handoff skill (rohitg00) for Claude Code: pick up where you left off

handoff makes your agent reload your last session in this folder and tell you where you stopped.

Installs: 13.7K; on skills.sh
Stars: 30.1K; on GitHub
Files: 2; besides SKILL.md
Updated: Oct 2026; last repo commit

handoff is one of the skills in rohitg00's agentmemory project. When you say "where were we", "resume" or "pick up where I left off", it finds the most recent session recorded for your current folder. It puts any unanswered question first, summarizes the key files, decisions and errors, and ends with one next step.

It does not work on its own. It calls two tools from agentmemory (memory_sessions and memory_recall), so the agentmemory server and its connection to your agent must be running first. Install it from the box above, which covers the server too. It matches sessions by folder boundary, so a sibling folder like repo-a-staging is never mistaken for repo-a. If no session matches, it says so and stops. If the matched session is empty, it reports that and offers a fresh start.

Good for

Developers who switch between agents or restart sessions often
Anyone already running agentmemory who wants a one-word resume
People who lose track of the last open question after a break

Install handoff

Claude Code

Copy this prompt.

Install the handoff skill in Claude Code. Run this command:

npx -y @agentmemory/agentmemory@latest
/plugin marketplace add rohitg00/agentmemory
/plugin install agentmemory

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

Requires Node.js 20 or newer.

From the project README. Source.

What's in the skill

When it loads

Resume the most recent agent session for the current working directory, leading with any unanswered question. Use when the user says "where were we", "resume", "handoff", "pick up where I left off", or starts a session with no fresh context.

Sections in SKILL.md

Quick start
Why
Workflow
Anti-patterns
Checklist
See also
Troubleshooting

Key facts

Author: rohitg00 (GitHub account that owns the repo)
Source repo: rohitg00/agentmemory
Installs on skills.sh: 13,691 as of 2026-11-02
GitHub stars: 30,069 (whole repo) as of 2026-11-02
License: Apache-2.0
Last repo commit: 2026-10-29
Files: 3 files including SKILL.md (plus EXAMPLES.md and TROUBLESHOOTING.md)
skills.sh audits: Gen Agent Trust Hub pass, Socket pass, Snyk pass
Part of: agentmemory plugin (17 skills, 12 hooks, bundled MCP bridge)
Requires: A running agentmemory server (Node.js 20+) and its memory_sessions and memory_recall tools

What it includes

Calls memory_sessions and picks the newest session whose folder matches.
Matches by exact folder or parent/child folder, never a raw prefix. Prefers completed over abandoned sessions. Stops if nothing matches.
Network · Find the session
Pass a path after the command to resume a different folder.
The argument hint is [optional cwd override]. The path is turned into an absolute path first.
Read · Optional folder override
Puts an unanswered user-facing question first in the reply.
It looks in the session summary and recent conversation notes for a narrative that ends in a question mark.
Read · Surface the open question
Calls memory_recall on the top concepts, limit 10, within this project.
Checks each result belongs to this session or project before using it.
Network · Recall context
Ends with one concrete next-step pointer.
An empty session is reported plainly with an offer to start fresh.
Next step

From the official docs

Permissions

Medium risk

The handoff skill only reads saved sessions, but the agentmemory plugin it depends on records your prompts and tool use and runs a local server. Install the plugin only if you are comfortable with that.

What it can do

Read your data
Create and change things
Delete things
Read-only mode

Watch out for

The plugin hooks record your sessions
The Claude Code plugin registers 12 hooks (plugin/hooks/hooks.json): SessionStart, UserPromptSubmit, PreToolUse (Edit|Write|Read|Glob|Grep), PostToolUse, PostToolUseFailure, PreCompact, SubagentStart, SubagentStop, Notification, TaskCompleted, Stop and SessionEnd. The scripts send your prompts, tool names and outputs, and the last assistant reply (cut to 4,000 characters), to the memory server. The README says API keys, secrets and <private> tags are stripped before storage, and the Codex plugin registers 6 hooks instead. Review what is stored in the viewer. github.com
Hooks post to AGENTMEMORY_URL
The hook scripts send events to AGENTMEMORY_URL, which defaults to http://localhost:3111. If you set it to another host, your session content goes there. Failed posts are spooled to disk under the agentmemory data folder (capture-spool) for replay, unless AGENTMEMORY_CAPTURE_SPOOL is false. Setting AGENTMEMORY_INJECT_CONTEXT=true makes SessionStart and PreToolUse inject recalled memory into your sessions. github.com
Setup runs npx and starts a local server
The install runs @agentmemory/agentmemory with npx, writes configuration and starts a daemon on port 3111 with a viewer on 3113. The README says both bind to loopback by default. github.com
The plugin adds a bundled MCP bridge and 17 skills
Its .mcp.json starts a node script, plugin-bridge.mjs, from the plugin folder. The README says the plugin exposes 54 MCP tools. github.com
Optional LLM provider
Compression of observations needs a provider key and AGENTMEMORY_AUTO_COMPRESS=true. Leave it off if you do not want session content sent to a model provider. github.com

Technical details

Talks to	localhost:3111 (memory API; AGENTMEMORY_URL can point elsewhere). localhost:3113 (viewer)
Sign-in and keys	The README says the daemon generates a random secret on first start and stores it in ~/.agentmemory/secret with mode 0600.
Scopes	agentmemory daemon on localhost:3111 ~/.agentmemory/ (data, secret, .env)

Questions

How do I install the handoff skill?

Use the install box above. It starts the agentmemory server and installs the agentmemory plugin, which carries handoff and its 16 sibling skills. Installing only the skill folder gives you a skill with no memory tools to call.

Does handoff need an API key?

The skill itself does not ask for one. The README says keyword recall works without an account or model API key. An LLM provider key is only needed for optional features such as compressing observations, which also needs AGENTMEMORY_AUTO_COMPRESS=true.

What happens if there is no earlier session?

It tells you this project has no matching session and stops. It does not borrow a session from another project. If the matched session has no recorded observations, it says there is nothing to hand off and offers to start fresh.

Does handoff work in Codex or the Claude app?

Codex: yes, the README documents a Codex plugin route, shown in the Codex tab. Claude app: the README does not document it, and the skill needs local memory tools, so use Claude Code.

Is the handoff skill safe to install?

The skill folder is three Markdown files and bundles no scripts. The plugin around it is a bigger change: it registers hooks that record your prompts and tool use into a local memory store. Read the safety section before installing the plugin.

handoff vs recap, session-history and recall?

SKILL.md lists recap, session-history and recall as skills that read the same session data with broader views. handoff is the narrow one: it resumes the single most recent session for this folder and leads with the open question.

Information

updated Nov 2, 2026

Repository: rohitg00/agentmemory
Folder: plugin/skills/handoff

rohitg00/agentmemory skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 8

agentmemory README · github.com
rohitg00-handoff SKILL.md · github.com
handoff EXAMPLES.md · github.com
rohitg00-handoff on skills.sh · skills.sh
handoff TROUBLESHOOTING.md · github.com
hooks.json · github.com
.mcp.json · github.com
_capture.mjs · github.com