OmniNode-ai

/onex:cloud_delegate — delegate to the platform, keep the files

"Delegate a task to the OmniNode platform with a dashboard API key. Runs `onex cloud delegate \"<prompt>\" --task-type <type>`, which submits over HTTPS, polls to a terminal state, fetches the signed receipt, and saves the output to local files. Handled inline — no subagent, no HTTP from the plugin."

OmniNode-ai 4 3 Updated 1w ago

Resources

1
GitHub

Install

npx skillscat add omninode-ai/omniclaude/plugins-onex-skills-cloud-delegate

Install via the SkillsCat registry.

SKILL.md

/onex:cloud_delegate — delegate to the platform, keep the files

Skill ID: onex:cloud_delegate · Command: onex cloud delegate · Backing surface: omnimarket.cli.cli_cloud (gateway HTTPS)

A dispatch skill IS one CLI call. This skill runs that command and presents its
output. It makes no HTTP calls of its own, and it must not: the Claude Code
plugin never talks to the cloud (standing 2026-08-03 ruling). The onex CLI
is the sole gateway client; this skill is a shim over it.

See prompt.md for the one command and how to present the result.

cloud_delegate vs delegate — pick the right one

/onex:delegate /onex:cloud_delegate
Runs on a model you choose: your own model server, or a provider on your own key, called from your machine the OmniNode platform
Transport from your machine to that model (in-process, or over a local bus) HTTPS to the gateway
Credential none, or your own provider key registered with onex secret set a dashboard-minted onxk_ API key
Needs the onex tool and a declared model or registered provider key nothing but the key and a base URL
Receipt typed result on stdout, plus receipt.json under .onex_state/runs/<run id>/ a signed receipt saved to disk

The two are siblings, not modes of one command. delegate sends the work
from your machine to a model you chose; cloud_delegate sends it to the OmniNode platform.

Prerequisite: install the CLI, then log in

The cloud command group is contributed by the omnimarket package through
the onex.cli entry-point group — installing that package is what makes
onex cloud exist. Nothing hand-wires it.

Package Provides
omnibase-core the onex console script and the onex.cli entry-point loader
omnimarket the cloud command group (login, delegate, receipt, status, logout)
uv tool install --python 3.12 --with omnimarket omnibase-core

Verify: onex cloud --help must exit 0 from any directory. If it exits 2 with
No such command 'cloud', omnimarket is not installed in the same environment
as the onex script.

Then store the key created in the dashboard — read from stdin, never passed as a
flag value:

read -rs ONXK && printf '%s' "$ONXK" | \
  onex cloud login --base-url https://dev.api.omninode.ai --api-key-stdin

onex cloud status prints which key is configured without printing the key.

Task Types

summarization, research, document, test, code_generation,
code_review, refactor, reasoning, complex_reasoning, planning,
review.

Unlike /onex:delegate, --task-type is required — the gateway contract
requires it, and guessing it would silently change which model answers.

Usage

/onex:cloud_delegate --task-type summarization summarize this changelog
/onex:cloud_delegate --task-type code_generation write a Python retry helper
/onex:cloud_delegate --task-type research --max-tokens 2048 compare Kafka and NATS

The files are the point

Every run writes <output-dir>/<workflow_id>/:

File What it holds
result.txt the generated output, verbatim (omitted when the run produced no content)
receipt.json the server's signed receipt, unmodified, hashes included
run.json what was asked, where, and the terminal status — so the receipt can be tied back to its request later

This is why the terminal path is preferred over a browser demo: a browser cannot
keep what it generates. Always report the saved paths.

What This Skill Does NOT Do

  • Make any HTTP request from the plugin — the CLI is the only gateway client
  • Read, print, echo, or store the API key; the CLI resolves it from ~/.onex
  • Retry a failed run. A terminal failed with no content is a runtime failure
    class (a quota-dead model key, say) and is surfaced, not retried
  • Spawn a general-purpose subagent (skill_kind: dispatch)
  • Fall back to answering the prompt itself if the delegation fails

Related

  • CLI entrypoint: omnimarket/src/omnimarket/cli/cli_cloud.py, registered as [project.entry-points."onex.cli"] cloud in omnimarket/pyproject.toml
  • Registration ratchet: omnimarket/tests/unit/cli/test_cloud_cli_entry_point_registration_omn16967.py
  • Gateway client: omnimarket/src/omnimarket/cloud/transport_cloud_delegation.py
  • Console script: omnibase_core/pyproject.toml [project.scripts] onex; the onex.cli group loader is omnibase_core/src/omnibase_core/cli/cli_commands.py