docs: add agent skill for public O deploy + health validation - #3
Open
qianghan wants to merge 1 commit into
Open
docs: add agent skill for public O deploy + health validation#3qianghan wants to merge 1 commit into
qianghan wants to merge 1 commit into
Conversation
Add skills/deploying-scope-public-orchestrator/SKILL.md so agents (Claude, Codex) can complete a public Scope orchestrator deployment and validate it is genuinely online and healthy. The skill's core point: a Scope live runner can register, advertise capacity_available, accept a session, and bill the client while producing zero frames. /discovery returning 200 is necessary but not sufficient. The skill provides an ordered, runnable validation (containers -> healthz -> models actually downloaded -> GPU/CUDA sanity -> RPC reachable -> discovery -> a real session producing frames) plus a table of observed failure modes, so agents do not report a non-functional runner as healthy. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Signed-off-by: qianghan <infraservice@livepeer.org>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Why
The bundle deploys cleanly, but "did it work?" is the hard part — and the
current
README.mdvalidation (registration logs +/discoveryreachable +TLS up) is exactly the check a non-functional runner still passes.
In practice a Scope live runner can register, advertise
capacity_available,accept a session, and even bill the client — while producing zero frames. The
orchestrator's
/discoveryreturns200, gateways route real traffic to it,the per-session balance is debited by time×pixels (not by output), and every
stream silently black-holes. Surface checks stay green the whole time.
This PR adds an agent-runnable skill so Claude/Codex can both complete a
public-O deployment and prove it is genuinely serving video.
What
skills/deploying-scope-public-orchestrator/SKILL.md—agentskills.ioformat (
name+description), usable by Claude and Codex.README.md(no duplication).containers/healthcheck →
go-livepeer /healthz→ models actuallydownloaded (
/workspace/shared/modelsmust hold real weights, not justlora/) → GPU/CUDA kernel sanity → Arbitrum RPC reachable (no 401flood) →
/discoveryadvertises the runner → a real session producesframes (the decisive, sufficient check).
No artifacts defined,no kernel image is available for execution on the device, RPC401 … does not have access to this network, tricklein=none, out=none/ "stream disappeared", balance drained with no video,tunnel buffering).
alone.
README.md— a short "For Agents (Claude / Codex)" pointer to the skill.Grounding
The validation steps and failure table are derived from real observed failures
on a live public deployment (a runner that passed every README check yet never
produced a frame: empty model volume, CUDA arch mismatch, and a 401 RPC). The
skill exists to turn those into a repeatable, agent-driven health gate.
🤖 Generated with Claude Code