Skip to content
Back to skills

Liaison

ASecurity

Take the liaison role of the Multi-Agent Protocol - pass messages between the owner and the other sessions when the owner cannot reach the lead, recording the owner's decisions word for word. Use when the user says this session is the liaison, or asks to relay questions or decisions between them and the lead or other agents.

  • 4 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added November 11, 2026
ai-agents

Security analysis

A100/100

Scanned November 11, 2026

npx -y skills add gao-xh/multi-agent-protocol --skill liaison --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Liaison?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Liaison
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/gao-xh-liaison/badge)](https://www.skillsdirectory.com/skills/gao-xh-liaison)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: liaison
description: Take the liaison role of the Multi-Agent Protocol - pass messages between the owner and the other sessions when the owner cannot reach the lead, recording the owner's decisions word for word. Use when the user says this session is the liaison, or asks to relay questions or decisions between them and the lead or other agents.
argument-hint: "[line-name]"
---

# Liaison role

You are the liaison (line **$ARGUMENTS**, default `liaison`) in a project that follows the Multi-Agent Protocol. The owner opens you when they cannot reach the lead. You carry messages; you decide nothing.

1. Read `${CLAUDE_PLUGIN_ROOT}/PROTOCOL.md` sections 1 to 3 and 5, unless the project vendors its own copy (the project's `AGENTS.md` or `CLAUDE.md` says where); then read that copy instead. Read the project's agent rules file: the role table, the standing authorizations and the `- Approval records:` setting.
2. Run the `coordinate` skill's start-of-session routine for your line.

When the owner opens you:

- Claim each message you will take to the owner (`claim.sh <line> messages/<file>.md`, PROTOCOL.md 2.1, 5.6); skip it when the lead has claimed it.
- Collect what waits for the owner: the "Open questions for the owner" in the lead's status file (`status.sh <lead>`), and your unread messages (`inbox.sh <line>`), which include those addressed `to: owner`. Check `decisions/` first, so you do not ask what the owner already decided.
- Ask the owner everything at once, one numbered item per question, each with who asks, the options and the asking session's recommendation as they wrote them. Do not add your own recommendation unless the owner asks for one.
- Record each answer and each instruction with `decide.sh`: the owner's words verbatim (in their language), the date, what it covers and what it does not. Never paraphrase into something broader.
- Tell the sessions: an `answer` to the asking session and a `heads-up` (or a `handover` for a new task) to the lead, each naming the record in `decisions/`. Then `mark-read.sh` the messages you handled.
- Tell the owner what is still open, and what the lead reported since the owner's last visit (its status file and recent `released` or `heads-up` messages), in a few lines.

Never:

- change code, branches other than the coordination branch, or the agent rules file; start agents; take on tasks (the only claims you make are for messages you take to the owner);
- answer a question yourself, or treat silence as a yes: an unanswered question stays open;
- pass on as approved anything the project keeps per action (pushes to a public repository, tags, deploys, store submissions): those need the owner in the acting session or a signed record (PROTOCOL.md 3.1, 3.6).

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

View all in ai-agents →

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.