command
/end
Before you stop, Claude writes a short handoff to disk: what got done, what’s unfinished, the exact next step, and the key files.
Stop re-explaining your project. Continuation stores decisions, learnings, conventions, and handoffs as Markdown in your repo. It works with any agent that can read a skill file; the Claude Code plugin adds /end, /continue, and /next.
A tiny, dependency-free plugin that adds three slash commands. That’s the whole thing — no linting, no commits, no automation. Just handoff, resume, and launch, so a future session picks up exactly where the last one stopped.
/endBefore you stop, Claude writes a short handoff to disk: what got done, what’s unfinished, the exact next step, and the key files.
/continueNext time, Claude reads the latest handoff, sanity-checks your git state, and orients you in about five seconds. Read-only — it never writes.
/nextOpens the next session in a fresh terminal (worktree-aware), auto-running /continue, with a distinct colour + title so parallel worktrees are easy to tell apart. macOS.
What it is, why it exists, when to use it, who it’s for, where it stores things, and how a full cycle goes.
Three slash commands — /end, /continue, /next — that persist a session’s state to disk and resume it cleanly later.
Open a project and Claude has no memory of yesterday — you waste ten minutes re-explaining. This removes that, and lets you reset a full context window without losing the thread.
Run /end before you stop, /continue when you return, and /next when you’re juggling several git worktrees and want one labelled session each.
Any language, any stack, solo or team. No dependencies, no runtime. Works in git repos and plain directories alike.
Handoffs live in .claude/continuations/ and optional gotchas in .claude/learnings/ — one file per session, dated. Nothing leaves your machine.
/end captures & confirms then writes; /continue finds, reality-checks git, and orients; /next opens the next labelled window. Read on for the full cycle.
The natural loop is /end → /next → /continue. Each command does one job, well.
/end — capture & confirm, then writeClaude reviews your git branch, recent commits, uncommitted changes, and the conversation; drafts a handoff for you to confirm; then writes .claude/continuations/YYYY-MM-DD-{topic}.md (and a learnings file only if something non-obvious came up). No commits, no lint.
/continue — find, reality-check, orientReads the most recent handoff, runs read-only git status/git log to flag branch mismatches or new commits, and prints a tight orientation: what was done, where you left off, the next task, the key files. It writes nothing.
/next — launch the next sessionPicks a target worktree, detects your terminal (Terminal.app, Ghostty, iTerm2, Warp), and opens a new window at that directory with a stable per-worktree colour + title, auto-running /continue. macOS only; it only launches — never commits.
Filenames start with the date, so ls produces a natural timeline. It’s your call whether to commit them — gitignore them for private notes, or commit for team-wide session history.
your-project/ └── .claude/ ├── continuations/ │ ├── 2026-05-27-user-profile-endpoint.md │ ├── 2026-05-28-fix-validation-bug.md │ └── 2026-05-30-add-pagination.md └── learnings/ └── 2026-05-27-zod-error-handler.md
The scope is intentionally narrow. Keeping each command sharp on its one job is the whole point. Pair it with other commands for the rest.
/end yourself.Use skills.sh for the portable skill, or add the marketplace for Claude Code’s slash commands. After installing the plugin, type / to see /end and /continue.
/plugin marketplace add amjad1233/ai-skills/plugin install continuation@amjad1233