Thin Codex skill shim for node_session_orchestrator. Use to run the session health gate, queue scoring, and dispatch planning loop.
68%
Does it follow best practices?
Run evals on this skill
Adds up to 20 points to the overall score
Passed
No findings from the security scan
Fix and improve this skill with Tessl
tessl review fix ./plugins/onex/skills/session-orchestrator/SKILL.mdThis skill is a thin Codex runtime request adapter shim over the OmniMarket
node_session_orchestrator node. The node owns health gating, queue scoring,
and dispatch planning. Do not add local health probes, dispatch compilation, or
ticket triage logic to this skill.
This section is generated from the Codex adapter contract and the target node contract. Keep the two surfaces distinct:
session_orchestratorscripts/run_codex_runtime_request.pyonex.cmd.omnimarket.pattern-b-dispatch.v1onex.evt.omnimarket.pattern-b-dispatch-completed.v1--compile-only to validate the request and binding
without publishing an event or starting a runtime. This is adapter preflight,
not evidence that a target runtime executed the command.runtime_evidence.runtime_observation and
runtime_evidence.adapter_dispatch_binding; compile-only is UNOBSERVED
with reason compile_only.runtime_evidence.schema_version is
runtime-evidence/v2; v2 requires runtime_observation and carries the
resolved node contract under adapter_dispatch_binding.node_contract.adapter_dispatch_binding reports
adapter_command_topic, requested_response_topic,
selected_terminal_topic, and terminal_selection (NODE_CONTRACT,
DIRECT_DELEGATE_SKILL_CONTRACT, or EXPLICIT_RESPONSE_OVERRIDE); its
node_contract is the resolved, typed contract binding.node_session_orchestratorsession_orchestratoronex.cmd.omnimarket.session-orchestrator-start.v1onex.evt.omnimarket.session-orchestrator-completed.v1The skill must use the request wrapper for dispatch. It must not publish the target node command topic directly as its generic adapter transport; the target topics above are contract metadata selected by the runtime adapter.
| Argument | Description | Default |
|---|---|---|
correlation_id | UUID v4 correlation id for the session run | Required |
session_id | Explicit session id. Node auto-generates one when omitted. | Auto |
mode | interactive or autonomous | interactive |
dry_run | Produce plan and receipts without dispatching workers | false |
skip_health | Skip Phase 1 health checks | false |
standing_orders_path | Standing orders input path | .onex_state/session/standing_orders.json |
state_dir | Session state directory | .onex_state/session |
phase | Run one phase only (1, 2, 3) or 0 for full loop | 0 |
target_runtime_address | Optional runtime://... runtime target | Uses ONEX_TARGET_RUNTIME_ADDRESS when set |
Run from the omnimarket repo or an omnimarket worktree:
env -u PYTHONPATH uv run python scripts/run_codex_runtime_request.py \
--command-name "session_orchestrator" \
--payload '<json-payload>' \
--timeout-ms 300000If the user supplies a runtime://... target, add
--target-runtime-address '<runtime-address>' to the request command. If the
argument is omitted, the wrapper uses ONEX_TARGET_RUNTIME_ADDRESS when set.
For event-bus-free preflight, add --compile-only. This validates the payload,
command topic, response topic, correlation id, timeout, and target runtime
address without publishing to Kafka or starting a runtime.
Map user inputs into a JSON payload using the same field names:
correlation_idsession_idmodedry_runskip_healthstanding_orders_pathstate_dirphaseGenerate a UUIDv4 correlation_id when the user does not supply one. Omit
session_id only when you want the node to auto-generate it.
If ok is true and output_payloads is present, treat output_payloads[0]
as the primary node result.
If ok is false, surface error.code and error.message directly.
src/omnimarket/nodes/node_session_orchestrator/scripts/run_codex_runtime_request.pysession_orchestratoronex.cmd.omnimarket.session-orchestrator-start.v1onex.evt.omnimarket.session-orchestrator-completed.v1Prefer output_payloads[0]. Render the node result with status,
halt_reason, health_report, dispatch_queue, and dispatch_receipts. For
dry runs, report queue length and receipt count without inventing worker
execution details.
OmniNode-ai/omnimarket
13a7e30
If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.