Agent skill

Relay Runner

by jmfederico in jmfederico/pi-web

Prepare and run Relay legs in Pi: keep the goal, baton, and log durable, then hand off to one fresh session or stop visibly.

MITAuto-check passedAgent Workflows

Install Relay Runner

skills CLI
$ npx skills add jmfederico/pi-web --skill relay-runner -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install jmfederico/pi-web relay-runner --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/jmfederico/pi-web.git skills-src && mkdir -p .claude/skills && cp -r skills-src/skills/relay-runner .claude/skills/relay-runner && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
relay-runner
GitHub stars
869
Token cost
~2.2k tokens
SKILL.md length
1,223 words
Files
1
Skills in repo
7
Repo updated
First seen
Licence
MIT

At a glance

Prepare and run Relay legs in Pi: keep the goal, baton, and log durable, then hand off to one fresh session or stop visibly.

  • Works in 5 steps: Read the charter and status, then… → Compare reality with the finish line and… → Do the work and the checks appropriate… → …
  • Tasks that involve Git worktrees
  • SKILL.md covers Packet, Preparation and dispatch, Run a leg and Final review, plus 1 more section
  • Instructions only: no scripts, shell commands, URLs or credentials in SKILL.md

What it does

Relay Runner is an agent skill from jmfederico/pi-web. Prepare and run Relay legs in Pi: keep the goal, baton, and log durable, then hand off to one fresh session or stop visibly. Use with relay for /relay, /relay-worktree, or a handoff naming relay-runner. Supplies continuity and a bounded fresh-context final review without prescribing the development route.

Its SKILL.md is about 2.2k tokens, which your agent loads only when the skill is triggered. It is a single SKILL.md file with no bundled scripts.

It sits in Agent Workflows, covering Git worktrees. The repository describes itself as: Web UI for Pi Coding Agent that keeps sessions alive in real workspaces. The licence is MIT.

When your agent uses it

  • Tasks that involve Git worktrees

Example prompts

  • “/relay-runner”

Workflow steps

5 steps, taken from the first numbered list in SKILL.md.

  1. Read the charter and status, then inspect relevant project guidance and work. Consult only the history or supporting material needed to…
  2. Compare reality with the finish line and completion requirements, including the final review below. If all are satisfied, record…
  3. Do the work and the checks appropriate to it under project conventions. Record failed, skipped, or incomplete checks honestly.
  4. Make the result durable before handoff. In Git, commit the leg's changes when project policy and human instructions permit, without…
  5. Refresh the compact status and append the log entry. Record progress against the goal, important evidence and decisions, remaining gaps…

What it can do on your machine

Read from SKILL.md and the folder at commit 2961669. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    No scripts in the folder and no shell commands in SKILL.md.

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Relay Runner loads about 2.2k tokens when it runs. Until then it costs about 80 tokens; SKILL.md has 1,223 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~80
When it runs · the whole SKILL.md, loaded when a task matches
~2.2k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check passed

The automated check found no risky patterns in SKILL.md.

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from jmfederico/pi-web at commit 2961669, republished under its MIT licence (© jmfederico). 1,223 words, ~2,220 tokens.

Download SKILL.mdSave it as .claude/skills/relay-runner/SKILL.md (or your agent's skills folder).
name
relay-runner
description
Prepare and run Relay legs in Pi: keep the goal, baton, and log durable, then hand off to one fresh session or stop visibly. Use with relay for /relay, /relay-worktree, or a handoff naming relay-runner. Supplies continuity and a bounded fresh-context final review without prescribing the development route.

Relay runner

This skill binds the relay method to Pi sessions and a small file packet. It supplies continuity and one completion gate: a fresh-context final review. The development route remains adaptive. Each agent discovers applicable project instructions and skills through the repository and harness; do not copy them into the packet or preselect them for future legs.

