relay
Skill882 starsChanged 16 days agoMITView on GitHub
Carry work across a chain of fresh agent contexts using a shared goal, compact baton, and durable record. Use when a user invokes Relay, discusses the Relay method, prepares a Relay, or continues an active Relay. Do not load for generic multi-step work or ordinary delegation.
We split this skill into the rules it gives an agent
7 rules. 0 are common, 7 are distinctive, and 0 are very specific to this project.
0
Common
Most of the 19 edited versions we read keep these.
7
Distinctive
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 10
Follow the Relay Principle: give agents a destination and enough continuity to act, rather than recreating a management structure around …
- Distinctive: line 16
Keep the goal stable and the route adaptable.
- Distinctive: line 22
Do not invent an implementation roadmap, speculative requirements, or a new software process.
- Distinctive: line 34
Keep three kinds of durable information:
- Distinctive: line 40
Keep the baton compact. Use history for targeted lookup, not mandatory reconstruction.
- Distinctive: line 44
Stop when the finish line is met, with evidence.
- Distinctive: line 46
Stop visibly when progress needs human input: record the blocker, what was tried, and the specific decision or help needed.
“Common” is measured: kept by at least 6 in 10 of the 19 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Relay
A Relay carries work across fresh agent contexts until the agreed goal is reached or an agent needs human help. Each context owns one useful slice, called a leg. There is no standing coordinator or fixed set of agent roles.
Follow the Relay Principle: give agents a destination and enough continuity to act, rather than recreating a management structure around them.
Finish line over plan
Help the human express what they want built and how they will recognize it is done. Resolve material ambiguity before dispatch. An existing requirements document can supply the goal; reference it rather than rewriting it into a larger specification.
Keep the goal stable and the route adaptable. Each runner checks the finish line against the current work and chooses the next useful slice. Changing the goal requires human agreement; changing implementation or sequencing within scope does not.
Scope over script
Record the relay-specific boundaries, limitations, conditions, and clarifications needed to understand the request. Preserve technical requirements and any other details the human explicitly asks to include.
Do not invent an implementation roadmap, speculative requirements, or a new software process. Project instructions, skills, and harness conventions remain where they belong; each runner discovers and follows what applies to its work. A Relay does not replace or weaken those rules.
Handoff over handhold
A baton says where we are, not how you must work. Avoid suggesting or assigning the next slice: even a provisional suggestion can anchor the successor. Record remaining gaps and concrete dependencies instead, and trust the successor to inspect reality and choose its slice.
Choose a meaningful increment that fits one context, not the tiniest possible task. Investigation and related work can stay together. Leg boundaries protect context quality, not organizational ownership.
Finish ongoing work and preserve the state before starting at most one fresh successor. Handoff is the final operational action; only a user-facing summary follows. Do not watch or steer the successor.
Record over memory
Keep three kinds of durable information:
- Goal: the agreed finish line and relay-specific scope. Current status cannot override it.
- Baton: current position, completed work, what remains, relevant evidence and decisions, useful pointers, and any blocker.
- History: concise entries recording each leg's result, decisions, artifacts, checks, and handoff or stop.
Keep the baton compact. Use history for targeted lookup, not mandatory reconstruction. Inspect the actual work as needed; the record saves rediscovering prior decisions, not reading the project.
Signal over spin
Stop when the finish line is met, with evidence. Completion does not require finding another improvement.
Stop visibly when progress needs human input: record the blocker, what was tried, and the specific decision or help needed. If successive legs keep revisiting the same issue without new evidence or meaningful progress, surface that stall instead of spawning another attempt.
Optional cleanup, hypothetical risks, and a desire for perfection are not reasons to prolong the chain. Revisit a settled decision when facts change or concrete evidence shows it was wrong, not merely because a fresh agent has arrived.
Tools, storage, and session creation belong to the environment-specific runner skill, not the definition of Relay.
Other versions25 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 25
- sonichi/sutando★ 396same name
- gavishap/omnia-vault★ 53same name
- jaskirat1616/grok-skills★ 22same name
- bekhruztbx/ai-creative-team★ 18same name
- mhlaghari/stuntman★ 16same name
- gao-xh/multi-agent-protocol★ 4same name
- kurofune/claudit★ 4same name
- RelayMessenger/Relay-SDK★ 4same name
- theyoungastronauts/polaris★ 4same name
- zommuter/dotclaude-skills★ 4same name
15 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository8 other files it gives its agents.
- AGENTS.mdAGENTS.md
- changeset-changelogSkill
- code-quality-architectureSkill
- documentation-guideSkill
- npm-release-via-github-actionsSkill
- testing-guideSkill
- ux-guideSkill
- relay-runnerSkill
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
Reviews can't be read right now.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

