How do I install the Relay skill?

Run npx skillmds@latest add gavishap/relay in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Relay safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Relay?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Relay free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

gavishap (@gavishap) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Relay by gavishap — Agent Skill

Name: Relay
Author: gavishap

Relay

Hand work between Claude Code and Codex (or any two agents) with zero copy-paste, via the _relay/STATE.md baton. Use at the start of a session ("catch up", "resume", "where were we") and at the end ("save", "hand off", "wrap up"), or whenever the user says they're switching agents.

gavishap Updated Sep 21, 2026

File contents

Relay — agent-to-agent handoff

The other agent cannot see this chat. Everything it needs must live in the repo: the baton (_relay/STATE.md), the history (_relay/HISTORY.md), git, the Wiki, and the chat archive. Full protocol: _relay/PROTOCOL.md.

Catch up (session start)

Read _relay/STATE.md in full.
python scripts/relay_tool.py status — heed the stale-baton warning; if it fires, trust git log --oneline -15 + the newest Wiki/Logs/ note over STATE.md, then repair STATE.md.
Summarize to the user in a few sentences: where things stand, what's next. Do NOT re-read raw sources, sweep repos, or replay old chats to resume.

Hand off (session end, or "switching to Codex/Claude")

Rewrite the sections of _relay/STATE.md so they describe the PRESENT:
- Now — the one-paragraph state of the project.
- Just landed — what this session finished (facts, not effort).
- Next — checkboxed, concrete, ordered next actions.
- Open questions — decisions only the user can make.
- Watch out — anything broken, red, flaky, or half-done. Never omit.
Write it for a stranger: no "as discussed", no chat-local shorthand, name files by path, name commands verbatim.
Stamp: python scripts/relay_tool.py stamp --agent claude --summary "<one paragraph>" (Codex stamps with --agent codex.)
Archive transcripts (both agents): python scripts/import_chats.py.
If the session changed knowledge, add a Wiki/Logs/ note + wiki_tool.py log.
Run the maintenance gate (see AGENTS.md), then commit. The baton travels with the commit.

Rules

One baton — no private TODO files. Short-term state → STATE.md; durable knowledge → the Wiki.
Never hand off a lie: red tests and failing gates go in Watch out.
A handoff without a commit is half a handoff.

Sync mode

The baton is the relay's async mode (agents alternate across sessions). For the synchronous mode — both models in ONE session, the rival attacking the driver's plan and diff in bounded rounds — use the sparring skill (/spar). Its state lives beside the baton in _relay/spar/; an active or interrupted spar is part of the handoff (mention it in Now / Watch out, and python scripts/spar_tool.py status shows where it stands).

---name: relaydescription: Hand work between Claude Code and Codex (or any two agents) with zero copy-paste, via the _relay/STATE.md baton. Use at the start of a session ("catch up", "resume", "where were we") and at the end ("save", "hand off", "wrap up"), or whenever the user says they're switching agents.---# Relay — agent-to-agent handoffThe other agent cannot see this chat. Everything it needs must live in the repo:the baton (`_relay/STATE.md`), the history (`_relay/HISTORY.md`), git, the Wiki,and the chat archive. Full protocol: `_relay/PROTOCOL.md`.## Catch up (session start)1. Read `_relay/STATE.md` in full.2. `python scripts/relay_tool.py status` — heed the stale-baton warning; if it fires, trust `git log --oneline -15` + the newest `Wiki/Logs/` note over STATE.md, then repair STATE.md.3. Summarize to the user in a few sentences: where things stand, what's next. Do NOT re-read raw sources, sweep repos, or replay old chats to resume.## Hand off (session end, or "switching to Codex/Claude")1. Rewrite the sections of `_relay/STATE.md` so they describe the PRESENT: - **Now** — the one-paragraph state of the project. - **Just landed** — what this session finished (facts, not effort). - **Next** — checkboxed, concrete, ordered next actions. - **Open questions** — decisions only the user can make. - **Watch out** — anything broken, red, flaky, or half-done. Never omit.2. Write it for a stranger: no "as discussed", no chat-local shorthand, name files by path, name commands verbatim.3. Stamp: `python scripts/relay_tool.py stamp --agent claude --summary "<one paragraph>"` (Codex stamps with `--agent codex`.)4. Archive transcripts (both agents): `python scripts/import_chats.py`.5. If the session changed knowledge, add a `Wiki/Logs/` note + `wiki_tool.py log`.6. Run the maintenance gate (see `AGENTS.md`), then commit. The baton travels with the commit.## Rules- One baton — no private TODO files. Short-term state → STATE.md; durable knowledge → the Wiki.- Never hand off a lie: red tests and failing gates go in **Watch out**.- A handoff without a commit is half a handoff.## Sync modeThe baton is the relay's **async** mode (agents alternate across sessions).For the **synchronous** mode — both models in ONE session, the rival attackingthe driver's plan and diff in bounded rounds — use the `sparring` skill(`/spar`). Its state lives beside the baton in `_relay/spar/`; an active orinterrupted spar is part of the handoff (mention it in **Now** / **Watchout**, and `python scripts/spar_tool.py status` shows where it stands).

Frequently asked questions