Packet

Use .pi-web/relays/<name>/ in the working checkout unless the user selects another location. Start with three files:

  • charter.md — goal. The agreed outcome and recognizable finish line, relay-specific scope boundaries and clarifications, and any explicit human requirements. Reference an existing specification when useful. Do not add project rules, implementation plans, or generic quality or delivery procedures; include such details only when the human explicitly asks.
  • status.md — baton. State (Draft — awaiting approval; not dispatched, Active, Complete, or Needs help), last completed leg and next leg number, working location, current progress, remaining gaps to the goal, relevant checks, current review state, targeted pointers, and blockers. Include only what is useful now. Avoid suggesting the next slice; record remaining gaps and concrete dependencies instead.
  • log.md — history. Append a short entry per leg: what changed or was learned, important decisions and why, artifact or commit references, checks and actual results, and the next leg or stop. Record human approval and scope changes here too.

Charter = agreement; status = position; decisions = durable reasoning; log = history. Keep revisions in the log, not a separate revisions document; no decision template is required. Status, decisions, and the log cannot silently change the agreement. Keep relay-specific decisions durable without adding documents merely to satisfy a template. The packet is not a second home for project policy or this skill's instructions.

Add decisions.md only when useful: a compact, living record of consequential decisions and their reasons, including rejected approaches or review concerns worth remembering. Keep current conclusions there and record revisions in the log. Link it from status and consult relevant entries rather than duplicating the reasoning in the baton. It is not a plan, task list, or another home for project rules. Human changes to the goal or scope still belong in the charter. Do not create an empty register or impose a decision template.

Preparation and dispatch

The preparation prompt owns discussion and approval. Drafting is allowed before approval; spawn_session is not. After the human reviews the goal and boundaries, require an explicit Approve and dispatch response. If the approved goal, scope, or working target changes materially, obtain fresh approval.

Use the current checkout for /relay unless the user chooses another. For /relay-worktree, create a fresh branch and worktree after approval, following project conventions; use current HEAD as the starting point unless another base is agreed. Ask about a material target ambiguity rather than guessing. Move the draft packet into the target checkout before dispatch, update its location, and remove the stale drafting copy.

Record the actual checkout and, in Git, branch and starting commit in status. Keep packet files out of delivery commits using an existing ignore rule or a local Git exclusion. Leave unrelated existing work untouched. Bootstrap the target as needed under project conventions; setup does not automatically require a separate leg.

Record approval in status/log, mark status Active, and seed leg 1. The first runner chooses its own slice; preparation need not prescribe one. Dispatch with the same handoff mechanics used by later legs.

Run a leg

  1. Read the charter and status, then inspect relevant project guidance and work. Consult only the history or supporting material needed to understand the current state. Repair small record gaps with targeted inspection; ask for help if the goal or ownership is unclear.
  2. Compare reality with the finish line and completion requirements, including the final review below. If all are satisfied, record completion and stop. Otherwise choose one useful, context-sized slice that closes a remaining gap. Briefly state in the session which slice you chose and how it advances the goal.
  3. Do the work and the checks appropriate to it under project conventions. Record failed, skipped, or incomplete checks honestly.
  4. Make the result durable before handoff. In Git, commit the leg's changes when project policy and human instructions permit, without including unrelated changes or the packet. Otherwise record the exact saved work and why it is uncommitted. Finish any helpers before handing off.
  5. Refresh the compact status and append the log entry. Record progress against the goal, important evidence and decisions, remaining gaps, and any limitations the successor needs to know. Then complete, ask for help, or hand off once.

Other reviews, delivery steps, and gates apply only when the human or project requires them. Keep this skill's review procedure here, not copied into the packet.

Show full SKILL.md (444 more words)Show less

Final review

When a leg finishes substantive work and believes implementation is done, record that final review is pending and hand off to a fresh leg. A fresh runner finding implementation complete with review pending performs the review instead of passing it on again. Leave substantive corrections to another leg so the review stays independent.

Review the whole relay result, not just the last slice, against the charter and applicable project guidance. Approve when no concrete, evidence-backed in-scope blocker remains. Optional improvements, speculative concerns, and unrelated pre-existing shortcomings do not block completion. Record consequential finding dispositions in decisions.md when useful.

Allow at most three review attempts total: the initial review and up to two post-correction reviews, not three mandatory passes. Keep the attempt count, latest result, and reviewed work reference in status; put evidence and findings in the log or link to decisions. Subsequent attempts happen only after concrete corrections and focus on those corrections and changed evidence, carrying supported prior decisions forward rather than restarting an expanding audit.

