tool — Claude (Projects), ChatGPT (Custom GPT), Cursor, Copilot, Gemini CLI and more. No code needed.