AgentHandover

What AgentHandover does

You click Record in the menu bar, perform a task once, then stop. AgentHandover asks up to three questions about what the recording could not show, such as what starts the task or when it is done. You review the generated Skill in the menu bar app and click Approve for Agents. A Skill holds the steps, the strategy behind them, decision logic, guardrails and your writing voice, and it states plainly what it does not know yet.

An agent is connected with one command, for example agenthandover connect followed by claude-code, codex, openclaw or hermes. Agents report back after every run. Runs that follow the steps confirm them, and a run that deviates where a decision was only a guess becomes a question for you. Any MCP-compatible tool can also use the Skills. The README gives examples such as research routines, community engagement, support triage, data extraction and ops checklists.

Key features

• Menu bar recorder that learns a workflow from one demonstration
• Up to three follow-up questions about what the recording missed
• Skills with steps, strategy, decision logic, guardrails and writing voice
• One-command connection for Claude Code, Codex, OpenClaw and Hermes
• Skills that update from agent run reports
• Review and approval step before agents can use a Skill

When to use it

• Handing a daily community-engagement routine to a Hermes agent
• Teaching an agent your support triage steps
• Turning a release or deploy checklist into a runnable Skill

Who it is for: Mac users who want agents to repeat their own workflows without writing prompts or skills by hand.

How it fits with Hermes Agent

Supports Hermes Agent among several agents: once a Skill is approved, one connect command links it to Hermes.

Requirements: A Mac, since it observes your work there and installs from a .pkg release; onboarding requests permissions and downloads models sized to your Mac's memory

FAQ