If the third attempt still has blockers, or a required further review would exceed the budget, mark Needs help, record what remains, and stop without dispatching another review or correction leg. A valid approval satisfies the gate; a new context or packet-only update does not reopen it. Material changes to the reviewed result require re-review within the same budget.

Stop or hand off

Mark Complete and report the result and evidence when no required work remains and final review has passed for the current result. Do not spawn a successor to search for extra work.

Mark Needs help when a decision, permission, environment problem, or lack of progress prevents responsible continuation. Record the concrete blocker, attempts and evidence, and the smallest question or action needed from the human. Surface it clearly in the user-facing response; use ask_user when an answer is needed. Do not hand a human-dependent blocker to another context hoping it will disappear.

Before continuing, check that concrete required work remains, including any pending final review; leave slice selection to the successor.

For handoff, finish work and all packet writes first. Call spawn_session at most once, with the target checkout as cwd. Omit model and thinkingLevel to inherit them unless the human instructed otherwise; keep any relay-wide instruction in the packet. Use an independent session, not a tracked subsession, for the successor.

Keep the successor prompt short, substituting actual paths and the next leg number:

text
Continue Relay "<name>", leg <N>, in <checkout>.
Load the `relay` and `relay-runner` skills.
Read <packet>/charter.md and <packet>/status.md.
Check the finish line against the current work and choose the next useful slice.
Make progress and the record durable, then hand off once, complete, or signal that you need help.

spawn_session is the final operational action. After it returns, give only a brief handoff summary: no further tool use, packet writes, work, or downstream supervision. If dispatch fails, report the failure rather than claiming the next leg started.

© jmfederico, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

Just SKILL.md in skills/relay-runner of jmfederico/pi-web.

Open the folder on GitHubat commit 2961669

Compare with similar skills

Relay Runner next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Relay Runner compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Relay Runner this skilljmfederico/pi-web869—~2.2kAutomated safety check: PassMIT
Orca CLIstablyai/orca88k2 repos~593Automated safety check: PassMIT
ClawTeam Multi-Agent Swarmwin4r/ClawTeam-OpenClaw1.5k1 repos~2.9kAutomated safety check: PassMIT
Using Git Worktreesjulianromli/opencode-template14428 repos~1.4kAutomated safety check: PassNone
Agent of Empires Session Manageragent-of-empires/agent-of-empires3.3k—~2.5kAutomated safety check: PassMIT
Nagentdavidondrej/skills4.1k—~1.7kAutomated safety check: NotesMIT

Similar skills

  • Orca CLI

    stablyai/orca

    Operate Orca-managed worktrees, folder contexts, terminals, repos, automations, artifacts, skill sharing, worktree comments, and Orca's embedded browser…

    88k GitHub starsUsed in 2 repos~593 tokens
    Agent WorkflowsAuto-check passed
  • ClawTeam Multi-Agent Swarm

    win4r/ClawTeam-OpenClaw

    Launches a swarm of specialist Hermes agents in git-worktree-isolated tmux windows with a kanban board and file-based inboxes, using built-in templates like hedge-fund and code-review.

    1.5k GitHub starsUsed in 1 repo~2.9k tokens
    Agent WorkflowsAuto-check passed
  • Using Git Worktrees

    julianromli/opencode-template

    A skill your agent uses when starting feature work that needs isolation from current workspace or before executing implementation plans - creates isolated git worktrees with smart directory…

    144 GitHub starsUsed in 28 repos~1.4k tokens
    Agent WorkflowsAuto-check passed
  • Agent of Empires Session Manager

    agent-of-empires/agent-of-empires

    Launches, monitors and organizes AI coding agent sessions such as Claude Code or Codex inside tmux, tracking status, capturing output and managing git worktrees for parallel branches.

    3.3k GitHub stars~2.5k tokensUpdated today
    Agent WorkflowsAuto-check passed
  • Nagent

    davidondrej/skills

    Launch a new bb worker thread with the right project, model, worktree, and task brief.

