Install the Plexarm plugin

One install for Claude Code. It connects Claude Code to the Plexarm server and adds a skill that explains how the method works, an agent that keeps the record in order, and hooks you can switch off one by one.

What it adds

  • The connection to the Plexarm MCP server — the same URL and header as a manual connection, with the token read from your machine’s credential store at every connect.
  • A skill: how to pick up work, record what was done, and file what you noticed, in the words the server itself uses.
  • An agent, the Plexarm chief of staff: sweeps the record for what is late, unowned or unlinked, and reports the counts.
  • Six hooks, listed below with what each one sends. Each can be switched off.

What the hooks send

What you and your agents write through the Plexarm tools goes to Plexarm; that is what the tools are for. Beyond that, the hooks send only what is listed here, over HTTPS to api.plexarm.com, with your token.

  • Credential check

    At session start. Silent unless no token is stored, or your Claude Code settings hold a second token: then it asks Plexarm which account each token belongs to, and if they differ, names both and asks Claude to stop rather than record anything.

    Switch off: PLEXARM_HOOKS_OFF=session_start_credential_check

  • Agent roster sync

    At session start, sends your agents’ names and, where a file sets one, who each reports to. For agents in ~/.claude/agents/ it also sends the display name and role; that folder is not tied to one project, so those names reach your account from every project. It does not send an agent’s prompt, description or tools.

    Switch off: PLEXARM_HOOKS_OFF=session_start_agent_sync

  • Due notice

    At the start of a new session, one request with no body, to list any routines or reminders due for you. If none are due, it prints nothing.

    Switch off: PLEXARM_HOOKS_OFF=session_start_due

  • Dispatch capture

    Each time an agent starts another, it asks Plexarm whether you switched capture on in your profile. Only if you did, it sends the type of each of the two agents and Claude Code’s id for that call. Off until you switch it on in your profile; the line below also stops the question.

    Switch off: PLEXARM_HOOKS_OFF=post_tool_use_dispatch

  • Spawn notice

    When a subagent starts, adds a reminder to record its work to that subagent’s context. It sends nothing.

    Switch off: PLEXARM_HOOKS_OFF=subagent_start_context

  • Record guard

    Off by default. When switched on, at the end of a session and of each subagent it sends the project alias the repository declares and asks whether work was left in progress there; if so, it gives that agent one more turn with the list.

    Switch on: PLEXARM_HOOKS_ON=stop_record_guard

Put a hook’s line in your shell profile or in the env block of Claude Code’s settings.json; it applies to that machine. Several hook names can share one line, separated by commas.

The token

The plugin reads your token from your machine’s credential store at every connect: the Keychain on macOS, secret-tool on Linux, or a file at ~/.config/plexarm/token that only you can read. After installing, run /plexarm:connect: it walks you through putting the token there, without asking you to paste it into the chat. The plugin’s connection does not read PLEXARM_TOKEN; some hooks fall back to it when the store is empty. Create a token in API tokens

Install

Type these at the Claude Code prompt — not in a shell

/plugin marketplace add natalabs/plexarm-plugin
/plugin install plexarm@plexarm

Or from a shell

claude plugin marketplace add natalabs/plexarm-plugin
claude plugin install plexarm@plexarm

To get updates automatically, turn them on once: /plugin → Marketplaces → plexarm → Enable auto-update.

Other clients

The plugin is a Claude Code install. Every other client connects to the same server with a URL and a header — see Connect your agent. Connect your agent