    4.1k GitHub stars~1.7k tokensUpdated yesterday
    Agent WorkflowsAuto-check: notes
  • Agent Deck

    asheshgoplani/agent-deck

    agent-deck, the terminal session manager for AI coding agents.

    1k GitHub stars~1.7k tokensUpdated 3 days ago
    Agent WorkflowsAuto-check passed

More from jmfederico/pi-web

  • Changeset Changelog

    jmfederico/pi-web

    A skill your agent uses whenever the user asks about changelogs, Changesets, release notes, conventional commits, commit messages for release notes, or making user-visible project changes that…

    869 GitHub stars~1.5k tokensUpdated today
    Auto-check passed
  • A skill your agent uses whenever the user asks for a new npm version, npm release, package release, new release, version bump, publishing to npm, cutting a GitHub release, tagging a release, or…

    869 GitHub stars~2.9k tokensUpdated today
    Auto-check passed
  • Testing Guide

    jmfederico/pi-web

    Repository-specific testing guide. An agent skill from jmfederico/pi-web.

    869 GitHub stars~2.6k tokensUpdated today
    Auto-check passed
  • Code Quality Architecture

    jmfederico/pi-web

    Project code quality and architecture expectations for implementation, refactoring, planning, and code review.

    869 GitHub stars~2.1k tokensUpdated today
    Auto-check passed
  • Documentation Guide

    jmfederico/pi-web

    Repository documentation placement and writing guidance. An agent skill from jmfederico/pi-web.

    869 GitHub stars~1.7k tokensUpdated today
    Auto-check passed
  • Relay

    jmfederico/pi-web

    Carry work across a chain of fresh agent contexts using a shared goal, compact baton, and durable record.

    869 GitHub stars~959 tokensUpdated today
    Auto-check passed

Categories

Questions about Relay Runner

What does Relay Runner do?

Prepare and run Relay legs in Pi: keep the goal, baton, and log durable, then hand off to one fresh session or stop visibly. Relay Runner is an agent skill from jmfederico/pi-web. Prepare and run Relay legs in Pi: keep the goal, baton, and log durable, then hand off to one fresh session or stop visibly.

When should I use Relay Runner?

Relay Runner fits situations like: tasks that involve Git worktrees.

How do I install Relay Runner in Claude Code?

Run `npx skills add jmfederico/pi-web --skill relay-runner -a claude-code`. Or copy the skill folder (skills/relay-runner in jmfederico/pi-web) into .claude/skills/relay-runner in your project. Claude Code loads it when a task matches its description.

How do I install Relay Runner in Codex?

Run `npx skills add jmfederico/pi-web --skill relay-runner -a codex`. Or copy the skill folder (skills/relay-runner in jmfederico/pi-web) into .agents/skills/relay-runner in your project. Codex loads it when a task matches its description.

Can I use Relay Runner in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add jmfederico/pi-web --skill relay-runner -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/relay-runner, .gemini/skills/relay-runner, .github/skills/relay-runner and .opencode/skills/relay-runner in your project.

What does Relay Runner need to run?

SKILL.md names no scripts, command-line tools or credentials: Relay Runner is instructions for the agent only.

Does Relay Runner access the network?

SKILL.md contains no URLs. Any network use would come from the scripts or tools the agent runs. This is read from the text; nothing was executed.

Is Relay Runner safe to install?

Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.

What licence does Relay Runner use?

Relay Runner is published under the MIT licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Relay Runner use?

About 2.2k tokens (SKILL.md is roughly 8.9k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full.

What are the alternatives to Relay Runner?

Skills that share tags, products or a category with Relay Runner: Orca CLI (stablyai/orca, 88k stars), ClawTeam Multi-Agent Swarm (win4r/ClawTeam-OpenClaw, 1.5k stars), Using Git Worktrees (julianromli/opencode-template, 144 stars) and Agent of Empires Session Manager (agent-of-empires/agent-of-empires, 3.3k stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Relay Runner?

jmfederico (a GitHub user) maintains it in jmfederico/pi-web, which has 869 GitHub stars. The repository holds 7 skills in this directory. The repository was last updated on October 9, 2026.

Source: jmfederico/pi-web on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.